diff options
Diffstat (limited to 'src/mesa/drivers/dri/r300')
44 files changed, 3895 insertions, 8427 deletions
diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile index 6ca934204f..bdb09624be 100644 --- a/src/mesa/drivers/dri/r300/Makefile +++ b/src/mesa/drivers/dri/r300/Makefile @@ -3,6 +3,8 @@ TOP = ../../../../.. include $(TOP)/configs/current +CFLAGS += $(RADEON_CFLAGS) + LIBNAME = r300_dri.so MINIGLX_SOURCES = server/radeon_dri.c @@ -20,20 +22,26 @@ COMMON_SOURCES = \ ../common/xmlconfig.c \ ../common/dri_util.c +RADEON_COMMON_SOURCES = \ + radeon_texture.c \ + radeon_common_context.c \ + radeon_common.c \ + radeon_dma.c \ + radeon_lock.c \ + radeon_bo_legacy.c \ + radeon_cs_legacy.c \ + radeon_mipmap_tree.c \ + radeon_span.c \ + radeon_fbo.c + DRIVER_SOURCES = \ radeon_screen.c \ - radeon_context.c \ - radeon_ioctl.c \ - radeon_lock.c \ - radeon_span.c \ - radeon_state.c \ - r300_mem.c \ r300_context.c \ + r300_draw.c \ r300_ioctl.c \ r300_cmdbuf.c \ r300_state.c \ r300_render.c \ - r300_texmem.c \ r300_tex.c \ r300_texstate.c \ radeon_program.c \ @@ -41,6 +49,7 @@ DRIVER_SOURCES = \ radeon_program_pair.c \ radeon_nqssadce.c \ r300_vertprog.c \ + r300_fragprog_common.c \ r300_fragprog.c \ r300_fragprog_swizzle.c \ r300_fragprog_emit.c \ @@ -49,12 +58,15 @@ DRIVER_SOURCES = \ r300_shader.c \ r300_emit.c \ r300_swtcl.c \ + $(RADEON_COMMON_SOURCES) \ $(EGL_SOURCES) C_SOURCES = $(COMMON_SOURCES) $(DRIVER_SOURCES) DRIVER_DEFINES = -DCOMPILE_R300 -DR200_MERGED=0 \ - -DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R300 + -DRADEON_COMMON=1 -DRADEON_COMMON_FOR_R300 \ +# -DRADEON_BO_TRACK \ + -Wall SYMLINKS = \ server/radeon_dri.c \ @@ -68,7 +80,29 @@ COMMON_SYMLINKS = \ radeon_chipset.h \ radeon_screen.c \ radeon_screen.h \ - radeon_span.h + radeon_span.h \ + radeon_span.c \ + radeon_bo_legacy.c \ + radeon_cs_legacy.c \ + radeon_bo_legacy.h \ + radeon_cs_legacy.h \ + radeon_bocs_wrapper.h \ + radeon_lock.c \ + radeon_lock.h \ + radeon_common.c \ + radeon_common.h \ + radeon_common_context.c \ + radeon_common_context.h \ + radeon_cmdbuf.h \ + radeon_dma.c \ + radeon_dma.h \ + radeon_mipmap_tree.c \ + radeon_mipmap_tree.h \ + radeon_texture.c \ + radeon_texture.h \ + radeon_fbo.c + +DRI_LIB_DEPS += $(RADEON_LDFLAGS) ##### TARGETS ##### diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c index f4472756f1..0261a5b1d8 100644 --- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c +++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c @@ -44,245 +44,417 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drm.h" #include "radeon_drm.h" -#include "radeon_ioctl.h" #include "r300_context.h" #include "r300_ioctl.h" #include "radeon_reg.h" #include "r300_reg.h" #include "r300_cmdbuf.h" #include "r300_emit.h" +#include "radeon_bocs_wrapper.h" +#include "radeon_mipmap_tree.h" #include "r300_state.h" +#include "radeon_reg.h" -// Set this to 1 for extremely verbose debugging of command buffers -#define DEBUG_CMDBUF 0 - -/** - * Send the current command buffer via ioctl to the hardware. +/** # of dwords reserved for additional instructions that may need to be written + * during flushing. */ -int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller) +#define SPACE_FOR_FLUSHING 4 + +static unsigned packet0_count(r300ContextPtr r300, uint32_t *pkt) { - int ret; - int i; - drm_radeon_cmd_buffer_t cmd; - int start; - - if (r300->radeon.lost_context) { - start = 0; - r300->radeon.lost_context = GL_FALSE; - } else - start = r300->cmdbuf.count_reemit; - - if (RADEON_DEBUG & DEBUG_IOCTL) { - fprintf(stderr, "%s from %s - %i cliprects\n", - __FUNCTION__, caller, r300->radeon.numClipRects); - - if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_VERBOSE) - for (i = start; i < r300->cmdbuf.count_used; ++i) - fprintf(stderr, "%d: %08x\n", i, - r300->cmdbuf.cmd_buf[i]); - } + if (r300->radeon.radeonScreen->kernel_mm) { + return ((((*pkt) >> 16) & 0x3FFF) + 1); + } else { + drm_r300_cmd_header_t *t = (drm_r300_cmd_header_t*)pkt; + return t->packet0.count; + } +} - cmd.buf = (char *)(r300->cmdbuf.cmd_buf + start); - cmd.bufsz = (r300->cmdbuf.count_used - start) * 4; +#define vpu_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->vpu.count) +#define r500fp_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->r500fp.count) - if (r300->radeon.state.scissor.enabled) { - cmd.nbox = r300->radeon.state.scissor.numClipRects; - cmd.boxes = - (drm_clip_rect_t *) r300->radeon.state.scissor.pClipRects; - } else { - cmd.nbox = r300->radeon.numClipRects; - cmd.boxes = (drm_clip_rect_t *) r300->radeon.pClipRects; +void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + drm_r300_cmd_header_t cmd; + uint32_t addr, ndw, i; + + if (!r300->radeon.radeonScreen->kernel_mm) { + uint32_t dwords; + dwords = (*atom->check) (ctx, atom); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, dwords); + END_BATCH(); + return; } - ret = drmCommandWrite(r300->radeon.dri.fd, - DRM_RADEON_CMDBUF, &cmd, sizeof(cmd)); - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "Syncing in %s (from %s)\n\n", - __FUNCTION__, caller); - radeonWaitForIdleLocked(&r300->radeon); + cmd.u = atom->cmd[0]; + addr = (cmd.vpu.adrhi << 8) | cmd.vpu.adrlo; + ndw = cmd.vpu.count * 4; + if (ndw) { + + if (r300->vap_flush_needed) { + BEGIN_BATCH_NO_AUTOSTATE(15 + ndw); + + /* flush processing vertices */ + OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0); + OUT_BATCH_REGVAL(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); + OUT_BATCH_REGVAL(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); + OUT_BATCH_REGVAL(R300_SC_SCREENDOOR, 0xffffff); + OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); + r300->vap_flush_needed = GL_FALSE; + } else { + BEGIN_BATCH_NO_AUTOSTATE(5 + ndw); + } + OUT_BATCH_REGVAL(R300_VAP_PVS_VECTOR_INDX_REG, addr); + OUT_BATCH(CP_PACKET0(R300_VAP_PVS_UPLOAD_DATA, ndw-1) | RADEON_ONE_REG_WR); + for (i = 0; i < ndw; i++) { + OUT_BATCH(atom->cmd[i+1]); + } + OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); + END_BATCH(); } - - r300->dma.nr_released_bufs = 0; - r300->cmdbuf.count_used = 0; - r300->cmdbuf.count_reemit = 0; - - return ret; } -int r300FlushCmdBuf(r300ContextPtr r300, const char *caller) +void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom) { - int ret; + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + drm_r300_cmd_header_t cmd; + uint32_t addr, ndw, i, sz; + int type, clamp, stride; + + if (!r300->radeon.radeonScreen->kernel_mm) { + uint32_t dwords; + dwords = (*atom->check) (ctx, atom); + BEGIN_BATCH_NO_AUTOSTATE(dwords); + OUT_BATCH_TABLE(atom->cmd, dwords); + END_BATCH(); + return; + } - LOCK_HARDWARE(&r300->radeon); + cmd.u = atom->cmd[0]; + sz = cmd.r500fp.count; + addr = ((cmd.r500fp.adrhi_flags & 1) << 8) | cmd.r500fp.adrlo; + type = !!(cmd.r500fp.adrhi_flags & R500FP_CONSTANT_TYPE); + clamp = !!(cmd.r500fp.adrhi_flags & R500FP_CONSTANT_CLAMP); - ret = r300FlushCmdBufLocked(r300, caller); + addr |= (type << 16); + addr |= (clamp << 17); - UNLOCK_HARDWARE(&r300->radeon); + stride = type ? 4 : 6; - if (ret) { - fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret); - _mesa_exit(ret); - } + ndw = sz * stride; + if (ndw) { - return ret; + BEGIN_BATCH_NO_AUTOSTATE(3 + ndw); + OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_INDEX, 0)); + OUT_BATCH(addr); + OUT_BATCH(CP_PACKET0(R500_GA_US_VECTOR_DATA, ndw-1) | RADEON_ONE_REG_WR); + for (i = 0; i < ndw; i++) { + OUT_BATCH(atom->cmd[i+1]); + } + END_BATCH(); + } } -static void r300PrintStateAtom(r300ContextPtr r300, struct r300_state_atom *state) +static void emit_tex_offsets(GLcontext *ctx, struct radeon_state_atom * atom) { - int i, j, reg; - int dwords = (*state->check) (r300, state); - drm_r300_cmd_header_t cmd; + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + int numtmus = packet0_count(r300, r300->hw.tex.offset.cmd); + int notexture = 0; - fprintf(stderr, " emit %s %d/%d\n", state->name, dwords, - state->cmd_size); - - if (RADEON_DEBUG & DEBUG_VERBOSE) { - for (i = 0; i < dwords;) { - cmd = *((drm_r300_cmd_header_t *) &state->cmd[i]); - reg = (cmd.packet0.reghi << 8) | cmd.packet0.reglo; - fprintf(stderr, " %s[%d]: cmdpacket0 (first reg=0x%04x, count=%d)\n", - state->name, i, reg, cmd.packet0.count); - ++i; - for (j = 0; j < cmd.packet0.count; j++) { - fprintf(stderr, " %s[%d]: 0x%04x = %08x\n", - state->name, i, reg, state->cmd[i]); - reg += 4; - ++i; - } + if (numtmus) { + int i; + + for(i = 0; i < numtmus; ++i) { + radeonTexObj *t = r300->hw.textures[i]; + + if (!t) + notexture = 1; + } + + if (r300->radeon.radeonScreen->kernel_mm && notexture) { + return; + } + for(i = 0; i < numtmus; ++i) { + radeonTexObj *t = r300->hw.textures[i]; + if (t && !t->image_override) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH_RELOC(t->tile_bits, t->mt->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + END_BATCH(); + } else if (!t) { + /* Texture unit hasn't a texture bound nothings to do */ + } else { /* override cases */ + if (t->bo) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH_RELOC(t->tile_bits, t->bo, 0, + RADEON_GEM_DOMAIN_GTT|RADEON_GEM_DOMAIN_VRAM, 0, 0); + END_BATCH(); + } else if (!r300->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGSEQ(R300_TX_OFFSET_0 + (i * 4), 1); + OUT_BATCH(t->override_offset); + END_BATCH(); + } else { + /* Texture unit hasn't a texture bound nothings to do */ + } + } } } } -/** - * Emit all atoms with a dirty field equal to dirty. - * - * The caller must have ensured that there is enough space in the command - * buffer. - */ -static INLINE void r300EmitAtoms(r300ContextPtr r300, GLboolean dirty) +void r300_emit_scissor(GLcontext *ctx) { - struct r300_state_atom *atom; - uint32_t *dest; - int dwords; - - dest = r300->cmdbuf.cmd_buf + r300->cmdbuf.count_used; - - /* Emit WAIT */ - *dest = cmdwait(R300_WAIT_3D | R300_WAIT_3D_CLEAN); - dest++; - r300->cmdbuf.count_used++; - - /* Emit cache flush */ - *dest = cmdpacket0(R300_TX_INVALTAGS, 1); - dest++; - r300->cmdbuf.count_used++; - - *dest = R300_TX_FLUSH; - dest++; - r300->cmdbuf.count_used++; - - /* Emit END3D */ - *dest = cmdpacify(); - dest++; - r300->cmdbuf.count_used++; - - /* Emit actual atoms */ - - foreach(atom, &r300->hw.atomlist) { - if ((atom->dirty || r300->hw.all_dirty) == dirty) { - dwords = (*atom->check) (r300, atom); - if (dwords) { - if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) { - r300PrintStateAtom(r300, atom); - } - memcpy(dest, atom->cmd, dwords * 4); - dest += dwords; - r300->cmdbuf.count_used += dwords; - atom->dirty = GL_FALSE; - } else { - if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) { - fprintf(stderr, " skip state %s\n", - atom->name); - } - } - } + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + unsigned x1, y1, x2, y2; + struct radeon_renderbuffer *rrb; + + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + return; + } + rrb = radeon_get_colorbuffer(&r300->radeon); + if (!rrb || !rrb->bo) { + fprintf(stderr, "no rrb\n"); + return; } + if (r300->radeon.state.scissor.enabled) { + x1 = r300->radeon.state.scissor.rect.x1; + y1 = r300->radeon.state.scissor.rect.y1; + x2 = r300->radeon.state.scissor.rect.x2 - 1; + y2 = r300->radeon.state.scissor.rect.y2 - 1; + } else { + x1 = 0; + y1 = 0; + x2 = rrb->width - 1; + y2 = rrb->height - 1; + } + if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) { + x1 += R300_SCISSORS_OFFSET; + y1 += R300_SCISSORS_OFFSET; + x2 += R300_SCISSORS_OFFSET; + y2 += R300_SCISSORS_OFFSET; + } + BEGIN_BATCH_NO_AUTOSTATE(3); + OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2); + OUT_BATCH((x1 << R300_SCISSORS_X_SHIFT)|(y1 << R300_SCISSORS_Y_SHIFT)); + OUT_BATCH((x2 << R300_SCISSORS_X_SHIFT)|(y2 << R300_SCISSORS_Y_SHIFT)); + END_BATCH(); } -/** - * Copy dirty hardware state atoms into the command buffer. - * - * We also copy out clean state if we're at the start of a buffer. That makes - * it easy to recover from lost contexts. - */ -void r300EmitState(r300ContextPtr r300) +static void emit_cb_offset(GLcontext *ctx, struct radeon_state_atom * atom) { - if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_PRIMS)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (r300->cmdbuf.count_used && !r300->hw.is_dirty - && !r300->hw.all_dirty) + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + struct radeon_renderbuffer *rrb; + uint32_t cbpitch; + uint32_t offset = r300->radeon.state.color.draw_offset; + uint32_t dw = 6; + int i; + + rrb = radeon_get_colorbuffer(&r300->radeon); + if (!rrb || !rrb->bo) { + fprintf(stderr, "no rrb\n"); return; + } - /* To avoid going across the entire set of states multiple times, just check - * for enough space for the case of emitting all state, and inline the - * r300AllocCmdBuf code here without all the checks. - */ - r300EnsureCmdBufSpace(r300, r300->hw.max_state_size, __FUNCTION__); + cbpitch = (rrb->pitch / rrb->cpp); + if (rrb->cpp == 4) + cbpitch |= R300_COLOR_FORMAT_ARGB8888; + else + cbpitch |= R300_COLOR_FORMAT_RGB565; + + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) + cbpitch |= R300_COLOR_TILE_ENABLE; + + if (r300->radeon.radeonScreen->kernel_mm) + dw += 2; + BEGIN_BATCH_NO_AUTOSTATE(dw); + OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1); + OUT_BATCH_RELOC(offset, rrb->bo, offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_BATCH_REGSEQ(R300_RB3D_COLORPITCH0, 1); + if (!r300->radeon.radeonScreen->kernel_mm) + OUT_BATCH(cbpitch); + else + OUT_BATCH_RELOC(cbpitch, rrb->bo, cbpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); + END_BATCH(); + if (r300->radeon.radeonScreen->driScreen->dri2.enabled) { + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + BEGIN_BATCH_NO_AUTOSTATE(3); + OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2); + OUT_BATCH(0); + OUT_BATCH(((rrb->width - 1) << R300_SCISSORS_X_SHIFT) | + ((rrb->height - 1) << R300_SCISSORS_Y_SHIFT)); + END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(16); + for (i = 0; i < 4; i++) { + OUT_BATCH_REGSEQ(R300_SC_CLIPRECT_TL_0 + (i * 8), 2); + OUT_BATCH((0 << R300_CLIPRECT_X_SHIFT) | (0 << R300_CLIPRECT_Y_SHIFT)); + OUT_BATCH(((rrb->width - 1) << R300_CLIPRECT_X_SHIFT) | ((rrb->height - 1) << R300_CLIPRECT_Y_SHIFT)); + } + OUT_BATCH_REGSEQ(R300_SC_CLIP_RULE, 1); + OUT_BATCH(0xAAAA); + OUT_BATCH_REGSEQ(R300_SC_SCREENDOOR, 1); + OUT_BATCH(0xffffff); + END_BATCH(); + } else { + BEGIN_BATCH_NO_AUTOSTATE(3); + OUT_BATCH_REGSEQ(R300_SC_SCISSORS_TL, 2); + OUT_BATCH((R300_SCISSORS_OFFSET << R300_SCISSORS_X_SHIFT) | + (R300_SCISSORS_OFFSET << R300_SCISSORS_Y_SHIFT)); + OUT_BATCH(((rrb->width + R300_SCISSORS_OFFSET - 1) << R300_SCISSORS_X_SHIFT) | + ((rrb->height + R300_SCISSORS_OFFSET - 1) << R300_SCISSORS_Y_SHIFT)); + END_BATCH(); + BEGIN_BATCH_NO_AUTOSTATE(16); + for (i = 0; i < 4; i++) { + OUT_BATCH_REGSEQ(R300_SC_CLIPRECT_TL_0 + (i * 8), 2); + OUT_BATCH((R300_SCISSORS_OFFSET << R300_CLIPRECT_X_SHIFT) | (R300_SCISSORS_OFFSET << R300_CLIPRECT_Y_SHIFT)); + OUT_BATCH(((R300_SCISSORS_OFFSET + rrb->width - 1) << R300_CLIPRECT_X_SHIFT) | + ((R300_SCISSORS_OFFSET + rrb->height - 1) << R300_CLIPRECT_Y_SHIFT)); + } + OUT_BATCH_REGSEQ(R300_SC_CLIP_RULE, 1); + OUT_BATCH(0xAAAA); + OUT_BATCH_REGSEQ(R300_SC_SCREENDOOR, 1); + OUT_BATCH(0xffffff); + END_BATCH(); + } + } +} - if (!r300->cmdbuf.count_used) { - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Begin reemit state\n"); +static void emit_zb_offset(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + struct radeon_renderbuffer *rrb; + uint32_t zbpitch; - r300EmitAtoms(r300, GL_FALSE); - r300->cmdbuf.count_reemit = r300->cmdbuf.count_used; + rrb = radeon_get_depthbuffer(&r300->radeon); + if (!rrb) + return; + + zbpitch = (rrb->pitch / rrb->cpp); + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE) { + zbpitch |= R300_DEPTHMACROTILE_ENABLE; + } + if (rrb->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){ + zbpitch |= R300_DEPTHMICROTILE_TILED; } - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Begin dirty state\n"); + BEGIN_BATCH_NO_AUTOSTATE(6); + OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1); + OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, zbpitch); + END_BATCH(); +} - r300EmitAtoms(r300, GL_TRUE); +static void emit_gb_misc(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH(atom->cmd[0]); + OUT_BATCH(atom->cmd[1]); + OUT_BATCH(atom->cmd[2]); + OUT_BATCH(atom->cmd[3]); + END_BATCH(); + } +} - assert(r300->cmdbuf.count_used < r300->cmdbuf.size); +static void emit_threshold_misc(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + BEGIN_BATCH_NO_AUTOSTATE(3); + OUT_BATCH(atom->cmd[0]); + OUT_BATCH(atom->cmd[1]); + OUT_BATCH(atom->cmd[2]); + END_BATCH(); + } +} - r300->hw.is_dirty = GL_FALSE; - r300->hw.all_dirty = GL_FALSE; +static void emit_shade_misc(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH(atom->cmd[0]); + OUT_BATCH(atom->cmd[1]); + END_BATCH(); + } } -#define packet0_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->packet0.count) -#define vpu_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->vpu.count) -#define r500fp_count(ptr) (((drm_r300_cmd_header_t*)(ptr))->r500fp.count) +static void emit_zstencil_format(GLcontext *ctx, struct radeon_state_atom * atom) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + BATCH_LOCALS(&r300->radeon); + struct radeon_renderbuffer *rrb; + uint32_t format = 0; + + rrb = radeon_get_depthbuffer(&r300->radeon); + if (!rrb) + format = 0; + else { + if (rrb->cpp == 2) + format = R300_DEPTHFORMAT_16BIT_INT_Z; + else if (rrb->cpp == 4) + format = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; + } + + OUT_BATCH(atom->cmd[0]); + atom->cmd[1] &= ~0xf; + atom->cmd[1] |= format; + OUT_BATCH(atom->cmd[1]); + OUT_BATCH(atom->cmd[2]); + OUT_BATCH(atom->cmd[3]); + OUT_BATCH(atom->cmd[4]); +} -static int check_always(r300ContextPtr r300, struct r300_state_atom *atom) +static int check_always(GLcontext *ctx, struct radeon_state_atom *atom) { return atom->cmd_size; } -static int check_variable(r300ContextPtr r300, struct r300_state_atom *atom) +static int check_variable(GLcontext *ctx, struct radeon_state_atom *atom) { + r300ContextPtr r300 = R300_CONTEXT(ctx); int cnt; - cnt = packet0_count(atom->cmd); + if (atom->cmd[0] == CP_PACKET2) { + return 0; + } + cnt = packet0_count(r300, atom->cmd); return cnt ? cnt + 1 : 0; } -static int check_vpu(r300ContextPtr r300, struct r300_state_atom *atom) +int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom) { int cnt; + cnt = vpu_count(atom->cmd); return cnt ? (cnt * 4) + 1 : 0; } -static int check_r500fp(r300ContextPtr r300, struct r300_state_atom *atom) +int check_r500fp(GLcontext *ctx, struct radeon_state_atom *atom) { int cnt; + cnt = r500fp_count(atom->cmd); return cnt ? (cnt * 6) + 1 : 0; } -static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom) +int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom) { int cnt; + cnt = r500fp_count(atom->cmd); return cnt ? (cnt * 4) + 1 : 0; } @@ -295,8 +467,8 @@ static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom) r300->hw.ATOM.idx = (IDX); \ r300->hw.ATOM.check = check_##CHK; \ r300->hw.ATOM.dirty = GL_FALSE; \ - r300->hw.max_state_size += (SZ); \ - insert_at_tail(&r300->hw.atomlist, &r300->hw.ATOM); \ + r300->radeon.hw.max_state_size += (SZ); \ + insert_at_tail(&r300->radeon.hw.atomlist, &r300->hw.ATOM); \ } while (0) /** * Allocate memory for the command buffer and initialize the state atom @@ -304,18 +476,16 @@ static int check_r500fp_const(r300ContextPtr r300, struct r300_state_atom *atom) */ void r300InitCmdBuf(r300ContextPtr r300) { - int size, mtu; - int has_tcl = 1; + int mtu; + int has_tcl; int is_r500 = 0; - int i; - if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) - has_tcl = 0; + has_tcl = r300->options.hw_tcl_enabled; if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) is_r500 = 1; - r300->hw.max_state_size = 2 + 2; /* reserve extra space for WAIT_IDLE and tex cache flush */ + r300->radeon.hw.max_state_size = 2 + 2; /* reserve extra space for WAIT_IDLE and tex cache flush */ mtu = r300->radeon.glCtx->Const.MaxTextureUnits; if (RADEON_DEBUG & DEBUG_TEXTURE) { @@ -323,232 +493,248 @@ void r300InitCmdBuf(r300ContextPtr r300) } /* Setup the atom linked list */ - make_empty_list(&r300->hw.atomlist); - r300->hw.atomlist.name = "atom-list"; + make_empty_list(&r300->radeon.hw.atomlist); + r300->radeon.hw.atomlist.name = "atom-list"; /* Initialize state atoms */ ALLOC_STATE(vpt, always, R300_VPT_CMDSIZE, 0); - r300->hw.vpt.cmd[R300_VPT_CMD_0] = cmdpacket0(R300_SE_VPORT_XSCALE, 6); + r300->hw.vpt.cmd[R300_VPT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_SE_VPORT_XSCALE, 6); ALLOC_STATE(vap_cntl, always, R300_VAP_CNTL_SIZE, 0); - r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH] = cmdpacket0(R300_VAP_PVS_STATE_FLUSH_REG, 1); + r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PVS_STATE_FLUSH_REG, 1); r300->hw.vap_cntl.cmd[R300_VAP_CNTL_FLUSH_1] = 0; - r300->hw.vap_cntl.cmd[R300_VAP_CNTL_CMD] = cmdpacket0(R300_VAP_CNTL, 1); + r300->hw.vap_cntl.cmd[R300_VAP_CNTL_CMD] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CNTL, 1); if (is_r500) { ALLOC_STATE(vap_index_offset, always, 2, 0); - r300->hw.vap_index_offset.cmd[0] = cmdpacket0(R500_VAP_INDEX_OFFSET, 1); + r300->hw.vap_index_offset.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_VAP_INDEX_OFFSET, 1); r300->hw.vap_index_offset.cmd[1] = 0; } ALLOC_STATE(vte, always, 3, 0); - r300->hw.vte.cmd[0] = cmdpacket0(R300_SE_VTE_CNTL, 2); + r300->hw.vte.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SE_VTE_CNTL, 2); ALLOC_STATE(vap_vf_max_vtx_indx, always, 3, 0); - r300->hw.vap_vf_max_vtx_indx.cmd[0] = cmdpacket0(R300_VAP_VF_MAX_VTX_INDX, 2); + r300->hw.vap_vf_max_vtx_indx.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_VF_MAX_VTX_INDX, 2); ALLOC_STATE(vap_cntl_status, always, 2, 0); - r300->hw.vap_cntl_status.cmd[0] = cmdpacket0(R300_VAP_CNTL_STATUS, 1); + r300->hw.vap_cntl_status.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CNTL_STATUS, 1); ALLOC_STATE(vir[0], variable, R300_VIR_CMDSIZE, 0); r300->hw.vir[0].cmd[R300_VIR_CMD_0] = - cmdpacket0(R300_VAP_PROG_STREAM_CNTL_0, 1); + cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PROG_STREAM_CNTL_0, 1); ALLOC_STATE(vir[1], variable, R300_VIR_CMDSIZE, 1); r300->hw.vir[1].cmd[R300_VIR_CMD_0] = - cmdpacket0(R300_VAP_PROG_STREAM_CNTL_EXT_0, 1); + cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PROG_STREAM_CNTL_EXT_0, 1); ALLOC_STATE(vic, always, R300_VIC_CMDSIZE, 0); - r300->hw.vic.cmd[R300_VIC_CMD_0] = cmdpacket0(R300_VAP_VTX_STATE_CNTL, 2); + r300->hw.vic.cmd[R300_VIC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_VTX_STATE_CNTL, 2); ALLOC_STATE(vap_psc_sgn_norm_cntl, always, 2, 0); - r300->hw.vap_psc_sgn_norm_cntl.cmd[0] = cmdpacket0(R300_VAP_PSC_SGN_NORM_CNTL, SGN_NORM_ZERO_CLAMP_MINUS_ONE); + r300->hw.vap_psc_sgn_norm_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PSC_SGN_NORM_CNTL, SGN_NORM_ZERO_CLAMP_MINUS_ONE); if (has_tcl) { ALLOC_STATE(vap_clip_cntl, always, 2, 0); - r300->hw.vap_clip_cntl.cmd[0] = cmdpacket0(R300_VAP_CLIP_CNTL, 1); + r300->hw.vap_clip_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_CLIP_CNTL, 1); ALLOC_STATE(vap_clip, always, 5, 0); - r300->hw.vap_clip.cmd[0] = cmdpacket0(R300_VAP_GB_VERT_CLIP_ADJ, 4); + r300->hw.vap_clip.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_VAP_GB_VERT_CLIP_ADJ, 4); ALLOC_STATE(vap_pvs_vtx_timeout_reg, always, 2, 0); - r300->hw.vap_pvs_vtx_timeout_reg.cmd[0] = cmdpacket0(VAP_PVS_VTX_TIMEOUT_REG, 1); + r300->hw.vap_pvs_vtx_timeout_reg.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, VAP_PVS_VTX_TIMEOUT_REG, 1); } ALLOC_STATE(vof, always, R300_VOF_CMDSIZE, 0); r300->hw.vof.cmd[R300_VOF_CMD_0] = - cmdpacket0(R300_VAP_OUTPUT_VTX_FMT_0, 2); + cmdpacket0(r300->radeon.radeonScreen, R300_VAP_OUTPUT_VTX_FMT_0, 2); if (has_tcl) { ALLOC_STATE(pvs, always, R300_PVS_CMDSIZE, 0); r300->hw.pvs.cmd[R300_PVS_CMD_0] = - cmdpacket0(R300_VAP_PVS_CODE_CNTL_0, 3); + cmdpacket0(r300->radeon.radeonScreen, R300_VAP_PVS_CODE_CNTL_0, 3); } ALLOC_STATE(gb_enable, always, 2, 0); - r300->hw.gb_enable.cmd[0] = cmdpacket0(R300_GB_ENABLE, 1); + r300->hw.gb_enable.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_ENABLE, 1); ALLOC_STATE(gb_misc, always, R300_GB_MISC_CMDSIZE, 0); - r300->hw.gb_misc.cmd[0] = cmdpacket0(R300_GB_MSPOS0, 5); + r300->hw.gb_misc.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GB_MSPOS0, 3); + r300->hw.gb_misc.emit = emit_gb_misc; + ALLOC_STATE(gb_misc2, always, R300_GB_MISC2_CMDSIZE, 0); + r300->hw.gb_misc2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x401C, 2); ALLOC_STATE(txe, always, R300_TXE_CMDSIZE, 0); - r300->hw.txe.cmd[R300_TXE_CMD_0] = cmdpacket0(R300_TX_ENABLE, 1); + r300->hw.txe.cmd[R300_TXE_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_ENABLE, 1); ALLOC_STATE(ga_point_s0, always, 5, 0); - r300->hw.ga_point_s0.cmd[0] = cmdpacket0(R300_GA_POINT_S0, 4); + r300->hw.ga_point_s0.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_S0, 4); ALLOC_STATE(ga_triangle_stipple, always, 2, 0); - r300->hw.ga_triangle_stipple.cmd[0] = cmdpacket0(R300_GA_TRIANGLE_STIPPLE, 1); + r300->hw.ga_triangle_stipple.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_TRIANGLE_STIPPLE, 1); ALLOC_STATE(ps, always, R300_PS_CMDSIZE, 0); - r300->hw.ps.cmd[0] = cmdpacket0(R300_GA_POINT_SIZE, 1); + r300->hw.ps.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_SIZE, 1); ALLOC_STATE(ga_point_minmax, always, 4, 0); - r300->hw.ga_point_minmax.cmd[0] = cmdpacket0(R300_GA_POINT_MINMAX, 3); + r300->hw.ga_point_minmax.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POINT_MINMAX, 3); ALLOC_STATE(lcntl, always, 2, 0); - r300->hw.lcntl.cmd[0] = cmdpacket0(R300_GA_LINE_CNTL, 1); + r300->hw.lcntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_CNTL, 1); ALLOC_STATE(ga_line_stipple, always, 4, 0); - r300->hw.ga_line_stipple.cmd[0] = cmdpacket0(R300_GA_LINE_STIPPLE_VALUE, 3); - ALLOC_STATE(shade, always, 5, 0); - r300->hw.shade.cmd[0] = cmdpacket0(R300_GA_ENHANCE, 4); + r300->hw.ga_line_stipple.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_LINE_STIPPLE_VALUE, 3); + ALLOC_STATE(shade, always, 2, 0); + r300->hw.shade.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_ENHANCE, 1); + r300->hw.shade.emit = emit_shade_misc; + ALLOC_STATE(shade2, always, 4, 0); + r300->hw.shade2.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, 0x4278, 3); ALLOC_STATE(polygon_mode, always, 4, 0); - r300->hw.polygon_mode.cmd[0] = cmdpacket0(R300_GA_POLY_MODE, 3); + r300->hw.polygon_mode.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_POLY_MODE, 3); ALLOC_STATE(fogp, always, 3, 0); - r300->hw.fogp.cmd[0] = cmdpacket0(R300_GA_FOG_SCALE, 2); + r300->hw.fogp.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_GA_FOG_SCALE, 2); ALLOC_STATE(zbias_cntl, always, 2, 0); - r300->hw.zbias_cntl.cmd[0] = cmdpacket0(R300_SU_TEX_WRAP, 1); + r300->hw.zbias_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_TEX_WRAP, 1); ALLOC_STATE(zbs, always, R300_ZBS_CMDSIZE, 0); r300->hw.zbs.cmd[R300_ZBS_CMD_0] = - cmdpacket0(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); + cmdpacket0(r300->radeon.radeonScreen, R300_SU_POLY_OFFSET_FRONT_SCALE, 4); ALLOC_STATE(occlusion_cntl, always, 2, 0); - r300->hw.occlusion_cntl.cmd[0] = cmdpacket0(R300_SU_POLY_OFFSET_ENABLE, 1); + r300->hw.occlusion_cntl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_POLY_OFFSET_ENABLE, 1); ALLOC_STATE(cul, always, R300_CUL_CMDSIZE, 0); - r300->hw.cul.cmd[R300_CUL_CMD_0] = cmdpacket0(R300_SU_CULL_MODE, 1); + r300->hw.cul.cmd[R300_CUL_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_CULL_MODE, 1); ALLOC_STATE(su_depth_scale, always, 3, 0); - r300->hw.su_depth_scale.cmd[0] = cmdpacket0(R300_SU_DEPTH_SCALE, 2); + r300->hw.su_depth_scale.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SU_DEPTH_SCALE, 2); ALLOC_STATE(rc, always, R300_RC_CMDSIZE, 0); - r300->hw.rc.cmd[R300_RC_CMD_0] = cmdpacket0(R300_RS_COUNT, 2); + r300->hw.rc.cmd[R300_RC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_COUNT, 2); if (is_r500) { - ALLOC_STATE(ri, always, R500_RI_CMDSIZE, 0); - r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(R500_RS_IP_0, 16); - for (i = 0; i < 8; i++) { - r300->hw.ri.cmd[R300_RI_CMD_0 + i +1] = - (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) | - (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) | - (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) | - (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT); - } + ALLOC_STATE(ri, variable, R500_RI_CMDSIZE, 0); + r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_IP_0, 16); ALLOC_STATE(rr, variable, R300_RR_CMDSIZE, 0); - r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R500_RS_INST_0, 1); + r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, 1); } else { - ALLOC_STATE(ri, always, R300_RI_CMDSIZE, 0); - r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(R300_RS_IP_0, 8); + ALLOC_STATE(ri, variable, R300_RI_CMDSIZE, 0); + r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_IP_0, 8); ALLOC_STATE(rr, variable, R300_RR_CMDSIZE, 0); - r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R300_RS_INST_0, 1); + r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, 1); } ALLOC_STATE(sc_hyperz, always, 3, 0); - r300->hw.sc_hyperz.cmd[0] = cmdpacket0(R300_SC_HYPERZ, 2); + r300->hw.sc_hyperz.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_HYPERZ, 2); ALLOC_STATE(sc_screendoor, always, 2, 0); - r300->hw.sc_screendoor.cmd[0] = cmdpacket0(R300_SC_SCREENDOOR, 1); + r300->hw.sc_screendoor.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_SC_SCREENDOOR, 1); ALLOC_STATE(us_out_fmt, always, 6, 0); - r300->hw.us_out_fmt.cmd[0] = cmdpacket0(R300_US_OUT_FMT, 5); + r300->hw.us_out_fmt.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_OUT_FMT, 5); if (is_r500) { ALLOC_STATE(fp, always, R500_FP_CMDSIZE, 0); - r300->hw.fp.cmd[R500_FP_CMD_0] = cmdpacket0(R500_US_CONFIG, 2); + r300->hw.fp.cmd[R500_FP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_US_CONFIG, 2); r300->hw.fp.cmd[R500_FP_CNTL] = R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO; - r300->hw.fp.cmd[R500_FP_CMD_1] = cmdpacket0(R500_US_CODE_ADDR, 3); - r300->hw.fp.cmd[R500_FP_CMD_2] = cmdpacket0(R500_US_FC_CTRL, 1); + r300->hw.fp.cmd[R500_FP_CMD_1] = cmdpacket0(r300->radeon.radeonScreen, R500_US_CODE_ADDR, 3); + r300->hw.fp.cmd[R500_FP_CMD_2] = cmdpacket0(r300->radeon.radeonScreen, R500_US_FC_CTRL, 1); r300->hw.fp.cmd[R500_FP_FC_CNTL] = 0; /* FIXME when we add flow control */ ALLOC_STATE(r500fp, r500fp, R500_FPI_CMDSIZE, 0); - r300->hw.r500fp.cmd[R300_FPI_CMD_0] = cmdr500fp(0, 0, 0, 0); + r300->hw.r500fp.cmd[R300_FPI_CMD_0] = + cmdr500fp(r300->radeon.radeonScreen, 0, 0, 0, 0); + r300->hw.r500fp.emit = emit_r500fp; ALLOC_STATE(r500fp_const, r500fp_const, R500_FPP_CMDSIZE, 0); - r300->hw.r500fp_const.cmd[R300_FPI_CMD_0] = cmdr500fp(0, 0, 1, 0); + r300->hw.r500fp_const.cmd[R300_FPI_CMD_0] = + cmdr500fp(r300->radeon.radeonScreen, 0, 0, 1, 0); + r300->hw.r500fp_const.emit = emit_r500fp; } else { ALLOC_STATE(fp, always, R300_FP_CMDSIZE, 0); - r300->hw.fp.cmd[R300_FP_CMD_0] = cmdpacket0(R300_US_CONFIG, 3); - r300->hw.fp.cmd[R300_FP_CMD_1] = cmdpacket0(R300_US_CODE_ADDR_0, 4); + r300->hw.fp.cmd[R300_FP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_CONFIG, 3); + r300->hw.fp.cmd[R300_FP_CMD_1] = cmdpacket0(r300->radeon.radeonScreen, R300_US_CODE_ADDR_0, 4); + ALLOC_STATE(fpt, variable, R300_FPT_CMDSIZE, 0); - r300->hw.fpt.cmd[R300_FPT_CMD_0] = cmdpacket0(R300_US_TEX_INST_0, 0); + r300->hw.fpt.cmd[R300_FPT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_TEX_INST_0, 0); ALLOC_STATE(fpi[0], variable, R300_FPI_CMDSIZE, 0); - r300->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, 1); + r300->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_RGB_INST_0, 1); ALLOC_STATE(fpi[1], variable, R300_FPI_CMDSIZE, 1); - r300->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, 1); + r300->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_RGB_ADDR_0, 1); ALLOC_STATE(fpi[2], variable, R300_FPI_CMDSIZE, 2); - r300->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, 1); + r300->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_ALPHA_INST_0, 1); ALLOC_STATE(fpi[3], variable, R300_FPI_CMDSIZE, 3); - r300->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, 1); + r300->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_US_ALU_ALPHA_ADDR_0, 1); ALLOC_STATE(fpp, variable, R300_FPP_CMDSIZE, 0); - r300->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(R300_PFS_PARAM_0_X, 0); + r300->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_PFS_PARAM_0_X, 0); } ALLOC_STATE(fogs, always, R300_FOGS_CMDSIZE, 0); - r300->hw.fogs.cmd[R300_FOGS_CMD_0] = cmdpacket0(R300_FG_FOG_BLEND, 1); + r300->hw.fogs.cmd[R300_FOGS_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_FOG_BLEND, 1); ALLOC_STATE(fogc, always, R300_FOGC_CMDSIZE, 0); - r300->hw.fogc.cmd[R300_FOGC_CMD_0] = cmdpacket0(R300_FG_FOG_COLOR_R, 3); + r300->hw.fogc.cmd[R300_FOGC_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_FOG_COLOR_R, 3); ALLOC_STATE(at, always, R300_AT_CMDSIZE, 0); - r300->hw.at.cmd[R300_AT_CMD_0] = cmdpacket0(R300_FG_ALPHA_FUNC, 2); + r300->hw.at.cmd[R300_AT_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_ALPHA_FUNC, 2); ALLOC_STATE(fg_depth_src, always, 2, 0); - r300->hw.fg_depth_src.cmd[0] = cmdpacket0(R300_FG_DEPTH_SRC, 1); + r300->hw.fg_depth_src.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_FG_DEPTH_SRC, 1); ALLOC_STATE(rb3d_cctl, always, 2, 0); - r300->hw.rb3d_cctl.cmd[0] = cmdpacket0(R300_RB3D_CCTL, 1); + r300->hw.rb3d_cctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_CCTL, 1); ALLOC_STATE(bld, always, R300_BLD_CMDSIZE, 0); - r300->hw.bld.cmd[R300_BLD_CMD_0] = cmdpacket0(R300_RB3D_CBLEND, 2); + r300->hw.bld.cmd[R300_BLD_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_CBLEND, 2); ALLOC_STATE(cmk, always, R300_CMK_CMDSIZE, 0); - r300->hw.cmk.cmd[R300_CMK_CMD_0] = cmdpacket0(RB3D_COLOR_CHANNEL_MASK, 1); + r300->hw.cmk.cmd[R300_CMK_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, RB3D_COLOR_CHANNEL_MASK, 1); if (is_r500) { ALLOC_STATE(blend_color, always, 3, 0); - r300->hw.blend_color.cmd[0] = cmdpacket0(R500_RB3D_CONSTANT_COLOR_AR, 2); + r300->hw.blend_color.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_CONSTANT_COLOR_AR, 2); } else { ALLOC_STATE(blend_color, always, 2, 0); - r300->hw.blend_color.cmd[0] = cmdpacket0(R300_RB3D_BLEND_COLOR, 1); + r300->hw.blend_color.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_BLEND_COLOR, 1); } ALLOC_STATE(rop, always, 2, 0); - r300->hw.rop.cmd[0] = cmdpacket0(R300_RB3D_ROPCNTL, 1); + r300->hw.rop.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_ROPCNTL, 1); ALLOC_STATE(cb, always, R300_CB_CMDSIZE, 0); - r300->hw.cb.cmd[R300_CB_CMD_0] = cmdpacket0(R300_RB3D_COLOROFFSET0, 1); - r300->hw.cb.cmd[R300_CB_CMD_1] = cmdpacket0(R300_RB3D_COLORPITCH0, 1); + r300->hw.cb.emit = &emit_cb_offset; ALLOC_STATE(rb3d_dither_ctl, always, 10, 0); - r300->hw.rb3d_dither_ctl.cmd[0] = cmdpacket0(R300_RB3D_DITHER_CTL, 9); + r300->hw.rb3d_dither_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_DITHER_CTL, 9); ALLOC_STATE(rb3d_aaresolve_ctl, always, 2, 0); - r300->hw.rb3d_aaresolve_ctl.cmd[0] = cmdpacket0(R300_RB3D_AARESOLVE_CTL, 1); - ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0); - r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2); + r300->hw.rb3d_aaresolve_ctl.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_RB3D_AARESOLVE_CTL, 1); + ALLOC_STATE(rb3d_discard_src_pixel_lte_threshold, always, 3, 0); + r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 2); + r300->hw.rb3d_discard_src_pixel_lte_threshold.emit = emit_threshold_misc; ALLOC_STATE(zs, always, R300_ZS_CMDSIZE, 0); r300->hw.zs.cmd[R300_ZS_CMD_0] = - cmdpacket0(R300_ZB_CNTL, 3); + cmdpacket0(r300->radeon.radeonScreen, R300_ZB_CNTL, 3); + ALLOC_STATE(zstencil_format, always, 5, 0); r300->hw.zstencil_format.cmd[0] = - cmdpacket0(R300_ZB_FORMAT, 4); + cmdpacket0(r300->radeon.radeonScreen, R300_ZB_FORMAT, 4); + r300->hw.zstencil_format.emit = emit_zstencil_format; + ALLOC_STATE(zb, always, R300_ZB_CMDSIZE, 0); - r300->hw.zb.cmd[R300_ZB_CMD_0] = cmdpacket0(R300_ZB_DEPTHOFFSET, 2); + r300->hw.zb.emit = emit_zb_offset; ALLOC_STATE(zb_depthclearvalue, always, 2, 0); - r300->hw.zb_depthclearvalue.cmd[0] = cmdpacket0(R300_ZB_DEPTHCLEARVALUE, 1); - ALLOC_STATE(unk4F30, always, 3, 0); - r300->hw.unk4F30.cmd[0] = cmdpacket0(0x4F30, 2); + r300->hw.zb_depthclearvalue.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_DEPTHCLEARVALUE, 1); + ALLOC_STATE(zb_zmask, always, 3, 0); + r300->hw.zb_zmask.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_ZMASK_OFFSET, 2); ALLOC_STATE(zb_hiz_offset, always, 2, 0); - r300->hw.zb_hiz_offset.cmd[0] = cmdpacket0(R300_ZB_HIZ_OFFSET, 1); + r300->hw.zb_hiz_offset.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_HIZ_OFFSET, 1); ALLOC_STATE(zb_hiz_pitch, always, 2, 0); - r300->hw.zb_hiz_pitch.cmd[0] = cmdpacket0(R300_ZB_HIZ_PITCH, 1); + r300->hw.zb_hiz_pitch.cmd[0] = cmdpacket0(r300->radeon.radeonScreen, R300_ZB_HIZ_PITCH, 1); /* VPU only on TCL */ if (has_tcl) { int i; ALLOC_STATE(vpi, vpu, R300_VPI_CMDSIZE, 0); - r300->hw.vpi.cmd[R300_VPI_CMD_0] = - cmdvpu(R300_PVS_CODE_START, 0); + r300->hw.vpi.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R300_PVS_CODE_START, 0); + r300->hw.vpi.emit = emit_vpu; if (is_r500) { ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); - r300->hw.vpp.cmd[R300_VPP_CMD_0] = - cmdvpu(R500_PVS_CONST_START, 0); + r300->hw.vpp.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R500_PVS_CONST_START, 0); + r300->hw.vpp.emit = emit_vpu; ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); - r300->hw.vps.cmd[R300_VPS_CMD_0] = - cmdvpu(R500_POINT_VPORT_SCALE_OFFSET, 1); + r300->hw.vps.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R500_POINT_VPORT_SCALE_OFFSET, 1); + r300->hw.vps.emit = emit_vpu; for (i = 0; i < 6; i++) { - ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); - r300->hw.vpucp[i].cmd[R300_VPUCP_CMD_0] = - cmdvpu(R500_PVS_UCP_START + i, 1); + ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); + r300->hw.vpucp[i].cmd[0] = + cmdvpu(r300->radeon.radeonScreen, + R500_PVS_UCP_START + i, 1); + r300->hw.vpucp[i].emit = emit_vpu; } } else { ALLOC_STATE(vpp, vpu, R300_VPP_CMDSIZE, 0); - r300->hw.vpp.cmd[R300_VPP_CMD_0] = - cmdvpu(R300_PVS_CONST_START, 0); + r300->hw.vpp.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R300_PVS_CONST_START, 0); + r300->hw.vpp.emit = emit_vpu; ALLOC_STATE(vps, vpu, R300_VPS_CMDSIZE, 0); - r300->hw.vps.cmd[R300_VPS_CMD_0] = - cmdvpu(R300_POINT_VPORT_SCALE_OFFSET, 1); + r300->hw.vps.cmd[0] = + cmdvpu(r300->radeon.radeonScreen, R300_POINT_VPORT_SCALE_OFFSET, 1); + r300->hw.vps.emit = emit_vpu; for (i = 0; i < 6; i++) { ALLOC_STATE(vpucp[i], vpu, R300_VPUCP_CMDSIZE, 0); - r300->hw.vpucp[i].cmd[R300_VPUCP_CMD_0] = - cmdvpu(R300_PVS_UCP_START + i, 1); + r300->hw.vpucp[i].cmd[0] = + cmdvpu(r300->radeon.radeonScreen, + R300_PVS_UCP_START + i, 1); + r300->hw.vpucp[i].emit = emit_vpu; } } } @@ -556,130 +742,37 @@ void r300InitCmdBuf(r300ContextPtr r300) /* Textures */ ALLOC_STATE(tex.filter, variable, mtu + 1, 0); r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FILTER0_0, 0); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, 0); ALLOC_STATE(tex.filter_1, variable, mtu + 1, 0); r300->hw.tex.filter_1.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FILTER1_0, 0); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER1_0, 0); ALLOC_STATE(tex.size, variable, mtu + 1, 0); - r300->hw.tex.size.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_SIZE_0, 0); + r300->hw.tex.size.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_SIZE_0, 0); ALLOC_STATE(tex.format, variable, mtu + 1, 0); r300->hw.tex.format.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FORMAT_0, 0); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT_0, 0); ALLOC_STATE(tex.pitch, variable, mtu + 1, 0); - r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_FORMAT2_0, 0); + r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT2_0, 0); - ALLOC_STATE(tex.offset, variable, mtu + 1, 0); + ALLOC_STATE(tex.offset, variable, 1, 0); r300->hw.tex.offset.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_OFFSET_0, 0); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_OFFSET_0, 0); + r300->hw.tex.offset.emit = &emit_tex_offsets; ALLOC_STATE(tex.chroma_key, variable, mtu + 1, 0); r300->hw.tex.chroma_key.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_CHROMA_KEY_0, 0); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_CHROMA_KEY_0, 0); ALLOC_STATE(tex.border_color, variable, mtu + 1, 0); r300->hw.tex.border_color.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_BORDER_COLOR_0, 0); - - r300->hw.is_dirty = GL_TRUE; - r300->hw.all_dirty = GL_TRUE; - - /* Initialize command buffer */ - size = - 256 * driQueryOptioni(&r300->radeon.optionCache, - "command_buffer_size"); - if (size < 2 * r300->hw.max_state_size) { - size = 2 * r300->hw.max_state_size + 65535; - } - if (size > 64 * 256) - size = 64 * 256; - - if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) { - fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n", - sizeof(drm_r300_cmd_header_t)); - fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n", - sizeof(drm_radeon_cmd_buffer_t)); - fprintf(stderr, - "Allocating %d bytes command buffer (max state is %d bytes)\n", - size * 4, r300->hw.max_state_size * 4); - } - - r300->cmdbuf.size = size; - r300->cmdbuf.cmd_buf = (uint32_t *) CALLOC(size * 4); - r300->cmdbuf.count_used = 0; - r300->cmdbuf.count_reemit = 0; -} - -/** - * Destroy the command buffer and state atoms. - */ -void r300DestroyCmdBuf(r300ContextPtr r300) -{ - struct r300_state_atom *atom; - - FREE(r300->cmdbuf.cmd_buf); - - foreach(atom, &r300->hw.atomlist) { - FREE(atom->cmd); - } -} - -void r300EmitBlit(r300ContextPtr rmesa, - GLuint color_fmt, - GLuint src_pitch, - GLuint src_offset, - GLuint dst_pitch, - GLuint dst_offset, - GLint srcx, GLint srcy, - GLint dstx, GLint dsty, GLuint w, GLuint h) -{ - drm_r300_cmd_header_t *cmd; - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, - "%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n", - __FUNCTION__, src_pitch, src_offset, srcx, srcy, - dst_pitch, dst_offset, dstx, dsty, w, h); - - assert((src_pitch & 63) == 0); - assert((dst_pitch & 63) == 0); - assert((src_offset & 1023) == 0); - assert((dst_offset & 1023) == 0); - assert(w < (1 << 16)); - assert(h < (1 << 16)); - - cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 8, __FUNCTION__); - - cmd[0].header.cmd_type = R300_CMD_PACKET3; - cmd[0].header.pad0 = R300_CMD_PACKET3_RAW; - cmd[1].u = R300_CP_CMD_BITBLT_MULTI | (5 << 16); - cmd[2].u = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL | - RADEON_GMC_DST_PITCH_OFFSET_CNTL | - RADEON_GMC_BRUSH_NONE | - (color_fmt << 8) | - RADEON_GMC_SRC_DATATYPE_COLOR | - RADEON_ROP3_S | - RADEON_DP_SRC_SOURCE_MEMORY | - RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS); - - cmd[3].u = ((src_pitch / 64) << 22) | (src_offset >> 10); - cmd[4].u = ((dst_pitch / 64) << 22) | (dst_offset >> 10); - cmd[5].u = (srcx << 16) | srcy; - cmd[6].u = (dstx << 16) | dsty; /* dst */ - cmd[7].u = (w << 16) | h; -} - -void r300EmitWait(r300ContextPtr rmesa, GLuint flags) -{ - drm_r300_cmd_header_t *cmd; + cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, 0); - assert(!(flags & ~(R300_WAIT_2D | R300_WAIT_3D))); + r300->radeon.hw.is_dirty = GL_TRUE; + r300->radeon.hw.all_dirty = GL_TRUE; - cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__); - cmd[0].u = 0; - cmd[0].wait.cmd_type = R300_CMD_WAIT; - cmd[0].wait.flags = flags; + rcommonInitCmdBuf(&r300->radeon); } diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.h b/src/mesa/drivers/dri/r300/r300_cmdbuf.h index a8eaa580bd..53bcc0eeb4 100644 --- a/src/mesa/drivers/dri/r300/r300_cmdbuf.h +++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.h @@ -38,79 +38,14 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_context.h" -extern int r300FlushCmdBufLocked(r300ContextPtr r300, const char *caller); -extern int r300FlushCmdBuf(r300ContextPtr r300, const char *caller); - -extern void r300EmitState(r300ContextPtr r300); - extern void r300InitCmdBuf(r300ContextPtr r300); -extern void r300DestroyCmdBuf(r300ContextPtr r300); - -/** - * Make sure that enough space is available in the command buffer - * by flushing if necessary. - * - * \param dwords The number of dwords we need to be free on the command buffer - */ -static INLINE void r300EnsureCmdBufSpace(r300ContextPtr r300, - int dwords, const char *caller) -{ - assert(dwords < r300->cmdbuf.size); - - if (r300->cmdbuf.count_used + dwords > r300->cmdbuf.size) - r300FlushCmdBuf(r300, caller); -} - -/** - * Allocate the given number of dwords in the command buffer and return - * a pointer to the allocated area. - * When necessary, these functions cause a flush. r300AllocCmdBuf() also - * causes state reemission after a flush. This is necessary to ensure - * correct hardware state after an unlock. - */ -static INLINE uint32_t *r300RawAllocCmdBuf(r300ContextPtr r300, - int dwords, const char *caller) -{ - uint32_t *ptr; - - r300EnsureCmdBufSpace(r300, dwords, caller); - - ptr = &r300->cmdbuf.cmd_buf[r300->cmdbuf.count_used]; - r300->cmdbuf.count_used += dwords; - return ptr; -} - -static INLINE uint32_t *r300AllocCmdBuf(r300ContextPtr r300, - int dwords, const char *caller) -{ - uint32_t *ptr; - - r300EnsureCmdBufSpace(r300, dwords, caller); - - if (!r300->cmdbuf.count_used) { - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, - "Reemit state after flush (from %s)\n", caller); - r300EmitState(r300); - } - - ptr = &r300->cmdbuf.cmd_buf[r300->cmdbuf.count_used]; - r300->cmdbuf.count_used += dwords; - return ptr; -} +void r300_emit_scissor(GLcontext *ctx); -extern void r300EmitBlit(r300ContextPtr rmesa, - GLuint color_fmt, - GLuint src_pitch, - GLuint src_offset, - GLuint dst_pitch, - GLuint dst_offset, - GLint srcx, GLint srcy, - GLint dstx, GLint dsty, GLuint w, GLuint h); +void emit_vpu(GLcontext *ctx, struct radeon_state_atom * atom); +int check_vpu(GLcontext *ctx, struct radeon_state_atom *atom); -extern void r300EmitWait(r300ContextPtr rmesa, GLuint flags); -extern void r300EmitLOAD_VBPNTR(r300ContextPtr rmesa, int start); -extern void r300EmitVertexShader(r300ContextPtr rmesa); -extern void r300EmitPixelShader(r300ContextPtr rmesa); +void emit_r500fp(GLcontext *ctx, struct radeon_state_atom * atom); +int check_r500fp(GLcontext *ctx, struct radeon_state_atom *atom); +int check_r500fp_const(GLcontext *ctx, struct radeon_state_atom *atom); #endif /* __R300_CMDBUF_H__ */ diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c index 8f0effd83e..76881e4928 100644 --- a/src/mesa/drivers/dri/r300/r300_context.c +++ b/src/mesa/drivers/dri/r300/r300_context.c @@ -43,8 +43,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/matrix.h" #include "main/extensions.h" #include "main/state.h" -#include "main/texobj.h" #include "main/bufferobj.h" +#include "main/texobj.h" #include "swrast/swrast.h" #include "swrast_setup/swrast_setup.h" @@ -56,42 +56,40 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drivers/common/driverfuncs.h" -#include "radeon_ioctl.h" -#include "radeon_span.h" #include "r300_context.h" +#include "radeon_context.h" +#include "radeon_span.h" #include "r300_cmdbuf.h" #include "r300_state.h" #include "r300_ioctl.h" #include "r300_tex.h" #include "r300_emit.h" +#include "r300_render.h" #include "r300_swtcl.h" +#include "radeon_bocs_wrapper.h" -#ifdef USER_BUFFERS -#include "r300_mem.h" -#endif #include "vblank.h" #include "utils.h" #include "xmlpool.h" /* for symbolic values of enum-type options */ -/* hw_tcl_on derives from future_hw_tcl_on when its safe to change it. */ -int future_hw_tcl_on = 1; -int hw_tcl_on = 1; - #define need_GL_VERSION_2_0 #define need_GL_ARB_point_parameters #define need_GL_ARB_vertex_program #define need_GL_EXT_blend_equation_separate #define need_GL_EXT_blend_func_separate #define need_GL_EXT_blend_minmax +#define need_GL_EXT_framebuffer_object #define need_GL_EXT_fog_coord #define need_GL_EXT_gpu_program_parameters #define need_GL_EXT_secondary_color #define need_GL_EXT_stencil_two_side #define need_GL_ATI_separate_stencil #define need_GL_NV_vertex_program + #include "extension_helper.h" + const struct dri_extension card_extensions[] = { /* *INDENT-OFF* */ {"GL_ARB_depth_texture", NULL}, @@ -112,6 +110,7 @@ const struct dri_extension card_extensions[] = { {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions}, {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions}, {"GL_EXT_blend_subtract", NULL}, + {"GL_EXT_packed_depth_stencil", NULL}, {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions }, {"GL_EXT_gpu_program_parameters", GL_EXT_gpu_program_parameters_functions}, {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions}, @@ -125,6 +124,8 @@ const struct dri_extension card_extensions[] = { {"GL_EXT_texture_lod_bias", NULL}, {"GL_EXT_texture_mirror_clamp", NULL}, {"GL_EXT_texture_rectangle", NULL}, + {"GL_EXT_texture_sRGB", NULL}, + {"GL_EXT_vertex_array_bgra", NULL}, {"GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions}, {"GL_ATI_texture_env_combine3", NULL}, {"GL_ATI_texture_mirror_once", NULL}, @@ -139,6 +140,11 @@ const struct dri_extension card_extensions[] = { }; +const struct dri_extension mm_extensions[] = { + { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions }, + { NULL, NULL } +}; + /** * The GL 2.0 functions are needed to make display lists work with * functions added by GL_ATI_separate_stencil. @@ -147,16 +153,8 @@ const struct dri_extension gl_20_extension[] = { {"GL_VERSION_2_0", GL_VERSION_2_0_functions }, }; - -extern struct tnl_pipeline_stage _r300_render_stage; -extern const struct tnl_pipeline_stage _r300_tcl_stage; - static const struct tnl_pipeline_stage *r300_pipeline[] = { - /* Try and go straight to t&l - */ - &_r300_tcl_stage, - /* Catch any t&l fallbacks */ &_tnl_vertex_transform_stage, @@ -165,6 +163,7 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = { &_tnl_fog_coordinate_stage, &_tnl_texgen_stage, &_tnl_texture_transform_stage, + &_tnl_point_attenuation_stage, &_tnl_vertex_program_stage, /* Try again to go to tcl? @@ -184,6 +183,186 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = { 0, }; +static void r300_get_lock(radeonContextPtr rmesa) +{ + drm_radeon_sarea_t *sarea = rmesa->sarea; + + if (sarea->ctx_owner != rmesa->dri.hwContext) { + sarea->ctx_owner = rmesa->dri.hwContext; + if (!rmesa->radeonScreen->kernel_mm) + radeon_bo_legacy_texture_age(rmesa->radeonScreen->bom); + } +} + +static void r300_vtbl_emit_cs_header(struct radeon_cs *cs, radeonContextPtr rmesa) +{ + /* please flush pipe do all pending work */ + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_SC_SCREENDOOR, 1)); + radeon_cs_write_dword(cs, 0x0); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_SC_SCREENDOOR, 1)); + radeon_cs_write_dword(cs, 0x00FFFFFF); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_SC_HYPERZ, 1)); + radeon_cs_write_dword(cs, 0x0); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_US_CONFIG, 1)); + radeon_cs_write_dword(cs, 0x0); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_ZB_CNTL, 1)); + radeon_cs_write_dword(cs, 0x0); + radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen, R300_WAIT_3D)); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_RB3D_DSTCACHE_CTLSTAT, 1)); + radeon_cs_write_dword(cs, R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); + radeon_cs_write_dword(cs, cmdpacket0(rmesa->radeonScreen, + R300_ZB_ZCACHE_CTLSTAT, 1)); + radeon_cs_write_dword(cs, R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE); + radeon_cs_write_dword(cs, cmdwait(rmesa->radeonScreen, + R300_WAIT_3D | R300_WAIT_3D_CLEAN)); +} + +static void r300_vtbl_pre_emit_atoms(radeonContextPtr radeon) +{ + r300ContextPtr r300 = (r300ContextPtr)radeon; + BATCH_LOCALS(radeon); + + r300->vap_flush_needed = GL_TRUE; + + cp_wait(radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN); + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGVAL(R300_TX_INVALTAGS, R300_TX_FLUSH); + END_BATCH(); + end_3d(radeon); +} + +static void r300_fallback(GLcontext *ctx, GLuint bit, GLboolean mode) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + if (mode) + r300->radeon.Fallback |= bit; + else + r300->radeon.Fallback &= ~bit; +} + +static void r300_init_vtbl(radeonContextPtr radeon) +{ + radeon->vtbl.get_lock = r300_get_lock; + radeon->vtbl.update_viewport_offset = r300UpdateViewportOffset; + radeon->vtbl.emit_cs_header = r300_vtbl_emit_cs_header; + radeon->vtbl.swtcl_flush = r300_swtcl_flush; + radeon->vtbl.pre_emit_atoms = r300_vtbl_pre_emit_atoms; + radeon->vtbl.fallback = r300_fallback; +} + +static void r300InitConstValues(GLcontext *ctx, radeonScreenPtr screen) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + + ctx->Const.MaxTextureImageUnits = + driQueryOptioni(&r300->radeon.optionCache, "texture_image_units"); + ctx->Const.MaxTextureCoordUnits = + driQueryOptioni(&r300->radeon.optionCache, "texture_coord_units"); + ctx->Const.MaxTextureUnits = MIN2(ctx->Const.MaxTextureImageUnits, + ctx->Const.MaxTextureCoordUnits); + + ctx->Const.MaxTextureMaxAnisotropy = 16.0; + ctx->Const.MaxTextureLodBias = 16.0; + + if (screen->chip_family >= CHIP_FAMILY_RV515) + ctx->Const.MaxTextureLevels = 13; + else + ctx->Const.MaxTextureLevels = 12; + + ctx->Const.MinPointSize = 1.0; + ctx->Const.MinPointSizeAA = 1.0; + ctx->Const.MaxPointSize = R300_POINTSIZE_MAX; + ctx->Const.MaxPointSizeAA = R300_POINTSIZE_MAX; + + ctx->Const.MinLineWidth = 1.0; + ctx->Const.MinLineWidthAA = 1.0; + ctx->Const.MaxLineWidth = R300_LINESIZE_MAX; + ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX; + + ctx->Const.MaxDrawBuffers = 1; + + /* currently bogus data */ + if (r300->options.hw_tcl_enabled) { + ctx->Const.VertexProgram.MaxInstructions = VSF_MAX_FRAGMENT_LENGTH / 4; + ctx->Const.VertexProgram.MaxNativeInstructions = + VSF_MAX_FRAGMENT_LENGTH / 4; + ctx->Const.VertexProgram.MaxNativeAttribs = 16; /* r420 */ + ctx->Const.VertexProgram.MaxTemps = 32; + ctx->Const.VertexProgram.MaxNativeTemps = + /*VSF_MAX_FRAGMENT_TEMPS */ 32; + ctx->Const.VertexProgram.MaxNativeParameters = 256; /* r420 */ + ctx->Const.VertexProgram.MaxNativeAddressRegs = 1; + } + + if (screen->chip_family >= CHIP_FAMILY_RV515) { + ctx->Const.FragmentProgram.MaxNativeTemps = R500_PFS_NUM_TEMP_REGS; + ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* copy i915... */ + ctx->Const.FragmentProgram.MaxNativeParameters = R500_PFS_NUM_CONST_REGS; + ctx->Const.FragmentProgram.MaxNativeAluInstructions = R500_PFS_MAX_INST; + ctx->Const.FragmentProgram.MaxNativeTexInstructions = R500_PFS_MAX_INST; + ctx->Const.FragmentProgram.MaxNativeInstructions = R500_PFS_MAX_INST; + ctx->Const.FragmentProgram.MaxNativeTexIndirections = R500_PFS_MAX_INST; + ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; + } else { + ctx->Const.FragmentProgram.MaxNativeTemps = R300_PFS_NUM_TEMP_REGS; + ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* copy i915... */ + ctx->Const.FragmentProgram.MaxNativeParameters = R300_PFS_NUM_CONST_REGS; + ctx->Const.FragmentProgram.MaxNativeAluInstructions = R300_PFS_MAX_ALU_INST; + ctx->Const.FragmentProgram.MaxNativeTexInstructions = R300_PFS_MAX_TEX_INST; + ctx->Const.FragmentProgram.MaxNativeInstructions = R300_PFS_MAX_ALU_INST + R300_PFS_MAX_TEX_INST; + ctx->Const.FragmentProgram.MaxNativeTexIndirections = R300_PFS_MAX_TEX_INDIRECT; + ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; + } +} + +static void r300ParseOptions(r300ContextPtr r300, radeonScreenPtr screen) +{ + struct r300_options options = { 0 }; + + driParseConfigFiles(&r300->radeon.optionCache, &screen->optionCache, + screen->driScreen->myNum, "r300"); + + r300->radeon.initialMaxAnisotropy = driQueryOptionf(&r300->radeon.optionCache, "def_max_anisotropy"); + + options.stencil_two_side_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_stencil_two_side"); + options.s3tc_force_enabled = driQueryOptionb(&r300->radeon.optionCache, "force_s3tc_enable"); + options.s3tc_force_disabled = driQueryOptionb(&r300->radeon.optionCache, "disable_s3tc"); + + if (!(screen->chip_flags & RADEON_CHIPSET_TCL) || driQueryOptioni(&r300->radeon.optionCache, "tcl_mode") == DRI_CONF_TCL_SW) + options.hw_tcl_enabled = 0; + else + options.hw_tcl_enabled = 1; + + options.conformance_mode = !driQueryOptionb(&r300->radeon.optionCache, "disable_lowimpact_fallback"); + + r300->options = options; +} + +static void r300InitGLExtensions(GLcontext *ctx) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + + driInitExtensions(ctx, card_extensions, GL_TRUE); + if (r300->radeon.radeonScreen->kernel_mm) + driInitExtensions(ctx, mm_extensions, GL_FALSE); + + if (r300->options.stencil_two_side_disabled) + _mesa_disable_extension(ctx, "GL_EXT_stencil_two_side"); + + if (ctx->Mesa_DXTn && !r300->options.s3tc_force_enabled) { + _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); + _mesa_enable_extension(ctx, "GL_S3_s3tc"); + } else if (r300->options.s3tc_force_disabled) { + _mesa_disable_extension(ctx, "GL_EXT_texture_compression_s3tc"); + } +} + /* Create the device specific rendering context. */ GLboolean r300CreateContext(const __GLcontextModes * glVisual, @@ -195,42 +374,25 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, struct dd_function_table functions; r300ContextPtr r300; GLcontext *ctx; - int tcl_mode, i; assert(glVisual); assert(driContextPriv); assert(screen); - /* Allocate the R300 context */ r300 = (r300ContextPtr) CALLOC(sizeof(*r300)); if (!r300) return GL_FALSE; - if (!(screen->chip_flags & RADEON_CHIPSET_TCL)) - hw_tcl_on = future_hw_tcl_on = 0; + r300ParseOptions(r300, screen); - /* Parse configuration files. - * Do this here so that initialMaxAnisotropy is set before we create - * the default textures. - */ - driParseConfigFiles(&r300->radeon.optionCache, &screen->optionCache, - screen->driScreen->myNum, "r300"); - r300->initialMaxAnisotropy = driQueryOptionf(&r300->radeon.optionCache, - "def_max_anisotropy"); + r300_init_vtbl(&r300->radeon); - /* Init default driver functions then plug in our R300-specific functions - * (the texture functions are especially important) - */ _mesa_init_driver_functions(&functions); r300InitIoctlFuncs(&functions); r300InitStateFuncs(&functions); r300InitTextureFuncs(&functions); r300InitShaderFuncs(&functions); -#ifdef USER_BUFFERS - r300_mem_init(r300); -#endif - if (!radeonInitContext(&r300->radeon, &functions, glVisual, driContextPriv, sharedContextPrivate)) { @@ -238,94 +400,15 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, return GL_FALSE; } - /* Init r300 context data */ - r300->dma.buf0_address = - r300->radeon.radeonScreen->buffers->list[0].address; - - (void)memset(r300->texture_heaps, 0, sizeof(r300->texture_heaps)); - make_empty_list(&r300->swapped); - - r300->nr_heaps = 1 /* screen->numTexHeaps */ ; - assert(r300->nr_heaps < RADEON_NR_TEX_HEAPS); - for (i = 0; i < r300->nr_heaps; i++) { - /* *INDENT-OFF* */ - r300->texture_heaps[i] = driCreateTextureHeap(i, r300, - screen-> - texSize[i], 12, - RADEON_NR_TEX_REGIONS, - (drmTextureRegionPtr) - r300->radeon.sarea-> - tex_list[i], - &r300->radeon.sarea-> - tex_age[i], - &r300->swapped, - sizeof - (r300TexObj), - (destroy_texture_object_t - *) - r300DestroyTexObj); - /* *INDENT-ON* */ - } - r300->texture_depth = driQueryOptioni(&r300->radeon.optionCache, - "texture_depth"); - if (r300->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB) - r300->texture_depth = (screen->cpp == 4) ? - DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16; - - /* Set the maximum texture size small enough that we can guarentee that - * all texture units can bind a maximal texture and have them both in - * texturable memory at once. - */ - ctx = r300->radeon.glCtx; - ctx->Const.MaxTextureImageUnits = - driQueryOptioni(&r300->radeon.optionCache, "texture_image_units"); - ctx->Const.MaxTextureCoordUnits = - driQueryOptioni(&r300->radeon.optionCache, "texture_coord_units"); - ctx->Const.MaxTextureUnits = - MIN2(ctx->Const.MaxTextureImageUnits, - ctx->Const.MaxTextureCoordUnits); - ctx->Const.MaxTextureMaxAnisotropy = 16.0; - ctx->Const.MaxTextureLodBias = 16.0; - - if (screen->chip_family >= CHIP_FAMILY_RV515) - ctx->Const.MaxTextureLevels = 13; - else - ctx->Const.MaxTextureLevels = 12; - - driCalculateMaxTextureLevels( r300->texture_heaps, - r300->nr_heaps, - & ctx->Const, - 4, - ctx->Const.MaxTextureLevels - 1, - MIN2(ctx->Const.MaxTextureLevels, - MAX_3D_TEXTURE_LEVELS) - 1, - ctx->Const.MaxTextureLevels - 1, - ctx->Const.MaxTextureLevels - 1, - ctx->Const.MaxTextureLevels - 1, - GL_FALSE, - 2 ); - - ctx->Const.MinPointSize = 1.0; - ctx->Const.MinPointSizeAA = 1.0; - ctx->Const.MaxPointSize = R300_POINTSIZE_MAX; - ctx->Const.MaxPointSizeAA = R300_POINTSIZE_MAX; + r300->fallback = 0; + if (r300->options.hw_tcl_enabled) + ctx->VertexProgram._MaintainTnlProgram = GL_TRUE; - ctx->Const.MinLineWidth = 1.0; - ctx->Const.MinLineWidthAA = 1.0; - ctx->Const.MaxLineWidth = R300_LINESIZE_MAX; - ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX; - -#ifdef USER_BUFFERS - /* Needs further modifications */ -#if 0 - ctx->Const.MaxArrayLockSize = - ( /*512 */ RADEON_BUFFER_SIZE * 16 * 1024) / (4 * 4); -#endif -#endif + ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE; - ctx->Const.MaxDrawBuffers = 1; + r300InitConstValues(ctx, screen); /* Initialize the software rasterizer and helper modules. */ @@ -334,16 +417,12 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, _tnl_CreateContext(ctx); _swsetup_CreateContext(ctx); _swsetup_Wakeup(ctx); - _ae_create_context(ctx); /* Install the customized pipeline: */ _tnl_destroy_pipeline(ctx); _tnl_install_pipeline(ctx, r300_pipeline); - - /* Try and keep materials and vertices separate: - */ -/* _tnl_isolate_materials(ctx, GL_TRUE); */ + TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; /* Configure swrast and TNL to match hardware characteristics: */ @@ -352,226 +431,25 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual, _tnl_allow_pixel_fog(ctx, GL_FALSE); _tnl_allow_vertex_fog(ctx, GL_TRUE); - /* currently bogus data */ - if (screen->chip_flags & RADEON_CHIPSET_TCL) { - ctx->Const.VertexProgram.MaxInstructions = VSF_MAX_FRAGMENT_LENGTH / 4; - ctx->Const.VertexProgram.MaxNativeInstructions = - VSF_MAX_FRAGMENT_LENGTH / 4; - ctx->Const.VertexProgram.MaxNativeAttribs = 16; /* r420 */ - ctx->Const.VertexProgram.MaxTemps = 32; - ctx->Const.VertexProgram.MaxNativeTemps = - /*VSF_MAX_FRAGMENT_TEMPS */ 32; - ctx->Const.VertexProgram.MaxNativeParameters = 256; /* r420 */ - ctx->Const.VertexProgram.MaxNativeAddressRegs = 1; - } - - ctx->Const.FragmentProgram.MaxNativeTemps = PFS_NUM_TEMP_REGS; - ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* copy i915... */ - ctx->Const.FragmentProgram.MaxNativeParameters = PFS_NUM_CONST_REGS; - ctx->Const.FragmentProgram.MaxNativeAluInstructions = PFS_MAX_ALU_INST; - ctx->Const.FragmentProgram.MaxNativeTexInstructions = PFS_MAX_TEX_INST; - ctx->Const.FragmentProgram.MaxNativeInstructions = - PFS_MAX_ALU_INST + PFS_MAX_TEX_INST; - ctx->Const.FragmentProgram.MaxNativeTexIndirections = - PFS_MAX_TEX_INDIRECT; - ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; /* and these are?? */ - ctx->VertexProgram._MaintainTnlProgram = GL_TRUE; - ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE; - - driInitExtensions(ctx, card_extensions, GL_TRUE); - - if (driQueryOptionb - (&r300->radeon.optionCache, "disable_stencil_two_side")) - _mesa_disable_extension(ctx, "GL_EXT_stencil_two_side"); - - if (r300->radeon.glCtx->Mesa_DXTn - && !driQueryOptionb(&r300->radeon.optionCache, "disable_s3tc")) { - _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); - _mesa_enable_extension(ctx, "GL_S3_s3tc"); - } else - if (driQueryOptionb(&r300->radeon.optionCache, "force_s3tc_enable")) - { - _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); + if (r300->options.hw_tcl_enabled) { + r300InitDraw(ctx); + } else { + r300InitSwtcl(ctx); } - r300->disable_lowimpact_fallback = - driQueryOptionb(&r300->radeon.optionCache, - "disable_lowimpact_fallback"); - - radeonInitSpanFuncs(ctx); + radeon_fbo_init(&r300->radeon); + radeonInitSpanFuncs( ctx ); r300InitCmdBuf(r300); r300InitState(r300); - if (!(screen->chip_flags & RADEON_CHIPSET_TCL)) - r300InitSwtcl(ctx); + r300InitShaderFunctions(r300); - TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline; - - tcl_mode = driQueryOptioni(&r300->radeon.optionCache, "tcl_mode"); - if (driQueryOptionb(&r300->radeon.optionCache, "no_rast")) { - fprintf(stderr, "disabling 3D acceleration\n"); -#if R200_MERGED - FALLBACK(&r300->radeon, RADEON_FALLBACK_DISABLE, 1); -#endif - } - if (tcl_mode == DRI_CONF_TCL_SW || - !(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) { - if (r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { - r300->radeon.radeonScreen->chip_flags &= - ~RADEON_CHIPSET_TCL; - fprintf(stderr, "Disabling HW TCL support\n"); - } - TCL_FALLBACK(r300->radeon.glCtx, - RADEON_TCL_FALLBACK_TCL_DISABLE, 1); + if (screen->chip_family == CHIP_FAMILY_RS600 || screen->chip_family == CHIP_FAMILY_RS690 || + screen->chip_family == CHIP_FAMILY_RS740) { + r300->radeon.texture_row_align = 64; } - return GL_TRUE; -} - -static void r300FreeGartAllocations(r300ContextPtr r300) -{ - int i, ret, tries = 0, done_age, in_use = 0; - drm_radeon_mem_free_t memfree; + r300InitGLExtensions(ctx); - memfree.region = RADEON_MEM_REGION_GART; - -#ifdef USER_BUFFERS - for (i = r300->rmm->u_last; i > 0; i--) { - if (r300->rmm->u_list[i].ptr == NULL) { - continue; - } - - /* check whether this buffer is still in use */ - if (r300->rmm->u_list[i].pending) { - in_use++; - } - } - /* Cannot flush/lock if no context exists. */ - if (in_use) - r300FlushCmdBuf(r300, __FUNCTION__); - - done_age = radeonGetAge((radeonContextPtr) r300); - - for (i = r300->rmm->u_last; i > 0; i--) { - if (r300->rmm->u_list[i].ptr == NULL) { - continue; - } - - /* check whether this buffer is still in use */ - if (!r300->rmm->u_list[i].pending) { - continue; - } - - assert(r300->rmm->u_list[i].h_pending == 0); - - tries = 0; - while (r300->rmm->u_list[i].age > done_age && tries++ < 1000) { - usleep(10); - done_age = radeonGetAge((radeonContextPtr) r300); - } - if (tries >= 1000) { - WARN_ONCE("Failed to idle region!"); - } - - memfree.region_offset = (char *)r300->rmm->u_list[i].ptr - - (char *)r300->radeon.radeonScreen->gartTextures.map; - - ret = drmCommandWrite(r300->radeon.radeonScreen->driScreen->fd, - DRM_RADEON_FREE, &memfree, - sizeof(memfree)); - if (ret) { - fprintf(stderr, "Failed to free at %p\nret = %s\n", - r300->rmm->u_list[i].ptr, strerror(-ret)); - } else { - if (i == r300->rmm->u_last) - r300->rmm->u_last--; - - r300->rmm->u_list[i].pending = 0; - r300->rmm->u_list[i].ptr = NULL; - } - } - r300->rmm->u_head = i; -#endif /* USER_BUFFERS */ + return GL_TRUE; } -/* Destroy the device specific context. - */ -void r300DestroyContext(__DRIcontextPrivate * driContextPriv) -{ - GET_CURRENT_CONTEXT(ctx); - r300ContextPtr r300 = (r300ContextPtr) driContextPriv->driverPrivate; - radeonContextPtr radeon = (radeonContextPtr) r300; - radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL; - int i; - - if (RADEON_DEBUG & DEBUG_DRI) { - fprintf(stderr, "Destroying context !\n"); - } - - /* check if we're deleting the currently bound context */ - if (&r300->radeon == current) { - radeonFlush(r300->radeon.glCtx); - _mesa_make_current(NULL, NULL, NULL); - } - - /* Free r300 context resources */ - assert(r300); /* should never be null */ - - if (r300) { - GLboolean release_texture_heaps; - - release_texture_heaps = - (r300->radeon.glCtx->Shared->RefCount == 1); - _swsetup_DestroyContext(r300->radeon.glCtx); - _tnl_DestroyContext(r300->radeon.glCtx); - _vbo_DestroyContext(r300->radeon.glCtx); - _swrast_DestroyContext(r300->radeon.glCtx); - - if (r300->dma.current.buf) { - r300ReleaseDmaRegion(r300, &r300->dma.current, - __FUNCTION__); -#ifndef USER_BUFFERS - r300FlushCmdBuf(r300, __FUNCTION__); -#endif - } - r300FreeGartAllocations(r300); - r300DestroyCmdBuf(r300); - - if (radeon->state.scissor.pClipRects) { - FREE(radeon->state.scissor.pClipRects); - radeon->state.scissor.pClipRects = NULL; - } - - if (release_texture_heaps) { - /* This share group is about to go away, free our private - * texture object data. - */ - int i; - - for (i = 0; i < r300->nr_heaps; i++) { - driDestroyTextureHeap(r300->texture_heaps[i]); - r300->texture_heaps[i] = NULL; - } - - assert(is_empty_list(&r300->swapped)); - } - - /* Drop texture object references from current hardware state */ - for (i = 0; i < 8; i++) { - _mesa_reference_texobj(&r300->state.texture.unit[i].texobj, NULL); - } - - radeonCleanupContext(&r300->radeon); - -#ifdef USER_BUFFERS - /* the memory manager might be accessed when Mesa frees the shared - * state, so don't destroy it earlier - */ - r300_mem_destroy(r300); -#endif - - /* free the option cache */ - driDestroyOptionCache(&r300->radeon.optionCache); - - FREE(r300); - } -} diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h index 96a3205f1a..026c33c67c 100644 --- a/src/mesa/drivers/dri/r300/r300_context.h +++ b/src/mesa/drivers/dri/r300/r300_context.h @@ -37,26 +37,20 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #ifndef __R300_CONTEXT_H__ #define __R300_CONTEXT_H__ -#include "tnl/t_vertex.h" #include "drm.h" #include "radeon_drm.h" #include "dri_util.h" -#include "texmem.h" +#include "radeon_common.h" -#include "main/macros.h" #include "main/mtypes.h" -#include "main/colormac.h" - -#define USER_BUFFERS +#include "shader/prog_instruction.h" struct r300_context; typedef struct r300_context r300ContextRec; typedef struct r300_context *r300ContextPtr; -#include "radeon_lock.h" -#include "main/mm.h" -/* From http://gcc.gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html . +/* From http://gcc. gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html . I suppose we could inline this and use macro to fetch out __LINE__ and stuff in case we run into trouble with other compilers ... GLUE! */ @@ -73,180 +67,14 @@ typedef struct r300_context *r300ContextPtr; } #include "r300_vertprog.h" -#include "r500_fragprog.h" - -/** - * This function takes a float and packs it into a uint32_t - */ -static INLINE uint32_t r300PackFloat32(float fl) -{ - union { - float fl; - uint32_t u; - } u; - - u.fl = fl; - return u.u; -} - -/* This is probably wrong for some values, I need to test this - * some more. Range checking would be a good idea also.. - * - * But it works for most things. I'll fix it later if someone - * else with a better clue doesn't - */ -static INLINE uint32_t r300PackFloat24(float f) -{ - float mantissa; - int exponent; - uint32_t float24 = 0; - - if (f == 0.0) - return 0; - - mantissa = frexpf(f, &exponent); - - /* Handle -ve */ - if (mantissa < 0) { - float24 |= (1 << 23); - mantissa = mantissa * -1.0; - } - /* Handle exponent, bias of 63 */ - exponent += 62; - float24 |= (exponent << 16); - /* Kill 7 LSB of mantissa */ - float24 |= (r300PackFloat32(mantissa) & 0x7FFFFF) >> 7; - - return float24; -} - -/************ DMA BUFFERS **************/ - -/* Need refcounting on dma buffers: - */ -struct r300_dma_buffer { - int refcount; /**< the number of retained regions in buf */ - drmBufPtr buf; - int id; -}; -#undef GET_START -#ifdef USER_BUFFERS -#define GET_START(rvb) (r300GartOffsetFromVirtual(rmesa, (rvb)->address+(rvb)->start)) -#else -#define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset + \ - (rvb)->address - rmesa->dma.buf0_address + \ - (rvb)->start) -#endif -/* A retained region, eg vertices for indexed vertices. - */ -struct r300_dma_region { - struct r300_dma_buffer *buf; - char *address; /* == buf->address */ - int start, end, ptr; /* offsets from start of buf */ - - int aos_offset; /* address in GART memory */ - int aos_stride; /* distance between elements, in dwords */ - int aos_size; /* number of components (1-4) */ -}; - -struct r300_dma { - /* Active dma region. Allocations for vertices and retained - * regions come from here. Also used for emitting random vertices, - * these may be flushed by calling flush_current(); - */ - struct r300_dma_region current; - void (*flush) (r300ContextPtr); - - char *buf0_address; /* start of buf[0], for index calcs */ - - /* Number of "in-flight" DMA buffers, i.e. the number of buffers - * for which a DISCARD command is currently queued in the command buffer. - */ - GLuint nr_released_bufs; -}; - - /* Texture related */ - -typedef struct r300_tex_obj r300TexObj, *r300TexObjPtr; - -/* Maximum number of mipmap levels supported by any supported GPU - */ -#define R300_MAX_TEXTURE_LEVELS 13 - -/* Texture object in locally shared texture space. - */ -struct r300_tex_obj { - driTextureObject base; - - GLuint bufAddr; /* Offset to start of locally - shared texture block */ - - drm_radeon_tex_image_t image[6][R300_MAX_TEXTURE_LEVELS]; - /* Six, for the cube faces */ - - GLboolean image_override; /* Image overridden by GLX_EXT_tfp */ - - GLuint pitch; /* this isn't sent to hardware just used in calculations */ - /* hardware register values */ - /* Note that R200 has 8 registers per texture and R300 only 7 */ - GLuint filter; - GLuint filter_1; - GLuint pitch_reg; - GLuint size; /* npot only */ - GLuint format; - GLuint offset; /* Image location in the card's address space. - All cube faces follow. */ - GLuint unknown4; - GLuint unknown5; - /* end hardware registers */ - - /* registers computed by r200 code - keep them here to - compare against what is actually written. - - to be removed later.. */ - GLuint pp_border_color; - GLuint pp_cubic_faces; /* cube face 1,2,3,4 log2 sizes */ - GLuint format_x; - - GLboolean border_fallback; - - GLuint tile_bits; /* hw texture tile bits used on this texture */ -}; - -struct r300_texture_env_state { - struct gl_texture_object *texobj; - GLenum format; - GLenum envMode; -}; /* The blit width for texture uploads */ #define R300_BLIT_WIDTH_BYTES 1024 #define R300_MAX_TEXTURE_UNITS 8 -struct r300_texture_state { - struct r300_texture_env_state unit[R300_MAX_TEXTURE_UNITS]; - int tc_count; /* number of incoming texture coordinates from VAP */ -}; -/** - * A block of hardware state. - * - * When check returns non-zero, the returned number of dwords must be - * copied verbatim into the command buffer in order to update a state atom - * when it is dirty. - */ -struct r300_state_atom { - struct r300_state_atom *next, *prev; - const char *name; /* for debug */ - int cmd_size; /* maximum size in dwords */ - GLuint idx; /* index in an array (e.g. textures) */ - uint32_t *cmd; - GLboolean dirty; - - int (*check) (r300ContextPtr, struct r300_state_atom * atom); -}; #define R300_VPT_CMD_0 0 #define R300_VPT_XSCALE 1 @@ -288,9 +116,11 @@ struct r300_state_atom { #define R300_GB_MISC_MSPOS_0 1 #define R300_GB_MISC_MSPOS_1 2 #define R300_GB_MISC_TILE_CONFIG 3 -#define R300_GB_MISC_SELECT 4 -#define R300_GB_MISC_AA_CONFIG 5 -#define R300_GB_MISC_CMDSIZE 6 +#define R300_GB_MISC_CMDSIZE 4 +#define R300_GB_MISC2_CMD_0 0 +#define R300_GB_MISC2_SELECT 1 +#define R300_GB_MISC2_AA_CONFIG 2 +#define R300_GB_MISC2_CMDSIZE 3 #define R300_TXE_CMD_0 0 #define R300_TXE_ENABLE 1 @@ -463,124 +293,100 @@ struct r300_state_atom { * Cache for hardware register state. */ struct r300_hw_state { - struct r300_state_atom atomlist; - - GLboolean is_dirty; - GLboolean all_dirty; - int max_state_size; /* in dwords */ - - struct r300_state_atom vpt; /* viewport (1D98) */ - struct r300_state_atom vap_cntl; - struct r300_state_atom vap_index_offset; /* 0x208c r5xx only */ - struct r300_state_atom vof; /* VAP output format register 0x2090 */ - struct r300_state_atom vte; /* (20B0) */ - struct r300_state_atom vap_vf_max_vtx_indx; /* Maximum Vertex Indx Clamp (2134) */ - struct r300_state_atom vap_cntl_status; - struct r300_state_atom vir[2]; /* vap input route (2150/21E0) */ - struct r300_state_atom vic; /* vap input control (2180) */ - struct r300_state_atom vap_psc_sgn_norm_cntl; /* Programmable Stream Control Signed Normalize Control (21DC) */ - struct r300_state_atom vap_clip_cntl; - struct r300_state_atom vap_clip; - struct r300_state_atom vap_pvs_vtx_timeout_reg; /* Vertex timeout register (2288) */ - struct r300_state_atom pvs; /* pvs_cntl (22D0) */ - struct r300_state_atom gb_enable; /* (4008) */ - struct r300_state_atom gb_misc; /* Multisampling position shifts ? (4010) */ - struct r300_state_atom ga_point_s0; /* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) (4200) */ - struct r300_state_atom ga_triangle_stipple; /* (4214) */ - struct r300_state_atom ps; /* pointsize (421C) */ - struct r300_state_atom ga_point_minmax; /* (4230) */ - struct r300_state_atom lcntl; /* line control */ - struct r300_state_atom ga_line_stipple; /* (4260) */ - struct r300_state_atom shade; - struct r300_state_atom polygon_mode; - struct r300_state_atom fogp; /* fog parameters (4294) */ - struct r300_state_atom ga_soft_reset; /* (429C) */ - struct r300_state_atom zbias_cntl; - struct r300_state_atom zbs; /* zbias (42A4) */ - struct r300_state_atom occlusion_cntl; - struct r300_state_atom cul; /* cull cntl (42B8) */ - struct r300_state_atom su_depth_scale; /* (42C0) */ - struct r300_state_atom rc; /* rs control (4300) */ - struct r300_state_atom ri; /* rs interpolators (4310) */ - struct r300_state_atom rr; /* rs route (4330) */ - struct r300_state_atom sc_hyperz; /* (43A4) */ - struct r300_state_atom sc_screendoor; /* (43E8) */ - struct r300_state_atom fp; /* fragment program cntl + nodes (4600) */ - struct r300_state_atom fpt; /* texi - (4620) */ - struct r300_state_atom us_out_fmt; /* (46A4) */ - struct r300_state_atom r500fp; /* r500 fp instructions */ - struct r300_state_atom r500fp_const; /* r500 fp constants */ - struct r300_state_atom fpi[4]; /* fp instructions (46C0/47C0/48C0/49C0) */ - struct r300_state_atom fogs; /* fog state (4BC0) */ - struct r300_state_atom fogc; /* fog color (4BC8) */ - struct r300_state_atom at; /* alpha test (4BD4) */ - struct r300_state_atom fg_depth_src; /* (4BD8) */ - struct r300_state_atom fpp; /* 0x4C00 and following */ - struct r300_state_atom rb3d_cctl; /* (4E00) */ - struct r300_state_atom bld; /* blending (4E04) */ - struct r300_state_atom cmk; /* colormask (4E0C) */ - struct r300_state_atom blend_color; /* constant blend color */ - struct r300_state_atom rop; /* ropcntl */ - struct r300_state_atom cb; /* colorbuffer (4E28) */ - struct r300_state_atom rb3d_dither_ctl; /* (4E50) */ - struct r300_state_atom rb3d_aaresolve_ctl; /* (4E88) */ - struct r300_state_atom rb3d_discard_src_pixel_lte_threshold; /* (4E88) I saw it only written on RV350 hardware.. */ - struct r300_state_atom zs; /* zstencil control (4F00) */ - struct r300_state_atom zstencil_format; - struct r300_state_atom zb; /* z buffer (4F20) */ - struct r300_state_atom zb_depthclearvalue; /* (4F28) */ - struct r300_state_atom unk4F30; /* (4F30) */ - struct r300_state_atom zb_hiz_offset; /* (4F44) */ - struct r300_state_atom zb_hiz_pitch; /* (4F54) */ - - struct r300_state_atom vpi; /* vp instructions */ - struct r300_state_atom vpp; /* vp parameters */ - struct r300_state_atom vps; /* vertex point size (?) */ - struct r300_state_atom vpucp[6]; /* vp user clip plane - 6 */ + struct radeon_state_atom vpt; /* viewport (1D98) */ + struct radeon_state_atom vap_cntl; + struct radeon_state_atom vap_index_offset; /* 0x208c r5xx only */ + struct radeon_state_atom vof; /* VAP output format register 0x2090 */ + struct radeon_state_atom vte; /* (20B0) */ + struct radeon_state_atom vap_vf_max_vtx_indx; /* Maximum Vertex Indx Clamp (2134) */ + struct radeon_state_atom vap_cntl_status; + struct radeon_state_atom vir[2]; /* vap input route (2150/21E0) */ + struct radeon_state_atom vic; /* vap input control (2180) */ + struct radeon_state_atom vap_psc_sgn_norm_cntl; /* Programmable Stream Control Signed Normalize Control (21DC) */ + struct radeon_state_atom vap_clip_cntl; + struct radeon_state_atom vap_clip; + struct radeon_state_atom vap_pvs_vtx_timeout_reg; /* Vertex timeout register (2288) */ + struct radeon_state_atom pvs; /* pvs_cntl (22D0) */ + struct radeon_state_atom gb_enable; /* (4008) */ + struct radeon_state_atom gb_misc; /* Multisampling position shifts ? (4010) */ + struct radeon_state_atom gb_misc2; /* Multisampling position shifts ? (4010) */ + struct radeon_state_atom ga_point_s0; /* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) (4200) */ + struct radeon_state_atom ga_triangle_stipple; /* (4214) */ + struct radeon_state_atom ps; /* pointsize (421C) */ + struct radeon_state_atom ga_point_minmax; /* (4230) */ + struct radeon_state_atom lcntl; /* line control */ + struct radeon_state_atom ga_line_stipple; /* (4260) */ + struct radeon_state_atom shade; + struct radeon_state_atom shade2; + struct radeon_state_atom polygon_mode; + struct radeon_state_atom fogp; /* fog parameters (4294) */ + struct radeon_state_atom ga_soft_reset; /* (429C) */ + struct radeon_state_atom zbias_cntl; + struct radeon_state_atom zbs; /* zbias (42A4) */ + struct radeon_state_atom occlusion_cntl; + struct radeon_state_atom cul; /* cull cntl (42B8) */ + struct radeon_state_atom su_depth_scale; /* (42C0) */ + struct radeon_state_atom rc; /* rs control (4300) */ + struct radeon_state_atom ri; /* rs interpolators (4310) */ + struct radeon_state_atom rr; /* rs route (4330) */ + struct radeon_state_atom sc_hyperz; /* (43A4) */ + struct radeon_state_atom sc_screendoor; /* (43E8) */ + struct radeon_state_atom fp; /* fragment program cntl + nodes (4600) */ + struct radeon_state_atom fpt; /* texi - (4620) */ + struct radeon_state_atom us_out_fmt; /* (46A4) */ + struct radeon_state_atom r500fp; /* r500 fp instructions */ + struct radeon_state_atom r500fp_const; /* r500 fp constants */ + struct radeon_state_atom fpi[4]; /* fp instructions (46C0/47C0/48C0/49C0) */ + struct radeon_state_atom fogs; /* fog state (4BC0) */ + struct radeon_state_atom fogc; /* fog color (4BC8) */ + struct radeon_state_atom at; /* alpha test (4BD4) */ + struct radeon_state_atom fg_depth_src; /* (4BD8) */ + struct radeon_state_atom fpp; /* 0x4C00 and following */ + struct radeon_state_atom rb3d_cctl; /* (4E00) */ + struct radeon_state_atom bld; /* blending (4E04) */ + struct radeon_state_atom cmk; /* colormask (4E0C) */ + struct radeon_state_atom blend_color; /* constant blend color */ + struct radeon_state_atom rop; /* ropcntl */ + struct radeon_state_atom cb; /* colorbuffer (4E28) */ + struct radeon_state_atom rb3d_dither_ctl; /* (4E50) */ + struct radeon_state_atom rb3d_aaresolve_ctl; /* (4E88) */ + struct radeon_state_atom rb3d_discard_src_pixel_lte_threshold; /* (4E88) I saw it only written on RV350 hardware.. */ + struct radeon_state_atom zs; /* zstencil control (4F00) */ + struct radeon_state_atom zstencil_format; + struct radeon_state_atom zb; /* z buffer (4F20) */ + struct radeon_state_atom zb_depthclearvalue; /* (4F28) */ + struct radeon_state_atom zb_zmask; /* (4F30) */ + struct radeon_state_atom zb_hiz_offset; /* (4F44) */ + struct radeon_state_atom zb_hiz_pitch; /* (4F54) */ + + struct radeon_state_atom vpi; /* vp instructions */ + struct radeon_state_atom vpp; /* vp parameters */ + struct radeon_state_atom vps; /* vertex point size (?) */ + struct radeon_state_atom vpucp[6]; /* vp user clip plane - 6 */ /* 8 texture units */ /* the state is grouped by function and not by texture unit. This makes single unit updates really awkward - we are much better off updating the whole thing at once */ struct { - struct r300_state_atom filter; - struct r300_state_atom filter_1; - struct r300_state_atom size; - struct r300_state_atom format; - struct r300_state_atom pitch; - struct r300_state_atom offset; - struct r300_state_atom chroma_key; - struct r300_state_atom border_color; + struct radeon_state_atom filter; + struct radeon_state_atom filter_1; + struct radeon_state_atom size; + struct radeon_state_atom format; + struct radeon_state_atom pitch; + struct radeon_state_atom offset; + struct radeon_state_atom chroma_key; + struct radeon_state_atom border_color; } tex; - struct r300_state_atom txe; /* tex enable (4104) */ -}; + struct radeon_state_atom txe; /* tex enable (4104) */ -/** - * This structure holds the command buffer while it is being constructed. - * - * The first batch of commands in the buffer is always the state that needs - * to be re-emitted when the context is lost. This batch can be skipped - * otherwise. - */ -struct r300_cmdbuf { - int size; /* DWORDs allocated for buffer */ - uint32_t *cmd_buf; - int count_used; /* DWORDs filled so far */ - int count_reemit; /* size of re-emission batch */ + radeonTexObj *textures[R300_MAX_TEXTURE_UNITS]; }; /** * State cache */ -struct r300_depthbuffer_state { - GLfloat scale; -}; - -struct r300_stencilbuffer_state { - GLboolean hw_stencil; -}; - /* Vertex shader state */ /* Perhaps more if we store programs in vmem? */ @@ -593,73 +399,55 @@ struct r300_stencilbuffer_state { #define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0) #define STATE_R300_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1) -struct r300_vertex_shader_fragment { - int length; - union { - GLuint d[VSF_MAX_FRAGMENT_LENGTH]; - float f[VSF_MAX_FRAGMENT_LENGTH]; - GLuint i[VSF_MAX_FRAGMENT_LENGTH]; - } body; -}; - -struct r300_vertex_shader_state { - struct r300_vertex_shader_fragment program; -}; - -extern int hw_tcl_on; - #define COLOR_IS_RGBA #define TAG(x) r300##x #include "tnl_dd/t_dd_vertex.h" #undef TAG -//#define CURRENT_VERTEX_SHADER(ctx) (ctx->VertexProgram._Current) -#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->selected_vp) - -/* Should but doesnt work */ -//#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->curr_vp) - -/* r300_vertex_shader_state and r300_vertex_program should probably be merged together someday. - * Keeping them them seperate for now should ensure fixed pipeline keeps functioning properly. - */ - -struct r300_vertex_program_key { - GLuint InputsRead; - GLuint OutputsWritten; - GLuint OutputsAdded; -}; - struct r300_vertex_program { struct r300_vertex_program *next; - struct r300_vertex_program_key key; - int translated; - struct r300_vertex_shader_fragment program; + struct r300_vertex_program_key { + GLuint InputsRead; + GLuint OutputsWritten; + GLuint OutputsAdded; + } key; + + struct r300_vertex_shader_hw_code { + int length; + union { + GLuint d[VSF_MAX_FRAGMENT_LENGTH]; + float f[VSF_MAX_FRAGMENT_LENGTH]; + } body; + } hw_code; + + GLboolean translated; + GLboolean error; int pos_end; int num_temporaries; /* Number of temp vars used by program */ int wpos_idx; int inputs[VERT_ATTRIB_MAX]; int outputs[VERT_RESULT_MAX]; - int native; - int ref_count; - int use_ref_count; }; struct r300_vertex_program_cont { struct gl_vertex_program mesa_program; /* Must be first */ - struct r300_vertex_shader_fragment params; struct r300_vertex_program *progs; }; -#define PFS_MAX_ALU_INST 64 -#define PFS_MAX_TEX_INST 64 -#define PFS_MAX_TEX_INDIRECT 4 -#define PFS_NUM_TEMP_REGS 32 -#define PFS_NUM_CONST_REGS 16 +#define R300_PFS_MAX_ALU_INST 64 +#define R300_PFS_MAX_TEX_INST 32 +#define R300_PFS_MAX_TEX_INDIRECT 4 +#define R300_PFS_NUM_TEMP_REGS 32 +#define R300_PFS_NUM_CONST_REGS 32 -struct r300_pfs_compile_state; +#define R500_PFS_MAX_INST 512 +#define R500_PFS_NUM_TEMP_REGS 128 +#define R500_PFS_NUM_CONST_REGS 256 +struct r300_pfs_compile_state; +struct r500_pfs_compile_state; /** * Stores state that influences the compilation of a fragment program. @@ -702,7 +490,7 @@ struct r300_fragment_program_node { struct r300_fragment_program_code { struct { int length; /**< total # of texture instructions used */ - GLuint inst[PFS_MAX_TEX_INST]; + GLuint inst[R300_PFS_MAX_TEX_INST]; } tex; struct { @@ -712,7 +500,7 @@ struct r300_fragment_program_code { GLuint inst1; GLuint inst2; GLuint inst3; - } inst[PFS_MAX_ALU_INST]; + } inst[R300_PFS_MAX_ALU_INST]; } alu; struct r300_fragment_program_node node[4]; @@ -723,53 +511,12 @@ struct r300_fragment_program_code { * Remember which program register a given hardware constant * belongs to. */ - struct prog_src_register constant[PFS_NUM_CONST_REGS]; + struct prog_src_register constant[R300_PFS_NUM_CONST_REGS]; int const_nr; int max_temp_idx; }; -/** - * Store everything about a fragment program that is needed - * to render with that program. - */ -struct r300_fragment_program { - struct gl_fragment_program mesa_program; - - GLboolean translated; - GLboolean error; - - struct r300_fragment_program_external_state state; - struct r300_fragment_program_code code; - - GLboolean WritesDepth; - GLuint optimization; -}; - -struct r500_pfs_compile_state; - -struct r500_fragment_program_external_state { - struct { - /** - * If the sampler is used as a shadow sampler, - * this field is: - * 0 - GL_LUMINANCE - * 1 - GL_INTENSITY - * 2 - GL_ALPHA - * depending on the depth texture mode. - */ - GLuint depth_texture_mode : 2; - - /** - * If the sampler is used as a shadow sampler, - * this field is (texture_compare_func - GL_NEVER). - * [e.g. if compare function is GL_LEQUAL, this field is 3] - * - * Otherwise, this field is 0. - */ - GLuint texture_compare_func : 3; - } unit[16]; -}; struct r500_fragment_program_code { struct { @@ -779,7 +526,7 @@ struct r500_fragment_program_code { GLuint inst3; GLuint inst4; GLuint inst5; - } inst[512]; + } inst[R500_PFS_MAX_INST]; int inst_offset; int inst_end; @@ -788,94 +535,46 @@ struct r500_fragment_program_code { * Remember which program register a given hardware constant * belongs to. */ - struct prog_src_register constant[PFS_NUM_CONST_REGS]; + struct prog_src_register constant[R500_PFS_NUM_CONST_REGS]; int const_nr; int max_temp_idx; }; -struct r500_fragment_program { - struct gl_fragment_program mesa_program; +/** +* Store everything about a fragment program that is needed +* to render with that program. +*/ +struct r300_fragment_program { + struct gl_fragment_program Base; - GLcontext *ctx; GLboolean translated; GLboolean error; - struct r500_fragment_program_external_state state; - struct r500_fragment_program_code code; + struct r300_fragment_program_external_state state; + union rX00_fragment_program_code { + struct r300_fragment_program_code r300; + struct r500_fragment_program_code r500; + } code; GLboolean writes_depth; - GLuint optimization; }; -#define R300_MAX_AOS_ARRAYS 16 - -#define REG_COORDS 0 -#define REG_COLOR0 1 -#define REG_TEX0 2 - -struct r300_state { - struct r300_depthbuffer_state depth; - struct r300_texture_state texture; - int sw_tcl_inputs[VERT_ATTRIB_MAX]; - struct r300_vertex_shader_state vertex_shader; - struct r300_dma_region aos[R300_MAX_AOS_ARRAYS]; - int aos_count; - - GLuint *Elts; - struct r300_dma_region elt_dma; - - struct r300_dma_region swtcl_dma; - DECLARE_RENDERINPUTS(render_inputs_bitset); /* actual render inputs that R300 was configured for. - They are the same as tnl->render_inputs for fixed pipeline */ - - struct r300_stencilbuffer_state stencil; - +struct r300_fragment_program_compiler { + r300ContextPtr r300; + struct r300_fragment_program *fp; + union rX00_fragment_program_code *code; + struct gl_program *program; }; -#define R300_FALLBACK_NONE 0 -#define R300_FALLBACK_TCL 1 -#define R300_FALLBACK_RAST 2 +#define R300_MAX_AOS_ARRAYS 16 + /* r300_swtcl.c */ struct r300_swtcl_info { - GLuint RenderIndex; - - /** - * Size of a hardware vertex. This is calculated when \c ::vertex_attrs is - * installed in the Mesa state vector. - */ - GLuint vertex_size; - - /** - * Attributes instructing the Mesa TCL pipeline where / how to put vertex - * data in the hardware buffer. - */ - struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX]; - - /** - * Number of elements of \c ::vertex_attrs that are actually used. - */ - GLuint vertex_attr_count; - - /** - * Cached pointer to the buffer where Mesa will store vertex data. - */ - GLubyte *verts; - - /* Fallback rasterization functions - */ - // r200_point_func draw_point; - // r200_line_func draw_line; - // r200_tri_func draw_tri; - - GLuint hw_primitive; - GLenum render_primitive; - GLuint numverts; - - /** + /* * Offset of the 4UB color data within a hardware (swtcl) vertex. */ GLuint coloroffset; @@ -884,15 +583,44 @@ struct r300_swtcl_info { * Offset of the 3UB specular color data within a hardware (swtcl) vertex. */ GLuint specoffset; +}; - /** - * Should Mesa project vertex data or will the hardware do it? - */ - GLboolean needproj; +struct r300_vtable { + void (* SetupRSUnit)(GLcontext *ctx); + void (* SetupFragmentShaderTextures)(GLcontext *ctx, int *tmu_mappings); + GLboolean (* BuildFragmentProgramHwCode)(struct r300_fragment_program_compiler *compiler); + void (* FragmentProgramDump)(union rX00_fragment_program_code *code); + void (* SetupPixelShader)(GLcontext *ctx); +}; - struct r300_dma_region indexed_verts; +struct r300_vertex_buffer { + struct vertex_attribute { + /* generic */ + GLubyte element; + GLvoid *data; + GLboolean free_needed; + GLuint stride; + GLuint dwords; + GLubyte size; /* number of components */ + + /* hw specific */ + uint32_t data_type:4; + uint32_t dst_loc:5; + uint32_t _signed:1; + uint32_t normalize:1; + uint32_t swizzle:12; + uint32_t write_mask:4; + } attribs[VERT_ATTRIB_MAX]; + + GLubyte num_attribs; }; +struct r300_index_buffer { + GLvoid *ptr; + GLboolean is_32bit; + GLboolean free_needed; + GLuint count; +}; /** * \brief R300 context structure. @@ -900,46 +628,33 @@ struct r300_swtcl_info { struct r300_context { struct radeon_context radeon; /* parent class, must be first */ + struct r300_vtable vtbl; + struct r300_hw_state hw; - struct r300_cmdbuf cmdbuf; - struct r300_state state; - struct gl_vertex_program *curr_vp; + struct r300_vertex_program *selected_vp; /* Vertex buffers */ - struct r300_dma dma; - GLboolean save_on_next_unlock; - GLuint NewGLState; - - /* Texture object bookkeeping - */ - unsigned nr_heaps; - driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS]; - driTextureObject swapped; - int texture_depth; - float initialMaxAnisotropy; - - /* Clientdata textures; - */ - GLuint prefer_gart_client_texturing; - -#ifdef USER_BUFFERS - struct r300_memory_manager *rmm; -#endif - GLvector4f dummy_attrib[_TNL_ATTRIB_MAX]; GLvector4f *temp_attrib[_TNL_ATTRIB_MAX]; - GLboolean disable_lowimpact_fallback; - - DECLARE_RENDERINPUTS(tnl_index_bitset); /* index of bits for last tnl_install_attrs */ + struct r300_options { + uint32_t conformance_mode:1; + uint32_t hw_tcl_enabled:1; + uint32_t s3tc_force_enabled:1; + uint32_t s3tc_force_disabled:1; + uint32_t stencil_two_side_disabled:1; + } options; + struct r300_swtcl_info swtcl; -}; + struct r300_vertex_buffer vbuf; + struct r300_index_buffer ind_buf; + GLboolean vap_flush_needed; + + uint32_t fallback; -struct r300_buffer_object { - struct gl_buffer_object mesa_obj; - int id; + DECLARE_RENDERINPUTS(render_inputs_bitset); }; #define R300_CONTEXT(ctx) ((r300ContextPtr)(ctx->DriverCtx)) @@ -955,9 +670,11 @@ extern int r300VertexProgUpdateParams(GLcontext * ctx, struct r300_vertex_program_cont *vp, float *dst); -#define RADEON_D_CAPTURE 0 -#define RADEON_D_PLAYBACK 1 -#define RADEON_D_PLAYBACK_RAW 2 -#define RADEON_D_T 3 +extern void r300InitShaderFunctions(r300ContextPtr r300); + +extern void r300InitDraw(GLcontext *ctx); + +#define r300PackFloat32 radeonPackFloat32 +#define r300PackFloat24 radeonPackFloat24 #endif /* __R300_CONTEXT_H__ */ diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c new file mode 100644 index 0000000000..cc5650fb7c --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_draw.c @@ -0,0 +1,475 @@ +/************************************************************************** + * + * Copyright 2009 Maciej Cencora + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL THE AUTHOR(S) AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include <stdlib.h> + +#include "main/glheader.h" +#include "main/context.h" +#include "main/state.h" +#include "main/api_validate.h" +#include "main/enums.h" + +#include "r300_reg.h" +#include "r300_context.h" +#include "r300_emit.h" +#include "r300_render.h" +#include "r300_state.h" +#include "r300_tex.h" + +#include "tnl/tnl.h" +#include "tnl/t_vp_build.h" +#include "vbo/vbo_context.h" +#include "swrast/swrast.h" +#include "swrast_setup/swrast_setup.h" + +static void r300FixupIndexBuffer(GLcontext *ctx, const struct _mesa_index_buffer *mesa_ind_buf, struct gl_buffer_object **bo, GLuint *nr_bo) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_index_buffer *ind_buf = &r300->ind_buf; + GLvoid *src_ptr; + + if (!mesa_ind_buf) { + ind_buf->ptr = NULL; + return; + } + + ind_buf->count = mesa_ind_buf->count; + if (mesa_ind_buf->obj->Name && !mesa_ind_buf->obj->Pointer) { + bo[*nr_bo] = mesa_ind_buf->obj; + (*nr_bo)++; + ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY_ARB, mesa_ind_buf->obj); + assert(mesa_ind_buf->obj->Pointer != NULL); + } + src_ptr = ADD_POINTERS(mesa_ind_buf->obj->Pointer, mesa_ind_buf->ptr); + + if (mesa_ind_buf->type == GL_UNSIGNED_BYTE) { + GLubyte *in = (GLubyte *)src_ptr; + GLuint *out = _mesa_malloc(sizeof(GLushort) * ((mesa_ind_buf->count + 1) & ~1)); + int i; + + ind_buf->ptr = out; + + for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) { + *out++ = in[i] | in[i + 1] << 16; + } + + if (i < mesa_ind_buf->count) { + *out++ = in[i]; + } + + ind_buf->free_needed = GL_TRUE; + ind_buf->is_32bit = GL_FALSE; + } else if (mesa_ind_buf->type == GL_UNSIGNED_SHORT) { +#if MESA_BIG_ENDIAN + GLushort *in = (GLushort *)src_ptr; + GLuint *out = _mesa_malloc(sizeof(GLushort) * + ((mesa_ind_buf->count + 1) & ~1)); + int i; + + ind_buf->ptr = out; + + for (i = 0; i + 1 < mesa_ind_buf->count; i += 2) { + *out++ = in[i] | in[i + 1] << 16; + } + + if (i < mesa_ind_buf->count) { + *out++ = in[i]; + } + + ind_buf->free_needed = GL_TRUE; +#else + ind_buf->ptr = src_ptr; + ind_buf->free_needed = GL_FALSE; +#endif + ind_buf->is_32bit = GL_FALSE; + } else { + ind_buf->ptr = src_ptr; + ind_buf->free_needed = GL_FALSE; + ind_buf->is_32bit = GL_TRUE; + } +} + +static int getTypeSize(GLenum type) +{ + switch (type) { + case GL_DOUBLE: + return sizeof(GLdouble); + case GL_FLOAT: + return sizeof(GLfloat); + case GL_INT: + return sizeof(GLint); + case GL_UNSIGNED_INT: + return sizeof(GLuint); + case GL_SHORT: + return sizeof(GLshort); + case GL_UNSIGNED_SHORT: + return sizeof(GLushort); + case GL_BYTE: + return sizeof(GLbyte); + case GL_UNSIGNED_BYTE: + return sizeof(GLubyte); + default: + assert(0); + return 0; + } +} + +#define CONVERT( TYPE, MACRO ) do { \ + GLuint i, j, sz; \ + sz = input->Size; \ + if (input->Normalized) { \ + for (i = 0; i < count; i++) { \ + const TYPE *in = (TYPE *)src_ptr; \ + for (j = 0; j < sz; j++) { \ + *dst_ptr++ = MACRO(*in); \ + in++; \ + } \ + src_ptr += stride; \ + } \ + } else { \ + for (i = 0; i < count; i++) { \ + const TYPE *in = (TYPE *)src_ptr; \ + for (j = 0; j < sz; j++) { \ + *dst_ptr++ = (GLfloat)(*in); \ + in++; \ + } \ + src_ptr += stride; \ + } \ + } \ +} while (0) + +static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const struct gl_client_array *input, struct gl_buffer_object **bo, GLuint *nr_bo) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + struct vertex_attribute r300_attr; + const void *src_ptr; + GLenum type; + GLuint stride; + + if (input->BufferObj->Name) { + if (!input->BufferObj->Pointer) { + bo[*nr_bo] = input->BufferObj; + (*nr_bo)++; + ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY_ARB, input->BufferObj); + assert(input->BufferObj->Pointer != NULL); + } + + src_ptr = ADD_POINTERS(input->BufferObj->Pointer, input->Ptr); + } else + src_ptr = input->Ptr; + + stride = (input->StrideB == 0) ? getTypeSize(input->Type) * input->Size : input->StrideB; + + if (input->Type == GL_DOUBLE || input->Type == GL_UNSIGNED_INT || input->Type == GL_INT || +#if MESA_BIG_ENDIAN + getTypeSize(input->Type) != 4 || +#endif + stride < 4) { + if (RADEON_DEBUG & DEBUG_FALLBACKS) { + fprintf(stderr, "%s: Converting vertex attributes, attribute data format %x,", __FUNCTION__, input->Type); + fprintf(stderr, "stride %d, components %d\n", stride, input->Size); + } + + GLfloat *dst_ptr, *tmp; + tmp = dst_ptr = _mesa_malloc(sizeof(GLfloat) * input->Size * count); + + switch (input->Type) { + case GL_DOUBLE: + CONVERT(GLdouble, (GLfloat)); + break; + case GL_UNSIGNED_INT: + CONVERT(GLuint, UINT_TO_FLOAT); + break; + case GL_INT: + CONVERT(GLint, INT_TO_FLOAT); + break; + case GL_UNSIGNED_SHORT: + CONVERT(GLushort, USHORT_TO_FLOAT); + break; + case GL_SHORT: + CONVERT(GLshort, SHORT_TO_FLOAT); + break; + case GL_UNSIGNED_BYTE: + assert(input->Format != GL_BGRA); + CONVERT(GLubyte, UBYTE_TO_FLOAT); + break; + case GL_BYTE: + CONVERT(GLbyte, BYTE_TO_FLOAT); + break; + default: + assert(0); + break; + } + + type = GL_FLOAT; + r300_attr.free_needed = GL_TRUE; + r300_attr.data = tmp; + r300_attr.stride = sizeof(GLfloat) * input->Size; + r300_attr.dwords = input->Size; + } else { + type = input->Type; + r300_attr.free_needed = GL_FALSE; + r300_attr.data = (GLvoid *)src_ptr; + r300_attr.stride = stride; + r300_attr.dwords = (getTypeSize(type) * input->Size + 3)/ 4; + } + + r300_attr.size = input->Size; + r300_attr.element = attr; + r300_attr.dst_loc = vbuf->num_attribs; + + switch (type) { + case GL_FLOAT: + switch (input->Size) { + case 1: r300_attr.data_type = R300_DATA_TYPE_FLOAT_1; break; + case 2: r300_attr.data_type = R300_DATA_TYPE_FLOAT_2; break; + case 3: r300_attr.data_type = R300_DATA_TYPE_FLOAT_3; break; + case 4: r300_attr.data_type = R300_DATA_TYPE_FLOAT_4; break; + } + r300_attr._signed = 0; + r300_attr.normalize = 0; + break; + case GL_SHORT: + r300_attr._signed = 1; + r300_attr.normalize = input->Normalized; + switch (input->Size) { + case 1: + case 2: + r300_attr.data_type = R300_DATA_TYPE_SHORT_2; + break; + case 3: + case 4: + r300_attr.data_type = R300_DATA_TYPE_SHORT_4; + break; + } + break; + case GL_BYTE: + r300_attr._signed = 1; + r300_attr.normalize = input->Normalized; + r300_attr.data_type = R300_DATA_TYPE_BYTE; + break; + case GL_UNSIGNED_SHORT: + r300_attr._signed = 0; + r300_attr.normalize = input->Normalized; + switch (input->Size) { + case 1: + case 2: + r300_attr.data_type = R300_DATA_TYPE_SHORT_2; + break; + case 3: + case 4: + r300_attr.data_type = R300_DATA_TYPE_SHORT_4; + break; + } + break; + case GL_UNSIGNED_BYTE: + r300_attr._signed = 0; + r300_attr.normalize = input->Normalized; + if (input->Format == GL_BGRA) + r300_attr.data_type = R300_DATA_TYPE_D3DCOLOR; + else + r300_attr.data_type = R300_DATA_TYPE_BYTE; + break; + + default: + case GL_DOUBLE: + case GL_INT: + case GL_UNSIGNED_INT: + assert(0); + break; + } + + switch (input->Size) { + case 4: + r300_attr.swizzle = SWIZZLE_XYZW; + break; + case 3: + r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); + break; + case 2: + r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE); + break; + case 1: + r300_attr.swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE); + break; + } + + r300_attr.write_mask = MASK_XYZW; + + vbuf->attribs[vbuf->num_attribs] = r300_attr; + ++vbuf->num_attribs; +} + +static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *arrays[], int count, struct gl_buffer_object **bo, GLuint *nr_bo) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + + { + int i, tmp; + + tmp = r300->selected_vp->key.InputsRead; + i = 0; + vbuf->num_attribs = 0; + while (tmp) { + /* find first enabled bit */ + while (!(tmp & 1)) { + tmp >>= 1; + ++i; + } + + r300TranslateAttrib(ctx, i, count, arrays[i], bo, nr_bo); + + tmp >>= 1; + ++i; + } + } + + r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS); + if (r300->fallback) + return; + + { + int i; + + for (i = 0; i < vbuf->num_attribs; i++) { + rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i], + vbuf->attribs[i].data, vbuf->attribs[i].dwords, + vbuf->attribs[i].stride, count); + } + + r300->radeon.tcl.aos_count = vbuf->num_attribs; + } +} + +static void r300FreeData(GLcontext *ctx, struct gl_buffer_object **bo, GLuint nr_bo) +{ + { + struct r300_vertex_buffer *vbuf = &R300_CONTEXT(ctx)->vbuf; + int i; + + for (i = 0; i < vbuf->num_attribs; i++) { + if (vbuf->attribs[i].free_needed) + _mesa_free(vbuf->attribs[i].data); + } + } + + { + struct r300_index_buffer *ind_buf = &R300_CONTEXT(ctx)->ind_buf; + if (ind_buf->free_needed) + _mesa_free(ind_buf->ptr); + } + + { + int i; + + for (i = 0; i < nr_bo; ++i) { + ctx->Driver.UnmapBuffer(ctx, 0, bo[i]); + } + } +} + +static GLboolean r300TryDrawPrims(GLcontext *ctx, + const struct gl_client_array *arrays[], + const struct _mesa_prim *prim, + GLuint nr_prims, + const struct _mesa_index_buffer *ib, + GLuint min_index, + GLuint max_index ) +{ + struct r300_context *r300 = R300_CONTEXT(ctx); + struct gl_buffer_object *bo[VERT_ATTRIB_MAX+1]; + GLuint i, nr_bo = 0; + + if (ctx->NewState) + _mesa_update_state( ctx ); + + if (r300->options.hw_tcl_enabled) + _tnl_UpdateFixedFunctionProgram(ctx); + + r300UpdateShaders(r300); + + r300SwitchFallback(ctx, R300_FALLBACK_INVALID_BUFFERS, !r300ValidateBuffers(ctx)); + + r300FixupIndexBuffer(ctx, ib, bo, &nr_bo); + + r300SetVertexFormat(ctx, arrays, max_index + 1, bo, &nr_bo); + + if (r300->fallback) + return GL_FALSE; + + r300SetupVAP(ctx, r300->selected_vp->key.InputsRead, r300->selected_vp->key.OutputsWritten); + + r300UpdateShaderStates(r300); + + r300EmitCacheFlush(r300); + radeonEmitState(&r300->radeon); + + for (i = 0; i < nr_prims; ++i) { + r300RunRenderPrimitive(ctx, prim[i].start, prim[i].start + prim[i].count, prim[i].mode); + } + + r300EmitCacheFlush(r300); + + radeonReleaseArrays(ctx, ~0); + + r300FreeData(ctx, bo, nr_bo); + + return GL_TRUE; +} + +/* TODO: rebase if number of indices in any of primitives is > 8192 for 32bit indices or 16384 for 16bit indices */ + +static void r300DrawPrims(GLcontext *ctx, + const struct gl_client_array *arrays[], + const struct _mesa_prim *prim, + GLuint nr_prims, + const struct _mesa_index_buffer *ib, + GLuint min_index, + GLuint max_index) +{ + GLboolean retval; + + if (min_index) { + vbo_rebase_prims( ctx, arrays, prim, nr_prims, ib, min_index, max_index, r300DrawPrims ); + return; + } + + /* Make an attempt at drawing */ + retval = r300TryDrawPrims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); + + /* If failed run tnl pipeline - it should take care of fallbacks */ + if (!retval) + _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); +} + +void r300InitDraw(GLcontext *ctx) +{ + struct vbo_context *vbo = vbo_context(ctx); + + vbo->draw_prims = r300DrawPrims; +} diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c index 28c3157427..c3817721dc 100644 --- a/src/mesa/drivers/dri/r300/r300_emit.c +++ b/src/mesa/drivers/dri/r300/r300_emit.c @@ -31,6 +31,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * \file * * \author Keith Whitwell <keith@tungstengraphics.com> + * \author Maciej Cencora <m.cencora@gmail.com> */ #include "main/glheader.h" @@ -46,222 +47,11 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "tnl/t_context.h" #include "r300_context.h" -#include "radeon_ioctl.h" #include "r300_state.h" #include "r300_emit.h" #include "r300_ioctl.h" - -#ifdef USER_BUFFERS -#include "r300_mem.h" -#endif - -#if SWIZZLE_X != R300_INPUT_ROUTE_SELECT_X || \ - SWIZZLE_Y != R300_INPUT_ROUTE_SELECT_Y || \ - SWIZZLE_Z != R300_INPUT_ROUTE_SELECT_Z || \ - SWIZZLE_W != R300_INPUT_ROUTE_SELECT_W || \ - SWIZZLE_ZERO != R300_INPUT_ROUTE_SELECT_ZERO || \ - SWIZZLE_ONE != R300_INPUT_ROUTE_SELECT_ONE -#error Cannot change these! -#endif - -#define DEBUG_ALL DEBUG_VERTS - -#if defined(USE_X86_ASM) -#define COPY_DWORDS( dst, src, nr ) \ -do { \ - int __tmp; \ - __asm__ __volatile__( "rep ; movsl" \ - : "=%c" (__tmp), "=D" (dst), "=S" (__tmp) \ - : "0" (nr), \ - "D" ((long)dst), \ - "S" ((long)src) ); \ -} while (0) -#else -#define COPY_DWORDS( dst, src, nr ) \ -do { \ - int j; \ - for ( j = 0 ; j < nr ; j++ ) \ - dst[j] = ((int *)src)[j]; \ - dst += nr; \ -} while (0) -#endif - -static void r300EmitVec4(GLcontext * ctx, struct r300_dma_region *rvb, - GLvoid * data, int stride, int count) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p data %p\n", - __FUNCTION__, count, stride, (void *)out, (void *)data); - - if (stride == 4) - COPY_DWORDS(out, data, count); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out++; - data += stride; - } -} - -static void r300EmitVec8(GLcontext * ctx, struct r300_dma_region *rvb, - GLvoid * data, int stride, int count) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p data %p\n", - __FUNCTION__, count, stride, (void *)out, (void *)data); - - if (stride == 8) - COPY_DWORDS(out, data, count * 2); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data + 4); - out += 2; - data += stride; - } -} - -static void r300EmitVec12(GLcontext * ctx, struct r300_dma_region *rvb, - GLvoid * data, int stride, int count) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p data %p\n", - __FUNCTION__, count, stride, (void *)out, (void *)data); - - if (stride == 12) - COPY_DWORDS(out, data, count * 3); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data + 4); - out[2] = *(int *)(data + 8); - out += 3; - data += stride; - } -} - -static void r300EmitVec16(GLcontext * ctx, struct r300_dma_region *rvb, - GLvoid * data, int stride, int count) -{ - int i; - int *out = (int *)(rvb->address + rvb->start); - - if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s count %d stride %d out %p data %p\n", - __FUNCTION__, count, stride, (void *)out, (void *)data); - - if (stride == 16) - COPY_DWORDS(out, data, count * 4); - else - for (i = 0; i < count; i++) { - out[0] = *(int *)data; - out[1] = *(int *)(data + 4); - out[2] = *(int *)(data + 8); - out[3] = *(int *)(data + 12); - out += 4; - data += stride; - } -} - -static void r300EmitVec(GLcontext * ctx, struct r300_dma_region *rvb, - GLvoid * data, int size, int stride, int count) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - - if (stride == 0) { - r300AllocDmaRegion(rmesa, rvb, size * 4, 4); - count = 1; - rvb->aos_offset = GET_START(rvb); - rvb->aos_stride = 0; - } else { - r300AllocDmaRegion(rmesa, rvb, size * count * 4, 4); - rvb->aos_offset = GET_START(rvb); - rvb->aos_stride = size; - } - - switch (size) { - case 1: - r300EmitVec4(ctx, rvb, data, stride, count); - break; - case 2: - r300EmitVec8(ctx, rvb, data, stride, count); - break; - case 3: - r300EmitVec12(ctx, rvb, data, stride, count); - break; - case 4: - r300EmitVec16(ctx, rvb, data, stride, count); - break; - default: - assert(0); - break; - } -} - -#define DW_SIZE(x) ((inputs[tab[(x)]] << R300_DST_VEC_LOC_SHIFT) | \ - (attribptr[tab[(x)]]->size - 1) << R300_DATA_TYPE_0_SHIFT) - -GLuint r300VAPInputRoute0(uint32_t * dst, GLvector4f ** attribptr, - int *inputs, GLint * tab, GLuint nr) -{ - GLuint i, dw; - - /* type, inputs, stop bit, size */ - for (i = 0; i < nr; i += 2) { - /* make sure input is valid, would lockup the gpu */ - assert(inputs[tab[i]] != -1); - dw = (R300_SIGNED | DW_SIZE(i)); - if (i + 1 == nr) { - dw |= R300_LAST_VEC << R300_DATA_TYPE_0_SHIFT; - } else { - assert(inputs[tab[i + 1]] != -1); - dw |= (R300_SIGNED | - DW_SIZE(i + 1)) << R300_DATA_TYPE_1_SHIFT; - if (i + 2 == nr) { - dw |= R300_LAST_VEC << R300_DATA_TYPE_1_SHIFT; - } - } - dst[i >> 1] = dw; - } - - return (nr + 1) >> 1; -} - -static GLuint r300VAPInputRoute1Swizzle(int swizzle[4]) -{ - return (swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) | - (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) | - (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) | - (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT); -} - -GLuint r300VAPInputRoute1(uint32_t * dst, int swizzle[][4], GLuint nr) -{ - GLuint i, dw; - - for (i = 0; i < nr; i += 2) { - dw = (r300VAPInputRoute1Swizzle(swizzle[i]) | - ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | - R300_WRITE_ENA_Z | R300_WRITE_ENA_W) << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE0_SHIFT; - if (i + 1 < nr) { - dw |= (r300VAPInputRoute1Swizzle(swizzle[i + 1]) | - ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | - R300_WRITE_ENA_Z | R300_WRITE_ENA_W) << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE1_SHIFT; - } - dst[i >> 1] = dw; - } - - return (nr + 1) >> 1; -} +#include "r300_render.h" +#include "r300_swtcl.h" GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead) { @@ -272,7 +62,6 @@ GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead) GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead) { - r300ContextPtr rmesa = R300_CONTEXT(ctx); GLuint i, vic_1 = 0; if (InputsRead & (1 << VERT_ATTRIB_POS)) @@ -284,281 +73,112 @@ GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead) if (InputsRead & (1 << VERT_ATTRIB_COLOR0)) vic_1 |= R300_INPUT_CNTL_COLOR; - rmesa->state.texture.tc_count = 0; for (i = 0; i < ctx->Const.MaxTextureUnits; i++) if (InputsRead & (1 << (VERT_ATTRIB_TEX0 + i))) { - rmesa->state.texture.tc_count++; vic_1 |= R300_INPUT_CNTL_TC0 << i; } return vic_1; } -GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten) +GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads) { GLuint ret = 0; - if (OutputsWritten & (1 << VERT_RESULT_HPOS)) + if (vp_writes & (1 << VERT_RESULT_HPOS)) ret |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; - if (OutputsWritten & (1 << VERT_RESULT_COL0)) + if (vp_writes & (1 << VERT_RESULT_COL0) && fp_reads & FRAG_BIT_COL0) ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT; - if (OutputsWritten & (1 << VERT_RESULT_COL1)) + if (vp_writes & (1 << VERT_RESULT_COL1) && fp_reads & FRAG_BIT_COL1) ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT; - if (OutputsWritten & (1 << VERT_RESULT_BFC0) - || OutputsWritten & (1 << VERT_RESULT_BFC1)) - ret |= - R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT | - R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT | - R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT; + /* Two sided lighting works only if all 4 colors are written */ + if (vp_writes & (1 << VERT_RESULT_BFC0) || vp_writes & (1 << VERT_RESULT_BFC1)) + ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT | + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT; - if (OutputsWritten & (1 << VERT_RESULT_PSIZ)) + if (vp_writes & (1 << VERT_RESULT_PSIZ)) ret |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; return ret; } -GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten) +GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads) { GLuint i, ret = 0, first_free_texcoord = 0; for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { - if (OutputsWritten & (1 << (VERT_RESULT_TEX0 + i))) { - ret |= (4 << (3 * i)); + if (vp_writes & (1 << (VERT_RESULT_TEX0 + i)) && fp_reads & FRAG_BIT_TEX(i)) { + ret |= (4 << (3 * first_free_texcoord)); ++first_free_texcoord; } } - if (OutputsWritten & (1 << VERT_RESULT_FOGC)) { - if (first_free_texcoord > 8) { - fprintf(stderr, "\tout of free texcoords to write fog coord\n"); - _mesa_exit(-1); - } - ret |= 4 << (3 * first_free_texcoord); - } - - return ret; -} - -/* Emit vertex data to GART memory - * Route inputs to the vertex processor - * This function should never return R300_FALLBACK_TCL when using software tcl. - */ -int r300EmitArrays(GLcontext * ctx) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; - GLuint nr; - GLuint count = vb->Count; - GLuint i; - GLuint InputsRead = 0, OutputsWritten = 0; - int *inputs = NULL; - int vir_inputs[VERT_ATTRIB_MAX]; - GLint tab[VERT_ATTRIB_MAX]; - int swizzle[VERT_ATTRIB_MAX][4]; - struct r300_vertex_program *prog = - (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx); - - if (hw_tcl_on) { - inputs = prog->inputs; - InputsRead = prog->key.InputsRead; - OutputsWritten = prog->key.OutputsWritten; - } else { - inputs = rmesa->state.sw_tcl_inputs; - - DECLARE_RENDERINPUTS(render_inputs_bitset); - RENDERINPUTS_COPY(render_inputs_bitset, tnl->render_inputs_bitset); - - vb->AttribPtr[VERT_ATTRIB_POS] = vb->ClipPtr; - - assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_POS)); - assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_NORMAL) == 0); - //assert(RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR0)); - - if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_POS)) { - InputsRead |= 1 << VERT_ATTRIB_POS; - OutputsWritten |= 1 << VERT_RESULT_HPOS; - } - - if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR0)) { - InputsRead |= 1 << VERT_ATTRIB_COLOR0; - OutputsWritten |= 1 << VERT_RESULT_COL0; - } - - if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_COLOR1)) { - InputsRead |= 1 << VERT_ATTRIB_COLOR1; - OutputsWritten |= 1 << VERT_RESULT_COL1; - } - - for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { - if (RENDERINPUTS_TEST(render_inputs_bitset, _TNL_ATTRIB_TEX(i))) { - InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i); - OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i); - } - } - - for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) { - if (InputsRead & (1 << i)) { - inputs[i] = nr++; - } else { - inputs[i] = -1; - } - } - - /* Fixed, apply to vir0 only */ - memcpy(vir_inputs, inputs, VERT_ATTRIB_MAX * sizeof(int)); - inputs = vir_inputs; - if (InputsRead & VERT_ATTRIB_POS) - inputs[VERT_ATTRIB_POS] = 0; - if (InputsRead & (1 << VERT_ATTRIB_COLOR0)) - inputs[VERT_ATTRIB_COLOR0] = 2; - if (InputsRead & (1 << VERT_ATTRIB_COLOR1)) - inputs[VERT_ATTRIB_COLOR1] = 3; - for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++) - if (InputsRead & (1 << i)) - inputs[i] = 6 + (i - VERT_ATTRIB_TEX0); - - RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset, render_inputs_bitset); + if (fp_reads & FRAG_BIT_WPOS) { + ret |= (4 << (3 * first_free_texcoord)); + ++first_free_texcoord; } - assert(InputsRead); - assert(OutputsWritten); - - for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) { - if (InputsRead & (1 << i)) { - tab[nr++] = i; - } + if (vp_writes & (1 << VERT_RESULT_FOGC) && fp_reads & FRAG_BIT_FOGC) { + ret |= 4 << (3 * first_free_texcoord); } - if (nr > R300_MAX_AOS_ARRAYS) { - return R300_FALLBACK_TCL; + if (first_free_texcoord > 8) { + fprintf(stderr, "\tout of free texcoords\n"); + _mesa_exit(-1); } - for (i = 0; i < nr; i++) { - int ci, fix, found = 0; - - swizzle[i][0] = SWIZZLE_ZERO; - swizzle[i][1] = SWIZZLE_ZERO; - swizzle[i][2] = SWIZZLE_ZERO; - swizzle[i][3] = SWIZZLE_ONE; + return ret; +} - for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++) { - swizzle[i][ci] = ci; - } +GLboolean r300EmitArrays(GLcontext * ctx) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_vertex_buffer *vbuf = &r300->vbuf; + GLuint InputsRead, OutputsWritten; - if (r300IsGartMemory(rmesa, vb->AttribPtr[tab[i]]->data, 4)) { - if (vb->AttribPtr[tab[i]]->stride % 4) { - return R300_FALLBACK_TCL; - } - rmesa->state.aos[i].address = (void *)(vb->AttribPtr[tab[i]]->data); - rmesa->state.aos[i].start = 0; - rmesa->state.aos[i].aos_offset = r300GartOffsetFromVirtual(rmesa, vb->AttribPtr[tab[i]]->data); - rmesa->state.aos[i].aos_stride = vb->AttribPtr[tab[i]]->stride / 4; - rmesa->state.aos[i].aos_size = vb->AttribPtr[tab[i]]->size; - } else { - r300EmitVec(ctx, &rmesa->state.aos[i], - vb->AttribPtr[tab[i]]->data, - vb->AttribPtr[tab[i]]->size, - vb->AttribPtr[tab[i]]->stride, count); - } + r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten); - rmesa->state.aos[i].aos_size = vb->AttribPtr[tab[i]]->size; + r300SwitchFallback(ctx, R300_FALLBACK_AOS_LIMIT, vbuf->num_attribs > R300_MAX_AOS_ARRAYS); + if (r300->fallback & R300_RASTER_FALLBACK_MASK) + return GL_FALSE; - for (fix = 0; fix <= 4 - vb->AttribPtr[tab[i]]->size; fix++) { - if ((rmesa->state.aos[i].aos_offset - _mesa_sizeof_type(GL_FLOAT) * fix) % 4) { - continue; - } - found = 1; - break; - } + { + struct vertex_buffer *mesa_vb = &TNL_CONTEXT(ctx)->vb; + GLuint attr, i; - if (found) { - if (fix > 0) { - WARN_ONCE("Feeling lucky?\n"); - } - rmesa->state.aos[i].aos_offset -= _mesa_sizeof_type(GL_FLOAT) * fix; - for (ci = 0; ci < vb->AttribPtr[tab[i]]->size; ci++) { - swizzle[i][ci] += fix; - } - } else { - WARN_ONCE - ("Cannot handle offset %x with stride %d, comp %d\n", - rmesa->state.aos[i].aos_offset, - rmesa->state.aos[i].aos_stride, - vb->AttribPtr[tab[i]]->size); - return R300_FALLBACK_TCL; + for (i = 0; i < vbuf->num_attribs; i++) { + attr = vbuf->attribs[i].element; + rcommon_emit_vector(ctx, &r300->radeon.tcl.aos[i], mesa_vb->AttribPtr[attr]->data, + mesa_vb->AttribPtr[attr]->size, mesa_vb->AttribPtr[attr]->stride, mesa_vb->Count); } - } - /* Setup INPUT_ROUTE. */ - R300_STATECHANGE(rmesa, vir[0]); - ((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count = - r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0], - vb->AttribPtr, inputs, tab, nr); - R300_STATECHANGE(rmesa, vir[1]); - ((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count = - r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle, - nr); - - /* Setup INPUT_CNTL. */ - R300_STATECHANGE(rmesa, vic); - rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead); - rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead); - - /* Setup OUTPUT_VTX_FMT. */ - R300_STATECHANGE(rmesa, vof); - rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = - r300VAPOutputCntl0(ctx, OutputsWritten); - rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = - r300VAPOutputCntl1(ctx, OutputsWritten); - - rmesa->state.aos_count = nr; - - return R300_FALLBACK_NONE; -} + r300->radeon.tcl.aos_count = vbuf->num_attribs; -#ifdef USER_BUFFERS -void r300UseArrays(GLcontext * ctx) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - int i; - - if (rmesa->state.elt_dma.buf) - r300_mem_use(rmesa, rmesa->state.elt_dma.buf->id); - - for (i = 0; i < rmesa->state.aos_count; i++) { - if (rmesa->state.aos[i].buf) - r300_mem_use(rmesa, rmesa->state.aos[i].buf->id); + /* Fill index buffer info */ + r300->ind_buf.ptr = mesa_vb->Elts; + r300->ind_buf.is_32bit = GL_TRUE; + r300->ind_buf.free_needed = GL_FALSE; } -} -#endif -void r300ReleaseArrays(GLcontext * ctx) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - int i; + r300SetupVAP(ctx, InputsRead, OutputsWritten); - r300ReleaseDmaRegion(rmesa, &rmesa->state.elt_dma, __FUNCTION__); - for (i = 0; i < rmesa->state.aos_count; i++) { - r300ReleaseDmaRegion(rmesa, &rmesa->state.aos[i], __FUNCTION__); - } + return GL_TRUE; } void r300EmitCacheFlush(r300ContextPtr rmesa) { - int cmd_reserved = 0; - int cmd_written = 0; - - drm_radeon_cmd_header_t *cmd = NULL; - - reg_start(R300_RB3D_DSTCACHE_CTLSTAT, 0); - e32(R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | - R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); - - reg_start(R300_ZB_ZCACHE_CTLSTAT, 0); - e32(R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | - R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); + BATCH_LOCALS(&rmesa->radeon); + + BEGIN_BATCH_NO_AUTOSTATE(4); + OUT_BATCH_REGVAL(R300_RB3D_DSTCACHE_CTLSTAT, + R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | + R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); + OUT_BATCH_REGVAL(R300_ZB_ZCACHE_CTLSTAT, + R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | + R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); + END_BATCH(); + COMMIT_BATCH(); } diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h index 89d738339f..2fb8b82d3a 100644 --- a/src/mesa/drivers/dri/r300/r300_emit.h +++ b/src/mesa/drivers/dri/r300/r300_emit.h @@ -44,28 +44,31 @@ #include "r300_cmdbuf.h" #include "radeon_reg.h" -/* TODO: move these defines (and the ones from DRM) into r300_reg.h and sync up - * with DRM */ -#define CP_PACKET0(reg, n) (RADEON_CP_PACKET0 | ((n)<<16) | ((reg)>>2)) -#define CP_PACKET3( pkt, n ) \ - (RADEON_CP_PACKET3 | (pkt) | ((n) << 16)) - -static INLINE uint32_t cmdpacket0(int reg, int count) +static INLINE uint32_t cmdpacket0(struct radeon_screen *rscrn, + int reg, int count) { - drm_r300_cmd_header_t cmd; - - cmd.packet0.cmd_type = R300_CMD_PACKET0; - cmd.packet0.count = count; - cmd.packet0.reghi = ((unsigned int)reg & 0xFF00) >> 8; - cmd.packet0.reglo = ((unsigned int)reg & 0x00FF); - - return cmd.u; + if (!rscrn->kernel_mm) { + drm_r300_cmd_header_t cmd; + + cmd.u = 0; + cmd.packet0.cmd_type = R300_CMD_PACKET0; + cmd.packet0.count = count; + cmd.packet0.reghi = ((unsigned int)reg & 0xFF00) >> 8; + cmd.packet0.reglo = ((unsigned int)reg & 0x00FF); + + return cmd.u; + } + if (count) { + return CP_PACKET0(reg, count - 1); + } + return CP_PACKET2; } -static INLINE uint32_t cmdvpu(int addr, int count) +static INLINE uint32_t cmdvpu(struct radeon_screen *rscrn, int addr, int count) { drm_r300_cmd_header_t cmd; + cmd.u = 0; cmd.vpu.cmd_type = R300_CMD_VPU; cmd.vpu.count = count; cmd.vpu.adrhi = ((unsigned int)addr & 0xFF00) >> 8; @@ -74,10 +77,12 @@ static INLINE uint32_t cmdvpu(int addr, int count) return cmd.u; } -static INLINE uint32_t cmdr500fp(int addr, int count, int type, int clamp) +static INLINE uint32_t cmdr500fp(struct radeon_screen *rscrn, + int addr, int count, int type, int clamp) { drm_r300_cmd_header_t cmd; + cmd.u = 0; cmd.r500fp.cmd_type = R300_CMD_R500FP; cmd.r500fp.count = count; cmd.r500fp.adrhi_flags = ((unsigned int)addr & 0x100) >> 8; @@ -88,181 +93,139 @@ static INLINE uint32_t cmdr500fp(int addr, int count, int type, int clamp) return cmd.u; } -static INLINE uint32_t cmdpacket3(int packet) +static INLINE uint32_t cmdpacket3(struct radeon_screen *rscrn, int packet) { drm_r300_cmd_header_t cmd; + cmd.u = 0; cmd.packet3.cmd_type = R300_CMD_PACKET3; cmd.packet3.packet = packet; return cmd.u; } -static INLINE uint32_t cmdcpdelay(unsigned short count) +static INLINE uint32_t cmdcpdelay(struct radeon_screen *rscrn, + unsigned short count) { drm_r300_cmd_header_t cmd; + cmd.u = 0; + cmd.delay.cmd_type = R300_CMD_CP_DELAY; cmd.delay.count = count; return cmd.u; } -static INLINE uint32_t cmdwait(unsigned char flags) +static INLINE uint32_t cmdwait(struct radeon_screen *rscrn, + unsigned char flags) { drm_r300_cmd_header_t cmd; + cmd.u = 0; cmd.wait.cmd_type = R300_CMD_WAIT; cmd.wait.flags = flags; return cmd.u; } -static INLINE uint32_t cmdpacify(void) +static INLINE uint32_t cmdpacify(struct radeon_screen *rscrn) { drm_r300_cmd_header_t cmd; + cmd.u = 0; cmd.header.cmd_type = R300_CMD_END3D; return cmd.u; } /** - * Prepare to write a register value to register at address reg. - * If num_extra > 0 then the following extra values are written - * to registers with address +4, +8 and so on.. - */ -#define reg_start(reg, num_extra) \ - do { \ - int _n; \ - _n=(num_extra); \ - cmd = (drm_radeon_cmd_header_t*) \ - r300AllocCmdBuf(rmesa, \ - (_n+2), \ - __FUNCTION__); \ - cmd_reserved=_n+2; \ - cmd_written=1; \ - cmd[0].i=cmdpacket0((reg), _n+1); \ - } while (0); - -/** - * Emit GLuint freestyle + * Write the header of a packet3 to the command buffer. + * Outputs 2 dwords and expects (num_extra+1) additional dwords afterwards. */ -#define e32(dword) \ - do { \ - if(cmd_written<cmd_reserved) { \ - cmd[cmd_written].i=(dword); \ - cmd_written++; \ - } else { \ - fprintf(stderr, \ - "e32 but no previous packet " \ - "declaration.\n" \ - "Aborting! in %s::%s at line %d, " \ - "cmd_written=%d cmd_reserved=%d\n", \ - __FILE__, __FUNCTION__, __LINE__, \ - cmd_written, cmd_reserved); \ - _mesa_exit(-1); \ - } \ +#define OUT_BATCH_PACKET3(packet, num_extra) do {\ + if (!b_l_rmesa->radeonScreen->kernel_mm) { \ + OUT_BATCH(cmdpacket3(b_l_rmesa->radeonScreen,\ + R300_CMD_PACKET3_RAW)); \ + } else b_l_rmesa->cmdbuf.cs->section_cdw++;\ + OUT_BATCH(CP_PACKET3((packet), (num_extra))); \ } while(0) -#define efloat(f) e32(r300PackFloat32(f)) - -#define vsf_start_fragment(dest, length) \ - do { \ - int _n; \ - _n = (length); \ - cmd = (drm_radeon_cmd_header_t*) \ - r300AllocCmdBuf(rmesa, \ - (_n+1), \ - __FUNCTION__); \ - cmd_reserved = _n+2; \ - cmd_written =1; \ - cmd[0].i = cmdvpu((dest), _n/4); \ - } while (0); - -#define r500fp_start_fragment(dest, length) \ - do { \ - int _n; \ - _n = (length); \ - cmd = (drm_radeon_cmd_header_t*) \ - r300AllocCmdBuf(rmesa, \ - (_n+1), \ - __FUNCTION__); \ - cmd_reserved = _n+1; \ - cmd_written =1; \ - cmd[0].i = cmdr500fp((dest), _n/6, 0, 0); \ - } while (0); - -#define start_packet3(packet, count) \ - { \ - int _n; \ - GLuint _p; \ - _n = (count); \ - _p = (packet); \ - cmd = (drm_radeon_cmd_header_t*) \ - r300AllocCmdBuf(rmesa, \ - (_n+3), \ - __FUNCTION__); \ - cmd_reserved = _n+3; \ - cmd_written = 2; \ - if(_n > 0x3fff) { \ - fprintf(stderr,"Too big packet3 %08x: cannot " \ - "store %d dwords\n", \ - _p, _n); \ - _mesa_exit(-1); \ - } \ - cmd[0].i = cmdpacket3(R300_CMD_PACKET3_RAW); \ - cmd[1].i = _p | ((_n & 0x3fff)<<16); \ - } - /** * Must be sent to switch to 2d commands */ -void static INLINE end_3d(r300ContextPtr rmesa) +void static INLINE end_3d(radeonContextPtr radeon) { - drm_radeon_cmd_header_t *cmd = NULL; + BATCH_LOCALS(radeon); - cmd = - (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__); - cmd[0].header.cmd_type = R300_CMD_END3D; + if (!radeon->radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(1); + OUT_BATCH(cmdpacify(radeon->radeonScreen)); + END_BATCH(); + } } void static INLINE cp_delay(r300ContextPtr rmesa, unsigned short count) { - drm_radeon_cmd_header_t *cmd = NULL; + BATCH_LOCALS(&rmesa->radeon); - cmd = - (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__); - cmd[0].i = cmdcpdelay(count); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(1); + OUT_BATCH(cmdcpdelay(rmesa->radeon.radeonScreen, count)); + END_BATCH(); + } } -void static INLINE cp_wait(r300ContextPtr rmesa, unsigned char flags) +void static INLINE cp_wait(radeonContextPtr radeon, unsigned char flags) { - drm_radeon_cmd_header_t *cmd = NULL; - - cmd = - (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 1, __FUNCTION__); - cmd[0].i = cmdwait(flags); + BATCH_LOCALS(radeon); + uint32_t wait_until; + + if (!radeon->radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(1); + OUT_BATCH(cmdwait(radeon->radeonScreen, flags)); + END_BATCH(); + } else { + switch(flags) { + case R300_WAIT_2D: + wait_until = (1 << 14); + break; + case R300_WAIT_3D: + wait_until = (1 << 15); + break; + case R300_NEW_WAIT_2D_3D: + wait_until = (1 << 14) | (1 << 15); + break; + case R300_NEW_WAIT_2D_2D_CLEAN: + wait_until = (1 << 14) | (1 << 16) | (1 << 18); + break; + case R300_NEW_WAIT_3D_3D_CLEAN: + wait_until = (1 << 15) | (1 << 17) | (1 << 18); + break; + case R300_NEW_WAIT_2D_2D_CLEAN_3D_3D_CLEAN: + wait_until = (1 << 14) | (1 << 16) | (1 << 18); + wait_until |= (1 << 15) | (1 << 17) | (1 << 18); + break; + default: + return; + } + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH(CP_PACKET0(RADEON_WAIT_UNTIL, 0)); + OUT_BATCH(wait_until); + END_BATCH(); + } } -extern int r300EmitArrays(GLcontext * ctx); - -#ifdef USER_BUFFERS -void r300UseArrays(GLcontext * ctx); -#endif +extern GLboolean r300EmitArrays(GLcontext * ctx); -extern void r300ReleaseArrays(GLcontext * ctx); extern int r300PrimitiveType(r300ContextPtr rmesa, int prim); extern int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim); extern void r300EmitCacheFlush(r300ContextPtr rmesa); -extern GLuint r300VAPInputRoute0(uint32_t * dst, GLvector4f ** attribptr, - int *inputs, GLint * tab, GLuint nr); -extern GLuint r300VAPInputRoute1(uint32_t * dst, int swizzle[][4], GLuint nr); extern GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead); extern GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead); -extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten); -extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten); +extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads); +extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads); #endif diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c index 2f45429cf2..55c1cfe631 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog.c @@ -25,32 +25,12 @@ * */ -/** - * \file - * - * Fragment program compiler. Perform transformations on the intermediate - * representation until the program is in a form where we can translate - * it more or less directly into machine-readable form. - * - * \author Ben Skeggs <darktama@iinet.net.au> - * \author Jerome Glisse <j.glisse@gmail.com> - */ +#include "r300_fragprog.h" -#include "main/glheader.h" -#include "main/macros.h" -#include "main/enums.h" -#include "shader/prog_instruction.h" #include "shader/prog_parameter.h" -#include "shader/prog_print.h" #include "r300_context.h" -#include "r300_fragprog.h" #include "r300_fragprog_swizzle.h" -#include "r300_state.h" - -#include "radeon_nqssadce.h" -#include "radeon_program_alu.h" - static void reset_srcreg(struct prog_src_register* reg) { @@ -81,7 +61,7 @@ static struct prog_src_register shadow_ambient(struct gl_program *program, int t * \todo If/when r5xx uses the radeon_program architecture, this can probably * be reused. */ -static GLboolean transform_TEX( +GLboolean r300_transform_TEX( struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data) { @@ -160,6 +140,8 @@ static GLboolean transform_TEX( inst.DstReg.Index = tempreg; inst.DstReg.WriteMask = WRITEMASK_XYZW; destredirect = GL_TRUE; + } else if (inst.SaturateMode) { + destredirect = GL_TRUE; } } @@ -175,7 +157,7 @@ static GLboolean transform_TEX( inst.SrcReg[0].File = PROGRAM_TEMPORARY; inst.SrcReg[0].Index = tmpreg; } - + tgt = radeonAppendInstructions(t->Program, 1); _mesa_copy_instructions(tgt, &inst, 1); @@ -239,6 +221,7 @@ static GLboolean transform_TEX( tgt->Opcode = OPCODE_MOV; tgt->DstReg = orig_inst->DstReg; + tgt->SaturateMode = inst.SaturateMode; tgt->SrcReg[0].File = PROGRAM_TEMPORARY; tgt->SrcReg[0].Index = inst.DstReg.Index; } @@ -246,242 +229,10 @@ static GLboolean transform_TEX( return GL_TRUE; } - -static void update_params(r300ContextPtr r300, struct r300_fragment_program *fp) -{ - struct gl_fragment_program *mp = &fp->mesa_program; - - /* Ask Mesa nicely to fill in ParameterValues for us */ - if (mp->Base.Parameters) - _mesa_load_state_parameters(r300->radeon.glCtx, mp->Base.Parameters); -} - - -/** - * Transform the program to support fragment.position. - * - * Introduce a small fragment at the start of the program that will be - * the only code that directly reads the FRAG_ATTRIB_WPOS input. - * All other code pieces that reference that input will be rewritten - * to read from a newly allocated temporary. - * - * \todo if/when r5xx supports the radeon_program architecture, this is a - * likely candidate for code sharing. - */ -static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) -{ - GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead; - - if (!(InputsRead & FRAG_BIT_WPOS)) - return; - - static gl_state_index tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0 - }; - struct prog_instruction *fpi; - GLuint window_index; - int i = 0; - GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY); - - _mesa_insert_instructions(compiler->program, 0, 3); - fpi = compiler->program->Instructions; - - /* perspective divide */ - fpi[i].Opcode = OPCODE_RCP; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_W; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW; - i++; - - fpi[i].Opcode = OPCODE_MUL; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW; - - fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[1].Index = tempregi; - fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW; - i++; - - /* viewport transformation */ - window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens); - - fpi[i].Opcode = OPCODE_MAD; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[0].Index = tempregi; - fpi[i].SrcReg[0].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[1].Index = window_index; - fpi[i].SrcReg[1].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[2].Index = window_index; - fpi[i].SrcReg[2].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - i++; - - for (; i < compiler->program->NumInstructions; ++i) { - int reg; - for (reg = 0; reg < 3; reg++) { - if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT && - fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) { - fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[reg].Index = tempregi; - } - } - } -} - - -static void nqssadce_init(struct nqssadce_state* s) -{ - s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW; - s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W; -} - - -static GLuint build_dtm(GLuint depthmode) -{ - switch(depthmode) { - default: - case GL_LUMINANCE: return 0; - case GL_INTENSITY: return 1; - case GL_ALPHA: return 2; - } -} - -static GLuint build_func(GLuint comparefunc) -{ - return comparefunc - GL_NEVER; -} - - -/** - * Collect all external state that is relevant for compiling the given - * fragment program. - */ -static void build_state( - r300ContextPtr r300, - struct r300_fragment_program *fp, - struct r300_fragment_program_external_state *state) -{ - int unit; - - _mesa_bzero(state, sizeof(*state)); - - for(unit = 0; unit < 16; ++unit) { - if (fp->mesa_program.Base.ShadowSamplers & (1 << unit)) { - struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; - - state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); - state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); - } - } -} - - -void r300TranslateFragmentShader(r300ContextPtr r300, - struct r300_fragment_program *fp) -{ - struct r300_fragment_program_external_state state; - - build_state(r300, fp, &state); - if (_mesa_memcmp(&fp->state, &state, sizeof(state))) { - /* TODO: cache compiled programs */ - fp->translated = GL_FALSE; - _mesa_memcpy(&fp->state, &state, sizeof(state)); - } - - if (!fp->translated) { - struct r300_fragment_program_compiler compiler; - - compiler.r300 = r300; - compiler.fp = fp; - compiler.code = &fp->code; - compiler.program = _mesa_clone_program(r300->radeon.glCtx, &fp->mesa_program.Base); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Fragment Program: Initial program:\n"); - _mesa_print_program(compiler.program); - } - - insert_WPOS_trailer(&compiler); - - struct radeon_program_transformation transformations[] = { - { &transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformTrigSimple, 0 } - }; - radeonLocalTransform( - r300->radeon.glCtx, - compiler.program, - 3, transformations); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Fragment Program: After native rewrite:\n"); - _mesa_print_program(compiler.program); - } - - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &r300FPIsNativeSwizzle, - .BuildSwizzle = &r300FPBuildSwizzle, - .RewriteDepthOut = GL_TRUE - }; - radeonNqssaDce(r300->radeon.glCtx, compiler.program, &nqssadce); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: after NqSSA-DCE:\n"); - _mesa_print_program(compiler.program); - } - - if (!r300FragmentProgramEmit(&compiler)) - fp->error = GL_TRUE; - - /* Subtle: Rescue any parameters that have been added during transformations */ - _mesa_free_parameter_list(fp->mesa_program.Base.Parameters); - fp->mesa_program.Base.Parameters = compiler.program->Parameters; - compiler.program->Parameters = 0; - - _mesa_reference_program(r300->radeon.glCtx, &compiler.program, NULL); - - if (!fp->error) - fp->translated = GL_TRUE; - if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL)) - r300FragmentProgramDump(fp, &fp->code); - r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM | - _NEW_PROGRAM_CONSTANTS); - } - - update_params(r300, fp); -} - /* just some random things... */ -void r300FragmentProgramDump( - struct r300_fragment_program *fp, - struct r300_fragment_program_code *code) +void r300FragmentProgramDump(union rX00_fragment_program_code *c) { + struct r300_fragment_program_code *code = &c->r300; int n, i, j; static int pc = 0; diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.h b/src/mesa/drivers/dri/r300/r300_fragprog.h index 94fb554fb3..5ce6f33cee 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog.h +++ b/src/mesa/drivers/dri/r300/r300_fragprog.h @@ -33,9 +33,6 @@ #ifndef __R300_FRAGPROG_H_ #define __R300_FRAGPROG_H_ -#include "main/glheader.h" -#include "main/macros.h" -#include "main/enums.h" #include "shader/program.h" #include "shader/prog_instruction.h" @@ -105,28 +102,10 @@ #endif -struct r300_fragment_program; +extern GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); -extern void r300TranslateFragmentShader(r300ContextPtr r300, - struct r300_fragment_program *fp); +extern void r300FragmentProgramDump(union rX00_fragment_program_code *c); - -/** - * Used internally by the r300 fragment program code to store compile-time - * only data. - */ -struct r300_fragment_program_compiler { - r300ContextPtr r300; - struct r300_fragment_program *fp; - struct r300_fragment_program_code *code; - struct gl_program *program; -}; - -extern GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler); - - -extern void r300FragmentProgramDump( - struct r300_fragment_program *fp, - struct r300_fragment_program_code *code); +extern GLboolean r300_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data); #endif diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c new file mode 100644 index 0000000000..abc8757ba1 --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c @@ -0,0 +1,291 @@ +/* + * Copyright (C) 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +/** + * \file + * + * Fragment program compiler. Perform transformations on the intermediate + * representation until the program is in a form where we can translate + * it more or less directly into machine-readable form. + * + * \author Ben Skeggs <darktama@iinet.net.au> + * \author Jerome Glisse <j.glisse@gmail.com> + */ + +#include "r300_fragprog_common.h" + +#include "shader/program.h" +#include "shader/prog_parameter.h" +#include "shader/prog_print.h" + +#include "r300_state.h" +#include "r300_fragprog.h" +#include "r300_fragprog_swizzle.h" +#include "r500_fragprog.h" + +#include "radeon_program.h" +#include "radeon_program_alu.h" + +static void update_params(GLcontext *ctx, struct gl_fragment_program *fp) +{ + /* Ask Mesa nicely to fill in ParameterValues for us */ + if (fp->Base.Parameters) + _mesa_load_state_parameters(ctx, fp->Base.Parameters); +} + +static void nqssadce_init(struct nqssadce_state* s) +{ + s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW; + s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W; +} + +/** + * Transform the program to support fragment.position. + * + * Introduce a small fragment at the start of the program that will be + * the only code that directly reads the FRAG_ATTRIB_WPOS input. + * All other code pieces that reference that input will be rewritten + * to read from a newly allocated temporary. + * + */ +static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler) +{ + GLuint InputsRead = compiler->fp->Base.Base.InputsRead; + + if (!(InputsRead & FRAG_BIT_WPOS)) + return; + + static gl_state_index tokens[STATE_LENGTH] = { + STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0 + }; + struct prog_instruction *fpi; + GLuint window_index; + int i = 0; + GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY); + + _mesa_insert_instructions(compiler->program, 0, 3); + fpi = compiler->program->Instructions; + + /* perspective divide */ + fpi[i].Opcode = OPCODE_RCP; + + fpi[i].DstReg.File = PROGRAM_TEMPORARY; + fpi[i].DstReg.Index = tempregi; + fpi[i].DstReg.WriteMask = WRITEMASK_W; + fpi[i].DstReg.CondMask = COND_TR; + + fpi[i].SrcReg[0].File = PROGRAM_INPUT; + fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; + fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW; + i++; + + fpi[i].Opcode = OPCODE_MUL; + + fpi[i].DstReg.File = PROGRAM_TEMPORARY; + fpi[i].DstReg.Index = tempregi; + fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; + fpi[i].DstReg.CondMask = COND_TR; + + fpi[i].SrcReg[0].File = PROGRAM_INPUT; + fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; + fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW; + + fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY; + fpi[i].SrcReg[1].Index = tempregi; + fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW; + i++; + + /* viewport transformation */ + window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens); + + fpi[i].Opcode = OPCODE_MAD; + + fpi[i].DstReg.File = PROGRAM_TEMPORARY; + fpi[i].DstReg.Index = tempregi; + fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; + fpi[i].DstReg.CondMask = COND_TR; + + fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY; + fpi[i].SrcReg[0].Index = tempregi; + fpi[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + + fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR; + fpi[i].SrcReg[1].Index = window_index; + fpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + + fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR; + fpi[i].SrcReg[2].Index = window_index; + fpi[i].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); + i++; + + for (; i < compiler->program->NumInstructions; ++i) { + int reg; + for (reg = 0; reg < 3; reg++) { + if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT && + fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) { + fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY; + fpi[i].SrcReg[reg].Index = tempregi; + } + } + } +} + +static GLuint build_dtm(GLuint depthmode) +{ + switch(depthmode) { + default: + case GL_LUMINANCE: return 0; + case GL_INTENSITY: return 1; + case GL_ALPHA: return 2; + } +} + +static GLuint build_func(GLuint comparefunc) +{ + return comparefunc - GL_NEVER; +} + +/** + * Collect all external state that is relevant for compiling the given + * fragment program. + */ +static void build_state( + r300ContextPtr r300, + struct r300_fragment_program *fp, + struct r300_fragment_program_external_state *state) +{ + int unit; + + _mesa_bzero(state, sizeof(*state)); + + for(unit = 0; unit < 16; ++unit) { + if (fp->Base.Base.ShadowSamplers & (1 << unit)) { + struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; + + state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); + state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); + } + } +} + +void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp) +{ + r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)fp; + struct r300_fragment_program_external_state state; + + build_state(r300, r300_fp, &state); + if (_mesa_memcmp(&r300_fp->state, &state, sizeof(state))) { + /* TODO: cache compiled programs */ + r300_fp->translated = GL_FALSE; + _mesa_memcpy(&r300_fp->state, &state, sizeof(state)); + } + + if (!r300_fp->translated) { + struct r300_fragment_program_compiler compiler; + + compiler.r300 = r300; + compiler.fp = r300_fp; + compiler.code = &r300_fp->code; + compiler.program = _mesa_clone_program(ctx, &fp->Base); + + if (RADEON_DEBUG & DEBUG_PIXEL) { + fflush(stdout); + _mesa_printf("Fragment Program: Initial program:\n"); + _mesa_print_program(compiler.program); + fflush(stdout); + } + + insert_WPOS_trailer(&compiler); + + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + struct radeon_program_transformation transformations[] = { + { &r500_transform_TEX, &compiler }, + { &radeonTransformALU, 0 }, + { &radeonTransformDeriv, 0 }, + { &radeonTransformTrigScale, 0 } + }; + radeonLocalTransform(ctx, compiler.program, 4, transformations); + } else { + struct radeon_program_transformation transformations[] = { + { &r300_transform_TEX, &compiler }, + { &radeonTransformALU, 0 }, + { &radeonTransformTrigSimple, 0 } + }; + radeonLocalTransform(ctx, compiler.program, 3, transformations); + } + + if (RADEON_DEBUG & DEBUG_PIXEL) { + _mesa_printf("Fragment Program: After native rewrite:\n"); + _mesa_print_program(compiler.program); + fflush(stdout); + } + + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + struct radeon_nqssadce_descr nqssadce = { + .Init = &nqssadce_init, + .IsNativeSwizzle = &r500FPIsNativeSwizzle, + .BuildSwizzle = &r500FPBuildSwizzle, + .RewriteDepthOut = GL_TRUE + }; + radeonNqssaDce(ctx, compiler.program, &nqssadce); + } else { + struct radeon_nqssadce_descr nqssadce = { + .Init = &nqssadce_init, + .IsNativeSwizzle = &r300FPIsNativeSwizzle, + .BuildSwizzle = &r300FPBuildSwizzle, + .RewriteDepthOut = GL_TRUE + }; + radeonNqssaDce(ctx, compiler.program, &nqssadce); + } + + if (RADEON_DEBUG & DEBUG_PIXEL) { + _mesa_printf("Compiler: after NqSSA-DCE:\n"); + _mesa_print_program(compiler.program); + fflush(stdout); + } + + if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler)) + r300_fp->error = GL_TRUE; + + /* Subtle: Rescue any parameters that have been added during transformations */ + _mesa_free_parameter_list(fp->Base.Parameters); + fp->Base.Parameters = compiler.program->Parameters; + compiler.program->Parameters = 0; + + _mesa_reference_program(ctx, &compiler.program, NULL); + + r300_fp->translated = GL_TRUE; + + r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS); + + if (r300_fp->error || (RADEON_DEBUG & DEBUG_PIXEL)) + r300->vtbl.FragmentProgramDump(&r300_fp->code); + } + + update_params(ctx, fp); +} diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.h b/src/mesa/drivers/dri/r300/r300_fragprog_common.h new file mode 100644 index 0000000000..85ea86fecb --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.h @@ -0,0 +1,35 @@ +/* + * Copyright (C) 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef __R300_FRAGPROG_COMMON_H_ +#define __R300_FRAGPROG_COMMON_H_ + +#include "main/mtypes.h" + +extern void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp); + +#endif diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c index 9f0b7e3534..b75656e7ee 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c @@ -47,7 +47,7 @@ #define PROG_CODE \ struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \ - struct r300_fragment_program_code *code = c->code + struct r300_fragment_program_code *code = &c->code->r300 #define error(fmt, args...) do { \ fprintf(stderr, "%s::%s(): " fmt "\n", \ @@ -66,7 +66,7 @@ static GLboolean emit_const(void* data, GLuint file, GLuint index, GLuint *hwind } if (*hwindex >= code->const_nr) { - if (*hwindex >= PFS_NUM_CONST_REGS) { + if (*hwindex >= R300_PFS_NUM_CONST_REGS) { error("Out of hw constants!\n"); return GL_FALSE; } @@ -138,7 +138,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) { PROG_CODE; - if (code->alu.length >= PFS_MAX_ALU_INST) { + if (code->alu.length >= R300_PFS_MAX_ALU_INST) { error("Too many ALU instructions"); return GL_FALSE; } @@ -201,7 +201,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) if (inst->Alpha.DepthWriteMask) { code->alu.inst[ip].inst3 |= R300_ALU_DSTA_DEPTH; code->node[code->cur_node].flags |= R300_W_OUT; - c->fp->WritesDepth = GL_TRUE; + c->fp->writes_depth = GL_TRUE; } return GL_TRUE; @@ -213,7 +213,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) */ static GLboolean finish_node(struct r300_fragment_program_compiler *c) { - struct r300_fragment_program_code *code = c->code; + struct r300_fragment_program_code *code = &c->code->r300; struct r300_fragment_program_node *node = &code->node[code->cur_node]; if (node->alu_end < 0) { @@ -275,7 +275,7 @@ static GLboolean emit_tex(void* data, struct prog_instruction* inst) { PROG_CODE; - if (code->tex.length >= PFS_MAX_TEX_INST) { + if (code->tex.length >= R300_PFS_MAX_TEX_INST) { error("Too many TEX instructions"); return GL_FALSE; } @@ -318,16 +318,16 @@ static const struct radeon_pair_handler pair_handler = { .EmitPaired = &emit_alu, .EmitTex = &emit_tex, .BeginTexBlock = &begin_tex, - .MaxHwTemps = PFS_NUM_TEMP_REGS + .MaxHwTemps = R300_PFS_NUM_TEMP_REGS }; /** * Final compilation step: Turn the intermediate radeon_program into * machine-readable instructions. */ -GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler) +GLboolean r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) { - struct r300_fragment_program_code *code = compiler->code; + struct r300_fragment_program_code *code = &compiler->code->r300; _mesa_bzero(code, sizeof(struct r300_fragment_program_code)); code->node[0].alu_end = -1; diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c index ee85e229f0..104079b4db 100644 --- a/src/mesa/drivers/dri/r300/r300_ioctl.c +++ b/src/mesa/drivers/dri/r300/r300_ioctl.c @@ -46,8 +46,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/context.h" #include "swrast/swrast.h" +#include "radeon_common.h" +#include "radeon_lock.h" #include "r300_context.h" -#include "radeon_ioctl.h" #include "r300_ioctl.h" #include "r300_cmdbuf.h" #include "r300_state.h" @@ -55,71 +56,90 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "radeon_reg.h" #include "r300_emit.h" #include "r300_fragprog.h" +#include "r300_context.h" #include "vblank.h" +#define R200_3D_DRAW_IMMD_2 0xC0003500 + #define CLEARBUFFER_COLOR 0x1 #define CLEARBUFFER_DEPTH 0x2 #define CLEARBUFFER_STENCIL 0x4 -static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer) +static void r300EmitClearState(GLcontext * ctx); + +static void r300UserClear(GLcontext *ctx, GLuint mask) { + radeon_clear_tris(ctx, mask); +} + +static void r300ClearBuffer(r300ContextPtr r300, int flags, + struct radeon_renderbuffer *rrb, + struct radeon_renderbuffer *rrbd) +{ + BATCH_LOCALS(&r300->radeon); GLcontext *ctx = r300->radeon.glCtx; - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; - GLuint cboffset, cbpitch; - drm_r300_cmd_header_t *cmd2; - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); + GLuint cbpitch = 0; r300ContextPtr rmesa = r300; if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s: %s buffer (%i,%i %ix%i)\n", - __FUNCTION__, buffer ? "back" : "front", - dPriv->x, dPriv->y, dPriv->w, dPriv->h); - - if (buffer) { - cboffset = r300->radeon.radeonScreen->backOffset; - cbpitch = r300->radeon.radeonScreen->backPitch; - } else { - cboffset = r300->radeon.radeonScreen->frontOffset; - cbpitch = r300->radeon.radeonScreen->frontPitch; + fprintf(stderr, "%s: buffer %p (%i,%i %ix%i)\n", + __FUNCTION__, rrb, dPriv->x, dPriv->y, + dPriv->w, dPriv->h); + + if (rrb) { + cbpitch = (rrb->pitch / rrb->cpp); + if (rrb->cpp == 4) + cbpitch |= R300_COLOR_FORMAT_ARGB8888; + else + cbpitch |= R300_COLOR_FORMAT_RGB565; + + if (rrb->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){ + cbpitch |= R300_COLOR_TILE_ENABLE; + } } - cboffset += r300->radeon.radeonScreen->fbLocation; - - cp_wait(r300, R300_WAIT_3D | R300_WAIT_3D_CLEAN); - end_3d(rmesa); - - R300_STATECHANGE(r300, cb); - reg_start(R300_RB3D_COLOROFFSET0, 0); - e32(cboffset); - - if (r300->radeon.radeonScreen->cpp == 4) - cbpitch |= R300_COLOR_FORMAT_ARGB8888; - else - cbpitch |= R300_COLOR_FORMAT_RGB565; - - if (r300->radeon.sarea->tiling_enabled) - cbpitch |= R300_COLOR_TILE_ENABLE; - - reg_start(R300_RB3D_COLORPITCH0, 0); - e32(cbpitch); - - R300_STATECHANGE(r300, cmk); - reg_start(RB3D_COLOR_CHANNEL_MASK, 0); + /* TODO in bufmgr */ + cp_wait(&r300->radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN); + end_3d(&rmesa->radeon); if (flags & CLEARBUFFER_COLOR) { - e32((ctx->Color.ColorMask[BCOMP] ? RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0 : 0) | - (ctx->Color.ColorMask[GCOMP] ? RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0 : 0) | - (ctx->Color.ColorMask[RCOMP] ? RB3D_COLOR_CHANNEL_MASK_RED_MASK0 : 0) | - (ctx->Color.ColorMask[ACOMP] ? RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0 : 0)); + assert(rrb != 0); + BEGIN_BATCH_NO_AUTOSTATE(6); + OUT_BATCH_REGSEQ(R300_RB3D_COLOROFFSET0, 1); + OUT_BATCH_RELOC(0, rrb->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_BATCH_REGVAL(R300_RB3D_COLORPITCH0, cbpitch); + END_BATCH(); + } +#if 1 + if (flags & (CLEARBUFFER_DEPTH | CLEARBUFFER_STENCIL)) { + assert(rrbd != 0); + cbpitch = (rrbd->pitch / rrbd->cpp); + if (rrbd->bo->flags & RADEON_BO_FLAGS_MACRO_TILE){ + cbpitch |= R300_DEPTHMACROTILE_ENABLE; + } + if (rrbd->bo->flags & RADEON_BO_FLAGS_MICRO_TILE){ + cbpitch |= R300_DEPTHMICROTILE_TILED; + } + BEGIN_BATCH_NO_AUTOSTATE(6); + OUT_BATCH_REGSEQ(R300_ZB_DEPTHOFFSET, 1); + OUT_BATCH_RELOC(0, rrbd->bo, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_BATCH_REGVAL(R300_ZB_DEPTHPITCH, cbpitch); + END_BATCH(); + } +#endif + BEGIN_BATCH_NO_AUTOSTATE(6); + OUT_BATCH_REGSEQ(RB3D_COLOR_CHANNEL_MASK, 1); + if (flags & CLEARBUFFER_COLOR) { + OUT_BATCH((ctx->Color.ColorMask[BCOMP] ? RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0 : 0) | + (ctx->Color.ColorMask[GCOMP] ? RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0 : 0) | + (ctx->Color.ColorMask[RCOMP] ? RB3D_COLOR_CHANNEL_MASK_RED_MASK0 : 0) | + (ctx->Color.ColorMask[ACOMP] ? RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0 : 0)); } else { - e32(0x0); + OUT_BATCH(0); } - R300_STATECHANGE(r300, zs); - reg_start(R300_ZB_CNTL, 2); { uint32_t t1, t2; @@ -146,73 +166,92 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer) R300_S_FRONT_ZFAIL_OP_SHIFT); } - e32(t1); - e32(t2); - e32(((ctx->Stencil.WriteMask[0] & R300_STENCILREF_MASK) << R300_STENCILWRITEMASK_SHIFT) | - (ctx->Stencil.Clear & R300_STENCILREF_MASK)); + OUT_BATCH_REGSEQ(R300_ZB_CNTL, 3); + OUT_BATCH(t1); + OUT_BATCH(t2); + OUT_BATCH(((ctx->Stencil.WriteMask[0] & R300_STENCILREF_MASK) << + R300_STENCILWRITEMASK_SHIFT) | + (ctx->Stencil.Clear & R300_STENCILREF_MASK)); + END_BATCH(); } - cmd2 = (drm_r300_cmd_header_t *) r300AllocCmdBuf(r300, 9, __FUNCTION__); - cmd2[0].packet3.cmd_type = R300_CMD_PACKET3; - cmd2[0].packet3.packet = R300_CMD_PACKET3_CLEAR; - cmd2[1].u = r300PackFloat32(dPriv->w / 2.0); - cmd2[2].u = r300PackFloat32(dPriv->h / 2.0); - cmd2[3].u = r300PackFloat32(ctx->Depth.Clear); - cmd2[4].u = r300PackFloat32(1.0); - cmd2[5].u = r300PackFloat32(ctx->Color.ClearColor[0]); - cmd2[6].u = r300PackFloat32(ctx->Color.ClearColor[1]); - cmd2[7].u = r300PackFloat32(ctx->Color.ClearColor[2]); - cmd2[8].u = r300PackFloat32(ctx->Color.ClearColor[3]); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(9); + OUT_BATCH(cmdpacket3(r300->radeon.radeonScreen, R300_CMD_PACKET3_CLEAR)); + OUT_BATCH_FLOAT32(dPriv->w / 2.0); + OUT_BATCH_FLOAT32(dPriv->h / 2.0); + OUT_BATCH_FLOAT32(ctx->Depth.Clear); + OUT_BATCH_FLOAT32(1.0); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[0]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[1]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[2]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[3]); + END_BATCH(); + } else { + OUT_BATCH(CP_PACKET3(R200_3D_DRAW_IMMD_2, 8)); + OUT_BATCH(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING | + (1 << R300_PRIM_NUM_VERTICES_SHIFT)); + OUT_BATCH_FLOAT32(dPriv->w / 2.0); + OUT_BATCH_FLOAT32(dPriv->h / 2.0); + OUT_BATCH_FLOAT32(ctx->Depth.Clear); + OUT_BATCH_FLOAT32(1.0); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[0]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[1]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[2]); + OUT_BATCH_FLOAT32(ctx->Color.ClearColor[3]); + } r300EmitCacheFlush(rmesa); - cp_wait(rmesa, R300_WAIT_3D | R300_WAIT_3D_CLEAN); + cp_wait(&r300->radeon, R300_WAIT_3D | R300_WAIT_3D_CLEAN); + + R300_STATECHANGE(r300, cb); + R300_STATECHANGE(r300, cmk); + R300_STATECHANGE(r300, zs); } static void r300EmitClearState(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); - r300ContextPtr rmesa = r300; - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; + BATCH_LOCALS(&r300->radeon); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); int i; - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; - int has_tcl = 1; + int has_tcl; int is_r500 = 0; GLuint vap_cntl; - if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) - has_tcl = 0; - - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) - is_r500 = 1; + has_tcl = r300->options.hw_tcl_enabled; + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) + is_r500 = 1; - /* FIXME: the values written to R300_VAP_INPUT_ROUTE_0_0 and - * R300_VAP_INPUT_ROUTE_0_1 are in fact known, however, the values are - * quite complex; see the functions in r300_emit.c. + /* State atom dirty tracking is a little subtle here. + * + * On the one hand, we need to make sure base state is emitted + * here if we start with an empty batch buffer, otherwise clear + * works incorrectly with multiple processes. Therefore, the first + * BEGIN_BATCH cannot be a BEGIN_BATCH_NO_AUTOSTATE. * - * I believe it would be a good idea to extend the functions in - * r300_emit.c so that they can be used to setup the default values for - * these registers, as well as the actual values used for rendering. + * On the other hand, implicit state emission clears the state atom + * dirty bits, so we have to call R300_STATECHANGE later than the + * first BEGIN_BATCH. + * + * The final trickiness is that, because we change state, we need + * to ensure that any stored swtcl primitives are flushed properly + * before we start changing state. See the R300_NEWPRIM in r300Clear + * for this. */ - R300_STATECHANGE(r300, vir[0]); - reg_start(R300_VAP_PROG_STREAM_CNTL_0, 0); + BEGIN_BATCH(31); + OUT_BATCH_REGSEQ(R300_VAP_PROG_STREAM_CNTL_0, 1); if (!has_tcl) - e32(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) | + OUT_BATCH(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) | ((R300_LAST_VEC | (2 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT))); else - e32(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) | + OUT_BATCH(((((0 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_0_SHIFT) | ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) | R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT))); - /* disable fog */ - R300_STATECHANGE(r300, fogs); - reg_start(R300_FG_FOG_BLEND, 0); - e32(0x0); - - R300_STATECHANGE(r300, vir[1]); - reg_start(R300_VAP_PROG_STREAM_CNTL_EXT_0, 0); - e32(((((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | + OUT_BATCH_REGVAL(R300_FG_FOG_BLEND, 0); + OUT_BATCH_REGVAL(R300_VAP_PROG_STREAM_CNTL_EXT_0, + ((((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_SHIFT) | (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_SHIFT) | (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_SHIFT) | @@ -226,619 +265,402 @@ static void r300EmitClearState(GLcontext * ctx) << R300_SWIZZLE1_SHIFT))); /* R300_VAP_INPUT_CNTL_0, R300_VAP_INPUT_CNTL_1 */ - R300_STATECHANGE(r300, vic); - reg_start(R300_VAP_VTX_STATE_CNTL, 1); - e32((R300_SEL_USER_COLOR_0 << R300_COLOR_0_ASSEMBLY_SHIFT)); - e32(R300_INPUT_CNTL_POS | R300_INPUT_CNTL_COLOR | R300_INPUT_CNTL_TC0); + OUT_BATCH_REGSEQ(R300_VAP_VTX_STATE_CNTL, 2); + OUT_BATCH((R300_SEL_USER_COLOR_0 << R300_COLOR_0_ASSEMBLY_SHIFT)); + OUT_BATCH(R300_INPUT_CNTL_POS | R300_INPUT_CNTL_COLOR | R300_INPUT_CNTL_TC0); - R300_STATECHANGE(r300, vte); /* comes from fglrx startup of clear */ - reg_start(R300_SE_VTE_CNTL, 1); - e32(R300_VTX_W0_FMT | R300_VPORT_X_SCALE_ENA | - R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA | - R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA | - R300_VPORT_Z_OFFSET_ENA); - e32(0x8); + OUT_BATCH_REGSEQ(R300_SE_VTE_CNTL, 2); + OUT_BATCH(R300_VTX_W0_FMT | R300_VPORT_X_SCALE_ENA | + R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA | + R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA | + R300_VPORT_Z_OFFSET_ENA); + OUT_BATCH(0x8); - reg_start(R300_VAP_PSC_SGN_NORM_CNTL, 0); - e32(0xaaaaaaaa); + OUT_BATCH_REGVAL(R300_VAP_PSC_SGN_NORM_CNTL, 0xaaaaaaaa); - R300_STATECHANGE(r300, vof); - reg_start(R300_VAP_OUTPUT_VTX_FMT_0, 1); - e32(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT | - R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT); - e32(0x0); /* no textures */ + OUT_BATCH_REGSEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); + OUT_BATCH(R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT | + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT); + OUT_BATCH(0); /* no textures */ - R300_STATECHANGE(r300, txe); - reg_start(R300_TX_ENABLE, 0); - e32(0x0); + OUT_BATCH_REGVAL(R300_TX_ENABLE, 0); - R300_STATECHANGE(r300, vpt); - reg_start(R300_SE_VPORT_XSCALE, 5); - efloat(1.0); - efloat(dPriv->x); - efloat(1.0); - efloat(dPriv->y); - efloat(1.0); - efloat(0.0); + OUT_BATCH_REGSEQ(R300_SE_VPORT_XSCALE, 6); + OUT_BATCH_FLOAT32(1.0); + OUT_BATCH_FLOAT32(dPriv->x); + OUT_BATCH_FLOAT32(1.0); + OUT_BATCH_FLOAT32(dPriv->y); + OUT_BATCH_FLOAT32(1.0); + OUT_BATCH_FLOAT32(0.0); - R300_STATECHANGE(r300, at); - reg_start(R300_FG_ALPHA_FUNC, 0); - e32(0x0); + OUT_BATCH_REGVAL(R300_FG_ALPHA_FUNC, 0); + + OUT_BATCH_REGSEQ(R300_RB3D_CBLEND, 2); + OUT_BATCH(0x0); + OUT_BATCH(0x0); + END_BATCH(); + R300_STATECHANGE(r300, vir[0]); + R300_STATECHANGE(r300, fogs); + R300_STATECHANGE(r300, vir[1]); + R300_STATECHANGE(r300, vic); + R300_STATECHANGE(r300, vte); + R300_STATECHANGE(r300, vof); + R300_STATECHANGE(r300, txe); + R300_STATECHANGE(r300, vpt); + R300_STATECHANGE(r300, at); R300_STATECHANGE(r300, bld); - reg_start(R300_RB3D_CBLEND, 1); - e32(0x0); - e32(0x0); + R300_STATECHANGE(r300, ps); if (has_tcl) { - R300_STATECHANGE(r300, vap_clip_cntl); - reg_start(R300_VAP_CLIP_CNTL, 0); - e32(R300_PS_UCP_MODE_CLIP_AS_TRIFAN | R300_CLIP_DISABLE); + R300_STATECHANGE(r300, vap_clip_cntl); + + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGVAL(R300_VAP_CLIP_CNTL, R300_PS_UCP_MODE_CLIP_AS_TRIFAN | R300_CLIP_DISABLE); + END_BATCH(); } - R300_STATECHANGE(r300, ps); - reg_start(R300_GA_POINT_SIZE, 0); - e32(((dPriv->w * 6) << R300_POINTSIZE_X_SHIFT) | - ((dPriv->h * 6) << R300_POINTSIZE_Y_SHIFT)); + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGVAL(R300_GA_POINT_SIZE, + ((dPriv->w * 6) << R300_POINTSIZE_X_SHIFT) | + ((dPriv->h * 6) << R300_POINTSIZE_Y_SHIFT)); + END_BATCH(); if (!is_r500) { R300_STATECHANGE(r300, ri); - reg_start(R300_RS_IP_0, 7); - for (i = 0; i < 8; ++i) { - e32(R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3)); - } - R300_STATECHANGE(r300, rc); - /* The second constant is needed to get glxgears display anything .. */ - reg_start(R300_RS_COUNT, 1); - e32((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN); - e32(0x0); - R300_STATECHANGE(r300, rr); - reg_start(R300_RS_INST_0, 0); - e32(R300_RS_INST_COL_CN_WRITE); + + BEGIN_BATCH(14); + OUT_BATCH_REGSEQ(R300_RS_IP_0, 8); + for (i = 0; i < 8; ++i) + OUT_BATCH(R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3)); + + OUT_BATCH_REGSEQ(R300_RS_COUNT, 2); + OUT_BATCH((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN); + OUT_BATCH(0x0); + + OUT_BATCH_REGVAL(R300_RS_INST_0, R300_RS_INST_COL_CN_WRITE); + END_BATCH(); } else { R300_STATECHANGE(r300, ri); - reg_start(R500_RS_IP_0, 7); + R300_STATECHANGE(r300, rc); + R300_STATECHANGE(r300, rr); + + BEGIN_BATCH(14); + OUT_BATCH_REGSEQ(R500_RS_IP_0, 8); for (i = 0; i < 8; ++i) { - e32((R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) | - (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) | - (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) | - (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT)); + OUT_BATCH((R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) | + (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) | + (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) | + (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT)); } - R300_STATECHANGE(r300, rc); - /* The second constant is needed to get glxgears display anything .. */ - reg_start(R300_RS_COUNT, 1); - e32((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN); - e32(0x0); - - R300_STATECHANGE(r300, rr); - reg_start(R500_RS_INST_0, 0); - e32(R500_RS_INST_COL_CN_WRITE); + OUT_BATCH_REGSEQ(R300_RS_COUNT, 2); + OUT_BATCH((1 << R300_IC_COUNT_SHIFT) | R300_HIRES_EN); + OUT_BATCH(0x0); + OUT_BATCH_REGVAL(R500_RS_INST_0, R500_RS_INST_COL_CN_WRITE); + END_BATCH(); } if (!is_r500) { R300_STATECHANGE(r300, fp); - reg_start(R300_US_CONFIG, 2); - e32(0x0); - e32(0x0); - e32(0x0); - reg_start(R300_US_CODE_ADDR_0, 3); - e32(0x0); - e32(0x0); - e32(0x0); - e32(R300_RGBA_OUT); - R300_STATECHANGE(r300, fpi[0]); R300_STATECHANGE(r300, fpi[1]); R300_STATECHANGE(r300, fpi[2]); R300_STATECHANGE(r300, fpi[3]); - reg_start(R300_US_ALU_RGB_INST_0, 0); - e32(FP_INSTRC(MAD, FP_ARGC(SRC0C_XYZ), FP_ARGC(ONE), FP_ARGC(ZERO))); - - reg_start(R300_US_ALU_RGB_ADDR_0, 0); - e32(FP_SELC(0, NO, XYZ, FP_TMP(0), 0, 0)); - - reg_start(R300_US_ALU_ALPHA_INST_0, 0); - e32(FP_INSTRA(MAD, FP_ARGA(SRC0A), FP_ARGA(ONE), FP_ARGA(ZERO))); - - reg_start(R300_US_ALU_ALPHA_ADDR_0, 0); - e32(FP_SELA(0, NO, W, FP_TMP(0), 0, 0)); + BEGIN_BATCH(17); + OUT_BATCH_REGSEQ(R300_US_CONFIG, 3); + OUT_BATCH(0x0); + OUT_BATCH(0x0); + OUT_BATCH(0x0); + OUT_BATCH_REGSEQ(R300_US_CODE_ADDR_0, 4); + OUT_BATCH(0x0); + OUT_BATCH(0x0); + OUT_BATCH(0x0); + OUT_BATCH(R300_RGBA_OUT); + + OUT_BATCH_REGVAL(R300_US_ALU_RGB_INST_0, + FP_INSTRC(MAD, FP_ARGC(SRC0C_XYZ), FP_ARGC(ONE), FP_ARGC(ZERO))); + OUT_BATCH_REGVAL(R300_US_ALU_RGB_ADDR_0, + FP_SELC(0, NO, XYZ, FP_TMP(0), 0, 0)); + OUT_BATCH_REGVAL(R300_US_ALU_ALPHA_INST_0, + FP_INSTRA(MAD, FP_ARGA(SRC0A), FP_ARGA(ONE), FP_ARGA(ZERO))); + OUT_BATCH_REGVAL(R300_US_ALU_ALPHA_ADDR_0, + FP_SELA(0, NO, W, FP_TMP(0), 0, 0)); + END_BATCH(); } else { - R300_STATECHANGE(r300, fp); - reg_start(R500_US_CONFIG, 1); - e32(R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); - e32(0x0); - reg_start(R500_US_CODE_ADDR, 2); - e32(R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(1)); - e32(R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(1)); - e32(R500_US_CODE_OFFSET_ADDR(0)); + struct radeon_state_atom r500fp; + uint32_t _cmd[10]; + R300_STATECHANGE(r300, fp); R300_STATECHANGE(r300, r500fp); - r500fp_start_fragment(0, 6); - - e32(R500_INST_TYPE_OUT | - R500_INST_TEX_SEM_WAIT | - R500_INST_LAST | - R500_INST_RGB_OMASK_R | - R500_INST_RGB_OMASK_G | - R500_INST_RGB_OMASK_B | - R500_INST_ALPHA_OMASK | - R500_INST_RGB_CLAMP | - R500_INST_ALPHA_CLAMP); - - e32(R500_RGB_ADDR0(0) | - R500_RGB_ADDR1(0) | - R500_RGB_ADDR1_CONST | - R500_RGB_ADDR2(0) | - R500_RGB_ADDR2_CONST); - - e32(R500_ALPHA_ADDR0(0) | - R500_ALPHA_ADDR1(0) | - R500_ALPHA_ADDR1_CONST | - R500_ALPHA_ADDR2(0) | - R500_ALPHA_ADDR2_CONST); - - e32(R500_ALU_RGB_SEL_A_SRC0 | - R500_ALU_RGB_R_SWIZ_A_R | - R500_ALU_RGB_G_SWIZ_A_G | - R500_ALU_RGB_B_SWIZ_A_B | - R500_ALU_RGB_SEL_B_SRC0 | - R500_ALU_RGB_R_SWIZ_B_R | - R500_ALU_RGB_B_SWIZ_B_G | - R500_ALU_RGB_G_SWIZ_B_B); - - e32(R500_ALPHA_OP_CMP | - R500_ALPHA_SWIZ_A_A | - R500_ALPHA_SWIZ_B_A); - - e32(R500_ALU_RGBA_OP_CMP | - R500_ALU_RGBA_R_SWIZ_0 | - R500_ALU_RGBA_G_SWIZ_0 | - R500_ALU_RGBA_B_SWIZ_0 | - R500_ALU_RGBA_A_SWIZ_0); + + BEGIN_BATCH(7); + OUT_BATCH_REGSEQ(R500_US_CONFIG, 2); + OUT_BATCH(R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); + OUT_BATCH(0x0); + OUT_BATCH_REGSEQ(R500_US_CODE_ADDR, 3); + OUT_BATCH(R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(1)); + OUT_BATCH(R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(1)); + OUT_BATCH(R500_US_CODE_OFFSET_ADDR(0)); + END_BATCH(); + + r500fp.check = check_r500fp; + r500fp.cmd = _cmd; + r500fp.cmd[0] = cmdr500fp(r300->radeon.radeonScreen, 0, 1, 0, 0); + r500fp.cmd[1] = R500_INST_TYPE_OUT | + R500_INST_TEX_SEM_WAIT | + R500_INST_LAST | + R500_INST_RGB_OMASK_R | + R500_INST_RGB_OMASK_G | + R500_INST_RGB_OMASK_B | + R500_INST_ALPHA_OMASK | + R500_INST_RGB_CLAMP | + R500_INST_ALPHA_CLAMP; + r500fp.cmd[2] = R500_RGB_ADDR0(0) | + R500_RGB_ADDR1(0) | + R500_RGB_ADDR1_CONST | + R500_RGB_ADDR2(0) | + R500_RGB_ADDR2_CONST; + r500fp.cmd[3] = R500_ALPHA_ADDR0(0) | + R500_ALPHA_ADDR1(0) | + R500_ALPHA_ADDR1_CONST | + R500_ALPHA_ADDR2(0) | + R500_ALPHA_ADDR2_CONST; + r500fp.cmd[4] = R500_ALU_RGB_SEL_A_SRC0 | + R500_ALU_RGB_R_SWIZ_A_R | + R500_ALU_RGB_G_SWIZ_A_G | + R500_ALU_RGB_B_SWIZ_A_B | + R500_ALU_RGB_SEL_B_SRC0 | + R500_ALU_RGB_R_SWIZ_B_R | + R500_ALU_RGB_B_SWIZ_B_G | + R500_ALU_RGB_G_SWIZ_B_B; + r500fp.cmd[5] = R500_ALPHA_OP_CMP | + R500_ALPHA_SWIZ_A_A | + R500_ALPHA_SWIZ_B_A; + r500fp.cmd[6] = R500_ALU_RGBA_OP_CMP | + R500_ALU_RGBA_R_SWIZ_0 | + R500_ALU_RGBA_G_SWIZ_0 | + R500_ALU_RGBA_B_SWIZ_0 | + R500_ALU_RGBA_A_SWIZ_0; + + r500fp.cmd[7] = 0; + emit_r500fp(ctx, &r500fp); } - reg_start(R300_VAP_PVS_STATE_FLUSH_REG, 0); - e32(0x00000000); + BEGIN_BATCH(2); + OUT_BATCH_REGVAL(R300_VAP_PVS_STATE_FLUSH_REG, 0); + END_BATCH(); + if (has_tcl) { - vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) | + vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) | (5 << R300_PVS_NUM_CNTLRS_SHIFT) | (12 << R300_VF_MAX_VTX_NUM_SHIFT)); - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) - vap_cntl |= R500_TCL_STATE_OPTIMIZATION; - } else - vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) | + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) + vap_cntl |= R500_TCL_STATE_OPTIMIZATION; + } else { + vap_cntl = ((10 << R300_PVS_NUM_SLOTS_SHIFT) | (5 << R300_PVS_NUM_CNTLRS_SHIFT) | (5 << R300_VF_MAX_VTX_NUM_SHIFT)); + } if (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV515) - vap_cntl |= (2 << R300_PVS_NUM_FPUS_SHIFT); + vap_cntl |= (2 << R300_PVS_NUM_FPUS_SHIFT); else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) || (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV560) || (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV570)) - vap_cntl |= (5 << R300_PVS_NUM_FPUS_SHIFT); + vap_cntl |= (5 << R300_PVS_NUM_FPUS_SHIFT); else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV410) || (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R420)) - vap_cntl |= (6 << R300_PVS_NUM_FPUS_SHIFT); + vap_cntl |= (6 << R300_PVS_NUM_FPUS_SHIFT); else if ((r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R520) || (r300->radeon.radeonScreen->chip_family == CHIP_FAMILY_R580)) - vap_cntl |= (8 << R300_PVS_NUM_FPUS_SHIFT); + vap_cntl |= (8 << R300_PVS_NUM_FPUS_SHIFT); else - vap_cntl |= (4 << R300_PVS_NUM_FPUS_SHIFT); + vap_cntl |= (4 << R300_PVS_NUM_FPUS_SHIFT); + + R300_STATECHANGE(r300, vap_cntl); - R300_STATECHANGE(rmesa, vap_cntl); - reg_start(R300_VAP_CNTL, 0); - e32(vap_cntl); + BEGIN_BATCH(2); + OUT_BATCH_REGVAL(R300_VAP_CNTL, vap_cntl); + END_BATCH(); if (has_tcl) { + struct radeon_state_atom vpu; + uint32_t _cmd[10]; R300_STATECHANGE(r300, pvs); - reg_start(R300_VAP_PVS_CODE_CNTL_0, 2); - - e32((0 << R300_PVS_FIRST_INST_SHIFT) | - (0 << R300_PVS_XYZW_VALID_INST_SHIFT) | - (1 << R300_PVS_LAST_INST_SHIFT)); - e32((0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | - (0 << R300_PVS_MAX_CONST_ADDR_SHIFT)); - e32(1 << R300_PVS_LAST_VTX_SRC_INST_SHIFT); - R300_STATECHANGE(r300, vpi); - vsf_start_fragment(0x0, 8); - e32(PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 0, 0xf, PVS_DST_REG_OUT)); - e32(PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE)); - e32(PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_REG_INPUT, VSF_FLAG_NONE)); - e32(0x0); - - e32(PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf, PVS_DST_REG_OUT)); - e32(PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE)); - e32(PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_SELECT_FORCE_0, PVS_SRC_REG_INPUT, VSF_FLAG_NONE)); - e32(0x0); + BEGIN_BATCH(4); + OUT_BATCH_REGSEQ(R300_VAP_PVS_CODE_CNTL_0, 3); + OUT_BATCH((0 << R300_PVS_FIRST_INST_SHIFT) | + (0 << R300_PVS_XYZW_VALID_INST_SHIFT) | + (1 << R300_PVS_LAST_INST_SHIFT)); + OUT_BATCH((0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | + (0 << R300_PVS_MAX_CONST_ADDR_SHIFT)); + OUT_BATCH(1 << R300_PVS_LAST_VTX_SRC_INST_SHIFT); + END_BATCH(); + + vpu.check = check_vpu; + vpu.cmd = _cmd; + vpu.cmd[0] = cmdvpu(r300->radeon.radeonScreen, 0, 2); + + vpu.cmd[1] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, + 0, 0xf, PVS_DST_REG_OUT); + vpu.cmd[2] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, + PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, + PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + vpu.cmd[3] = PVS_SRC_OPERAND(0, PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + vpu.cmd[4] = 0x0; + + vpu.cmd[5] = PVS_OP_DST_OPERAND(VE_ADD, GL_FALSE, GL_FALSE, 1, 0xf, + PVS_DST_REG_OUT); + vpu.cmd[6] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_X, + PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, + PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, + + VSF_FLAG_NONE); + vpu.cmd[7] = PVS_SRC_OPERAND(1, PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_SELECT_FORCE_0, + PVS_SRC_REG_INPUT, VSF_FLAG_NONE); + vpu.cmd[8] = 0x0; + + r300->vap_flush_needed = GL_TRUE; + emit_vpu(ctx, &vpu); } } -/** - * Buffer clear - */ -static void r300Clear(GLcontext * ctx, GLbitfield mask) +static void r300KernelClear(GLcontext *ctx, GLuint flags) { r300ContextPtr r300 = R300_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = r300->radeon.dri.drawable; - int flags = 0; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); + struct radeon_framebuffer *rfb = dPriv->driverPrivate; + struct radeon_renderbuffer *rrb; + struct radeon_renderbuffer *rrbd; int bits = 0; - int swapped; - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "r300Clear\n"); - - { - LOCK_HARDWARE(&r300->radeon); - UNLOCK_HARDWARE(&r300->radeon); - if (dPriv->numClipRects == 0) - return; - } - if (mask & BUFFER_BIT_FRONT_LEFT) { - flags |= BUFFER_BIT_FRONT_LEFT; - mask &= ~BUFFER_BIT_FRONT_LEFT; - } - - if (mask & BUFFER_BIT_BACK_LEFT) { - flags |= BUFFER_BIT_BACK_LEFT; - mask &= ~BUFFER_BIT_BACK_LEFT; - } - - if (mask & BUFFER_BIT_DEPTH) { + /* Make sure it fits there. */ + rcommonEnsureCmdBufSpace(&r300->radeon, 421 * 3, __FUNCTION__); + if (flags || bits) + r300EmitClearState(ctx); + rrbd = radeon_get_renderbuffer(&rfb->base, BUFFER_DEPTH); + if (rrbd && (flags & BUFFER_BIT_DEPTH)) bits |= CLEARBUFFER_DEPTH; - mask &= ~BUFFER_BIT_DEPTH; - } - if ((mask & BUFFER_BIT_STENCIL) && r300->state.stencil.hw_stencil) { + if (rrbd && (flags & BUFFER_BIT_STENCIL)) bits |= CLEARBUFFER_STENCIL; - mask &= ~BUFFER_BIT_STENCIL; - } - if (mask) { - if (RADEON_DEBUG & DEBUG_FALLBACKS) - fprintf(stderr, "%s: swrast clear, mask: %x\n", - __FUNCTION__, mask); - _swrast_Clear(ctx, mask); + if (flags & BUFFER_BIT_COLOR0) { + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_COLOR0); + r300ClearBuffer(r300, CLEARBUFFER_COLOR, rrb, NULL); + bits = 0; } - swapped = r300->radeon.sarea->pfCurrentPage == 1; - - /* Make sure it fits there. */ - r300EnsureCmdBufSpace(r300, 421 * 3, __FUNCTION__); - if (flags || bits) - r300EmitClearState(ctx); - if (flags & BUFFER_BIT_FRONT_LEFT) { - r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, swapped); + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_FRONT_LEFT); + r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd); bits = 0; } if (flags & BUFFER_BIT_BACK_LEFT) { - r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, swapped ^ 1); + rrb = radeon_get_renderbuffer(&rfb->base, BUFFER_BACK_LEFT); + r300ClearBuffer(r300, bits | CLEARBUFFER_COLOR, rrb, rrbd); bits = 0; } if (bits) - r300ClearBuffer(r300, bits, 0); + r300ClearBuffer(r300, bits, NULL, rrbd); + COMMIT_BATCH(); } -void r300Flush(GLcontext * ctx) +/** + * Buffer clear + */ +static void r300Clear(GLcontext * ctx, GLbitfield mask) { - r300ContextPtr rmesa = R300_CONTEXT(ctx); + r300ContextPtr r300 = R300_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&r300->radeon); + const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask); + GLbitfield swrast_mask = 0, tri_mask = 0; + int i; + struct gl_framebuffer *fb = ctx->DrawBuffer; if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); - - if (rmesa->cmdbuf.count_used > rmesa->cmdbuf.count_reemit) - r300FlushCmdBuf(rmesa, __FUNCTION__); -} - -#ifdef USER_BUFFERS -#include "r300_mem.h" - -void r300RefillCurrentDmaRegion(r300ContextPtr rmesa, int size) -{ - struct r300_dma_buffer *dmabuf; - size = MAX2(size, RADEON_BUFFER_SIZE * 16); - - if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->dma.flush) { - rmesa->dma.flush(rmesa); - } + fprintf(stderr, "r300Clear\n"); - if (rmesa->dma.current.buf) { -#ifdef USER_BUFFERS - r300_mem_use(rmesa, rmesa->dma.current.buf->id); -#endif - r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__); + if (!r300->radeon.radeonScreen->driScreen->dri2.enabled) { + LOCK_HARDWARE(&r300->radeon); + UNLOCK_HARDWARE(&r300->radeon); + if (dPriv->numClipRects == 0) + return; } - if (rmesa->dma.nr_released_bufs > 4) - r300FlushCmdBuf(rmesa, __FUNCTION__); - - dmabuf = CALLOC_STRUCT(r300_dma_buffer); - dmabuf->buf = (void *)1; /* hack */ - dmabuf->refcount = 1; - dmabuf->id = r300_mem_alloc(rmesa, 4, size); - if (dmabuf->id == 0) { - LOCK_HARDWARE(&rmesa->radeon); /* no need to validate */ - - r300FlushCmdBufLocked(rmesa, __FUNCTION__); - radeonWaitForIdleLocked(&rmesa->radeon); + /* Flush swtcl vertices if necessary, because we will change hardware + * state during clear. See also the state-related comment in + * r300EmitClearState. + */ + R300_NEWPRIM(r300); - dmabuf->id = r300_mem_alloc(rmesa, 4, size); + if (colorMask == ~0) + tri_mask |= (mask & BUFFER_BITS_COLOR); - UNLOCK_HARDWARE(&rmesa->radeon); - if (dmabuf->id == 0) { - fprintf(stderr, - "Error: Could not get dma buffer... exiting\n"); - _mesa_exit(-1); - } + /* HW stencil */ + if (mask & BUFFER_BIT_STENCIL) { + tri_mask |= BUFFER_BIT_STENCIL; } - rmesa->dma.current.buf = dmabuf; - rmesa->dma.current.address = r300_mem_ptr(rmesa, dmabuf->id); - rmesa->dma.current.end = size; - rmesa->dma.current.start = 0; - rmesa->dma.current.ptr = 0; -} - -void r300ReleaseDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, const char *caller) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (!region->buf) - return; - - if (rmesa->dma.flush) - rmesa->dma.flush(rmesa); - - if (--region->buf->refcount == 0) { - r300_mem_free(rmesa, region->buf->id); - FREE(region->buf); - rmesa->dma.nr_released_bufs++; + /* HW depth */ + if (mask & BUFFER_BIT_DEPTH) { + tri_mask |= BUFFER_BIT_DEPTH; } - region->buf = 0; - region->start = 0; -} - -/* Allocates a region from rmesa->dma.current. If there isn't enough - * space in current, grab a new buffer (and discard what was left of current) - */ -void r300AllocDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, - int bytes, int alignment) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, bytes); - - if (rmesa->dma.flush) - rmesa->dma.flush(rmesa); - - if (region->buf) - r300ReleaseDmaRegion(rmesa, region, __FUNCTION__); - - alignment--; - rmesa->dma.current.start = rmesa->dma.current.ptr = - (rmesa->dma.current.ptr + alignment) & ~alignment; - - if (rmesa->dma.current.ptr + bytes > rmesa->dma.current.end) - r300RefillCurrentDmaRegion(rmesa, (bytes + 0x7) & ~0x7); - - region->start = rmesa->dma.current.start; - region->ptr = rmesa->dma.current.start; - region->end = rmesa->dma.current.start + bytes; - region->address = rmesa->dma.current.address; - region->buf = rmesa->dma.current.buf; - region->buf->refcount++; - - rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */ - rmesa->dma.current.start = - rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7; - - assert(rmesa->dma.current.ptr <= rmesa->dma.current.end); -} + /* If we're doing a tri pass for depth/stencil, include a likely color + * buffer with it. + */ -#else -static void r300RefillCurrentDmaRegion(r300ContextPtr rmesa) -{ - struct r300_dma_buffer *dmabuf; - int fd = rmesa->radeon.dri.fd; - int index = 0; - int size = 0; - drmDMAReq dma; - int ret; - - if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->dma.flush) { - rmesa->dma.flush(rmesa); + for (i = 0; i < BUFFER_COUNT; i++) { + GLuint bufBit = 1 << i; + if ((tri_mask) & bufBit) { + if (!fb->Attachment[i].Renderbuffer->ClassID) { + tri_mask &= ~bufBit; + swrast_mask |= bufBit; + } + } } - if (rmesa->dma.current.buf) - r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__); - - if (rmesa->dma.nr_released_bufs > 4) - r300FlushCmdBuf(rmesa, __FUNCTION__); - - dma.context = rmesa->radeon.dri.hwContext; - dma.send_count = 0; - dma.send_list = NULL; - dma.send_sizes = NULL; - dma.flags = 0; - dma.request_count = 1; - dma.request_size = RADEON_BUFFER_SIZE; - dma.request_list = &index; - dma.request_sizes = &size; - dma.granted_count = 0; - - LOCK_HARDWARE(&rmesa->radeon); /* no need to validate */ - - ret = drmDMA(fd, &dma); - - if (ret != 0) { - /* Try to release some buffers and wait until we can't get any more */ - if (rmesa->dma.nr_released_bufs) { - r300FlushCmdBufLocked(rmesa, __FUNCTION__); - } - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Waiting for buffers\n"); - - radeonWaitForIdleLocked(&rmesa->radeon); - ret = drmDMA(fd, &dma); + /* SW fallback clearing */ + swrast_mask = mask & ~tri_mask; - if (ret != 0) { - UNLOCK_HARDWARE(&rmesa->radeon); - fprintf(stderr, - "Error: Could not get dma buffer... exiting\n"); - _mesa_exit(-1); - } + if (tri_mask) { + if (r300->radeon.radeonScreen->kernel_mm) + r300UserClear(ctx, tri_mask); + else + r300KernelClear(ctx, tri_mask); } - - UNLOCK_HARDWARE(&rmesa->radeon); - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Allocated buffer %d\n", index); - - dmabuf = CALLOC_STRUCT(r300_dma_buffer); - dmabuf->buf = &rmesa->radeon.radeonScreen->buffers->list[index]; - dmabuf->refcount = 1; - - rmesa->dma.current.buf = dmabuf; - rmesa->dma.current.address = dmabuf->buf->address; - rmesa->dma.current.end = dmabuf->buf->total; - rmesa->dma.current.start = 0; - rmesa->dma.current.ptr = 0; -} - -void r300ReleaseDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, const char *caller) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (!region->buf) - return; - - if (rmesa->dma.flush) - rmesa->dma.flush(rmesa); - - if (--region->buf->refcount == 0) { - drm_radeon_cmd_header_t *cmd; - - if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) - fprintf(stderr, "%s -- DISCARD BUF %d\n", - __FUNCTION__, region->buf->buf->idx); - cmd = - (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, - sizeof - (*cmd) / 4, - __FUNCTION__); - cmd->dma.cmd_type = R300_CMD_DMA_DISCARD; - cmd->dma.buf_idx = region->buf->buf->idx; - - FREE(region->buf); - rmesa->dma.nr_released_bufs++; + if (swrast_mask) { + if (RADEON_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, "%s: swrast clear, mask: %x\n", + __FUNCTION__, swrast_mask); + _swrast_Clear(ctx, swrast_mask); } - - region->buf = 0; - region->start = 0; -} - -/* Allocates a region from rmesa->dma.current. If there isn't enough - * space in current, grab a new buffer (and discard what was left of current) - */ -void r300AllocDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, - int bytes, int alignment) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, bytes); - - if (rmesa->dma.flush) - rmesa->dma.flush(rmesa); - - if (region->buf) - r300ReleaseDmaRegion(rmesa, region, __FUNCTION__); - - alignment--; - rmesa->dma.current.start = rmesa->dma.current.ptr = - (rmesa->dma.current.ptr + alignment) & ~alignment; - - if (rmesa->dma.current.ptr + bytes > rmesa->dma.current.end) - r300RefillCurrentDmaRegion(rmesa); - - region->start = rmesa->dma.current.start; - region->ptr = rmesa->dma.current.start; - region->end = rmesa->dma.current.start + bytes; - region->address = rmesa->dma.current.address; - region->buf = rmesa->dma.current.buf; - region->buf->refcount++; - - rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */ - rmesa->dma.current.start = - rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7; - - assert(rmesa->dma.current.ptr <= rmesa->dma.current.end); } -#endif - -GLboolean r300IsGartMemory(r300ContextPtr rmesa, const GLvoid * pointer, - GLint size) -{ - int offset = - (char *)pointer - - (char *)rmesa->radeon.radeonScreen->gartTextures.map; - int valid = (size >= 0 && offset >= 0 - && offset + size < - rmesa->radeon.radeonScreen->gartTextures.size); - - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "r300IsGartMemory( %p ) : %d\n", pointer, - valid); - - return valid; -} - -GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa, const GLvoid * pointer) -{ - int offset = - (char *)pointer - - (char *)rmesa->radeon.radeonScreen->gartTextures.map; - - //fprintf(stderr, "offset=%08x\n", offset); - - if (offset < 0 - || offset > rmesa->radeon.radeonScreen->gartTextures.size) - return ~0; - else - return rmesa->radeon.radeonScreen->gart_texture_offset + offset; -} void r300InitIoctlFuncs(struct dd_function_table *functions) { functions->Clear = r300Clear; functions->Finish = radeonFinish; - functions->Flush = r300Flush; + functions->Flush = radeonFlush; } diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.h b/src/mesa/drivers/dri/r300/r300_ioctl.h index e1143fb6c3..3abfa71a6e 100644 --- a/src/mesa/drivers/dri/r300/r300_ioctl.h +++ b/src/mesa/drivers/dri/r300/r300_ioctl.h @@ -39,22 +39,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_context.h" #include "radeon_drm.h" -extern GLboolean r300IsGartMemory(r300ContextPtr rmesa, - const GLvoid * pointer, GLint size); - -extern GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa, - const GLvoid * pointer); - -extern void r300Flush(GLcontext * ctx); - -extern void r300ReleaseDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, - const char *caller); -extern void r300AllocDmaRegion(r300ContextPtr rmesa, - struct r300_dma_region *region, int bytes, - int alignment); - extern void r300InitIoctlFuncs(struct dd_function_table *functions); -extern void r300RefillCurrentDmaRegion(r300ContextPtr rmesa, int size); #endif /* __R300_IOCTL_H__ */ diff --git a/src/mesa/drivers/dri/r300/r300_mem.c b/src/mesa/drivers/dri/r300/r300_mem.c deleted file mode 100644 index f8f9d4fcdf..0000000000 --- a/src/mesa/drivers/dri/r300/r300_mem.c +++ /dev/null @@ -1,385 +0,0 @@ -/* - * Copyright (C) 2005 Aapo Tahkola. - * - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining - * a copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sublicense, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial - * portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, - * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. - * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE - * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION - * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION - * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - */ - -/** - * \file - * - * \author Aapo Tahkola <aet@rasterburn.org> - */ - -#include <unistd.h> - -#include "r300_context.h" -#include "r300_cmdbuf.h" -#include "r300_ioctl.h" -#include "r300_mem.h" -#include "radeon_ioctl.h" - -#ifdef USER_BUFFERS - -static void resize_u_list(r300ContextPtr rmesa) -{ - void *temp; - int nsize; - - temp = rmesa->rmm->u_list; - nsize = rmesa->rmm->u_size * 2; - - rmesa->rmm->u_list = _mesa_malloc(nsize * sizeof(*rmesa->rmm->u_list)); - _mesa_memset(rmesa->rmm->u_list, 0, - nsize * sizeof(*rmesa->rmm->u_list)); - - if (temp) { - r300FlushCmdBuf(rmesa, __FUNCTION__); - - _mesa_memcpy(rmesa->rmm->u_list, temp, - rmesa->rmm->u_size * sizeof(*rmesa->rmm->u_list)); - _mesa_free(temp); - } - - rmesa->rmm->u_size = nsize; -} - -void r300_mem_init(r300ContextPtr rmesa) -{ - rmesa->rmm = malloc(sizeof(struct r300_memory_manager)); - memset(rmesa->rmm, 0, sizeof(struct r300_memory_manager)); - - rmesa->rmm->u_size = 128; - resize_u_list(rmesa); -} - -void r300_mem_destroy(r300ContextPtr rmesa) -{ - _mesa_free(rmesa->rmm->u_list); - rmesa->rmm->u_list = NULL; - - _mesa_free(rmesa->rmm); - rmesa->rmm = NULL; -} - -void *r300_mem_ptr(r300ContextPtr rmesa, int id) -{ - assert(id <= rmesa->rmm->u_last); - return rmesa->rmm->u_list[id].ptr; -} - -int r300_mem_find(r300ContextPtr rmesa, void *ptr) -{ - int i; - - for (i = 1; i < rmesa->rmm->u_size + 1; i++) - if (rmesa->rmm->u_list[i].ptr && - ptr >= rmesa->rmm->u_list[i].ptr && - ptr < - rmesa->rmm->u_list[i].ptr + rmesa->rmm->u_list[i].size) - break; - - if (i < rmesa->rmm->u_size + 1) - return i; - - fprintf(stderr, "%p failed\n", ptr); - return 0; -} - -//#define MM_DEBUG -int r300_mem_alloc(r300ContextPtr rmesa, int alignment, int size) -{ - drm_radeon_mem_alloc_t alloc; - int offset = 0, ret; - int i, free = -1; - int done_age; - drm_radeon_mem_free_t memfree; - int tries = 0; - static int bytes_wasted = 0, allocated = 0; - - if (size < 4096) - bytes_wasted += 4096 - size; - - allocated += size; - -#if 0 - static int t = 0; - if (t != time(NULL)) { - t = time(NULL); - fprintf(stderr, "slots used %d, wasted %d kb, allocated %d\n", - rmesa->rmm->u_last, bytes_wasted / 1024, - allocated / 1024); - } -#endif - - memfree.region = RADEON_MEM_REGION_GART; - - again: - - done_age = radeonGetAge((radeonContextPtr) rmesa); - - if (rmesa->rmm->u_last + 1 >= rmesa->rmm->u_size) - resize_u_list(rmesa); - - for (i = rmesa->rmm->u_last + 1; i > 0; i--) { - if (rmesa->rmm->u_list[i].ptr == NULL) { - free = i; - continue; - } - - if (rmesa->rmm->u_list[i].h_pending == 0 && - rmesa->rmm->u_list[i].pending - && rmesa->rmm->u_list[i].age <= done_age) { - memfree.region_offset = - (char *)rmesa->rmm->u_list[i].ptr - - (char *)rmesa->radeon.radeonScreen->gartTextures. - map; - - ret = - drmCommandWrite(rmesa->radeon.radeonScreen-> - driScreen->fd, DRM_RADEON_FREE, - &memfree, sizeof(memfree)); - - if (ret) { - fprintf(stderr, "Failed to free at %p\n", - rmesa->rmm->u_list[i].ptr); - fprintf(stderr, "ret = %s\n", strerror(-ret)); - exit(1); - } else { -#ifdef MM_DEBUG - fprintf(stderr, "really freed %d at age %x\n", - i, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - if (i == rmesa->rmm->u_last) - rmesa->rmm->u_last--; - - if (rmesa->rmm->u_list[i].size < 4096) - bytes_wasted -= - 4096 - rmesa->rmm->u_list[i].size; - - allocated -= rmesa->rmm->u_list[i].size; - rmesa->rmm->u_list[i].pending = 0; - rmesa->rmm->u_list[i].ptr = NULL; - free = i; - } - } - } - rmesa->rmm->u_head = i; - - if (free == -1) { - WARN_ONCE("Ran out of slots!\n"); - //usleep(100); - r300FlushCmdBuf(rmesa, __FUNCTION__); - tries++; - if (tries > 100) { - WARN_ONCE("Ran out of slots!\n"); - exit(1); - } - goto again; - } - - alloc.region = RADEON_MEM_REGION_GART; - alloc.alignment = alignment; - alloc.size = size; - alloc.region_offset = &offset; - - ret = - drmCommandWriteRead(rmesa->radeon.dri.fd, DRM_RADEON_ALLOC, &alloc, - sizeof(alloc)); - if (ret) { -#if 0 - WARN_ONCE("Ran out of mem!\n"); - r300FlushCmdBuf(rmesa, __FUNCTION__); - //usleep(100); - tries2++; - tries = 0; - if (tries2 > 100) { - WARN_ONCE("Ran out of GART memory!\n"); - exit(1); - } - goto again; -#else - WARN_ONCE - ("Ran out of GART memory (for %d)!\nPlease consider adjusting GARTSize option.\n", - size); - return 0; -#endif - } - - i = free; - - if (i > rmesa->rmm->u_last) - rmesa->rmm->u_last = i; - - rmesa->rmm->u_list[i].ptr = - ((GLubyte *) rmesa->radeon.radeonScreen->gartTextures.map) + offset; - rmesa->rmm->u_list[i].size = size; - rmesa->rmm->u_list[i].age = 0; - //fprintf(stderr, "alloc %p at id %d\n", rmesa->rmm->u_list[i].ptr, i); - -#ifdef MM_DEBUG - fprintf(stderr, "allocated %d at age %x\n", i, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - - return i; -} - -void r300_mem_use(r300ContextPtr rmesa, int id) -{ - uint64_t ull; -#ifdef MM_DEBUG - fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - drm_r300_cmd_header_t *cmd; - - assert(id <= rmesa->rmm->u_last); - - if (id == 0) - return; - - cmd = - (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, - 2 + sizeof(ull) / 4, - __FUNCTION__); - cmd[0].scratch.cmd_type = R300_CMD_SCRATCH; - cmd[0].scratch.reg = R300_MEM_SCRATCH; - cmd[0].scratch.n_bufs = 1; - cmd[0].scratch.flags = 0; - cmd++; - - ull = (uint64_t) (intptr_t) & rmesa->rmm->u_list[id].age; - _mesa_memcpy(cmd, &ull, sizeof(ull)); - cmd += sizeof(ull) / 4; - - cmd[0].u = /*id */ 0; - - LOCK_HARDWARE(&rmesa->radeon); /* Protect from DRM. */ - rmesa->rmm->u_list[id].h_pending++; - UNLOCK_HARDWARE(&rmesa->radeon); -} - -unsigned long r300_mem_offset(r300ContextPtr rmesa, int id) -{ - unsigned long offset; - - assert(id <= rmesa->rmm->u_last); - - offset = (char *)rmesa->rmm->u_list[id].ptr - - (char *)rmesa->radeon.radeonScreen->gartTextures.map; - offset += rmesa->radeon.radeonScreen->gart_texture_offset; - - return offset; -} - -void *r300_mem_map(r300ContextPtr rmesa, int id, int access) -{ -#ifdef MM_DEBUG - fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - void *ptr; - int tries = 0; - - assert(id <= rmesa->rmm->u_last); - - if (access == R300_MEM_R) { - - if (rmesa->rmm->u_list[id].mapped == 1) - WARN_ONCE("buffer %d already mapped\n", id); - - rmesa->rmm->u_list[id].mapped = 1; - ptr = r300_mem_ptr(rmesa, id); - - return ptr; - } - - if (rmesa->rmm->u_list[id].h_pending) - r300FlushCmdBuf(rmesa, __FUNCTION__); - - if (rmesa->rmm->u_list[id].h_pending) { - return NULL; - } - - while (rmesa->rmm->u_list[id].age > - radeonGetAge((radeonContextPtr) rmesa) && tries++ < 1000) - usleep(10); - - if (tries >= 1000) { - fprintf(stderr, "Idling failed (%x vs %x)\n", - rmesa->rmm->u_list[id].age, - radeonGetAge((radeonContextPtr) rmesa)); - return NULL; - } - - if (rmesa->rmm->u_list[id].mapped == 1) - WARN_ONCE("buffer %d already mapped\n", id); - - rmesa->rmm->u_list[id].mapped = 1; - ptr = r300_mem_ptr(rmesa, id); - - return ptr; -} - -void r300_mem_unmap(r300ContextPtr rmesa, int id) -{ -#ifdef MM_DEBUG - fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - - assert(id <= rmesa->rmm->u_last); - - if (rmesa->rmm->u_list[id].mapped == 0) - WARN_ONCE("buffer %d not mapped\n", id); - - rmesa->rmm->u_list[id].mapped = 0; -} - -void r300_mem_free(r300ContextPtr rmesa, int id) -{ -#ifdef MM_DEBUG - fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, - radeonGetAge((radeonContextPtr) rmesa)); -#endif - - assert(id <= rmesa->rmm->u_last); - - if (id == 0) - return; - - if (rmesa->rmm->u_list[id].ptr == NULL) { - WARN_ONCE("Not allocated!\n"); - return; - } - - if (rmesa->rmm->u_list[id].pending) { - WARN_ONCE("%p already pended!\n", rmesa->rmm->u_list[id].ptr); - return; - } - - rmesa->rmm->u_list[id].pending = 1; -} -#endif diff --git a/src/mesa/drivers/dri/r300/r300_mem.h b/src/mesa/drivers/dri/r300/r300_mem.h deleted file mode 100644 index 625a7f6d8d..0000000000 --- a/src/mesa/drivers/dri/r300/r300_mem.h +++ /dev/null @@ -1,37 +0,0 @@ -#ifndef __R300_MEM_H__ -#define __R300_MEM_H__ - -//#define R300_MEM_PDL 0 -#define R300_MEM_UL 1 - -#define R300_MEM_R 1 -#define R300_MEM_W 2 -#define R300_MEM_RW (R300_MEM_R | R300_MEM_W) - -#define R300_MEM_SCRATCH 2 - -struct r300_memory_manager { - struct { - void *ptr; - uint32_t size; - uint32_t age; - uint32_t h_pending; - int pending; - int mapped; - } *u_list; - int u_head, u_size, u_last; - -}; - -extern void r300_mem_init(r300ContextPtr rmesa); -extern void r300_mem_destroy(r300ContextPtr rmesa); -extern void *r300_mem_ptr(r300ContextPtr rmesa, int id); -extern int r300_mem_find(r300ContextPtr rmesa, void *ptr); -extern int r300_mem_alloc(r300ContextPtr rmesa, int alignment, int size); -extern void r300_mem_use(r300ContextPtr rmesa, int id); -extern unsigned long r300_mem_offset(r300ContextPtr rmesa, int id); -extern void *r300_mem_map(r300ContextPtr rmesa, int id, int access); -extern void r300_mem_unmap(r300ContextPtr rmesa, int id); -extern void r300_mem_free(r300ContextPtr rmesa, int id); - -#endif diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h index 8f1a6630d5..357c600af9 100644 --- a/src/mesa/drivers/dri/r300/r300_reg.h +++ b/src/mesa/drivers/dri/r300/r300_reg.h @@ -1467,6 +1467,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_3D (1 << 25) # define R300_TX_FORMAT_CUBIC_MAP (2 << 25) +# define R300_TX_FORMAT_GAMMA (1 << 21) + /* gap */ /* Floating point formats */ /* Note - hardware supports both 16 and 32 bit floating point */ @@ -1531,6 +1533,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R500_SEL_FILTER4_TC3 (3 << 18) #define R300_TX_OFFSET_0 0x4540 +#define R300_TX_OFFSET_1 0x4544 +#define R300_TX_OFFSET_2 0x4548 +#define R300_TX_OFFSET_3 0x454C +#define R300_TX_OFFSET_4 0x4550 +#define R300_TX_OFFSET_5 0x4554 +#define R300_TX_OFFSET_6 0x4558 +#define R300_TX_OFFSET_7 0x455C /* BEGIN: Guess from R200 */ # define R300_TXO_ENDIAN_NO_SWAP (0 << 0) # define R300_TXO_ENDIAN_BYTE_SWAP (1 << 0) @@ -2425,6 +2434,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* Z Buffer Clear Value */ #define R300_ZB_DEPTHCLEARVALUE 0x4f28 +#define R300_ZB_ZMASK_OFFSET 0x4f30 +#define R300_ZB_ZMASK_PITCH 0x4f34 +#define R300_ZB_ZMASK_WRINDEX 0x4f38 +#define R300_ZB_ZMASK_DWORD 0x4f3c +#define R300_ZB_ZMASK_RDINDEX 0x4f40 + /* Hierarchical Z Memory Offset */ #define R300_ZB_HIZ_OFFSET 0x4f44 @@ -3165,6 +3180,9 @@ enum { # define R300_W_SRC_RAS (1 << 2) +/* Packet0 field ordering to write all values to the same reg */ +#define RADEON_ONE_REG_WR (1 << 15) + /* Draw a primitive from vertex data in arrays loaded via 3D_LOAD_VBPNTR. * Two parameter dwords: * 0. VAP_VTX_FMT: The first parameter is not written to hardware diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c index 16ce4a1199..1356305a21 100644 --- a/src/mesa/drivers/dri/r300/r300_render.c +++ b/src/mesa/drivers/dri/r300/r300_render.c @@ -50,6 +50,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. * no bugs... */ +#include "r300_render.h" + #include "main/glheader.h" #include "main/state.h" #include "main/imports.h" @@ -66,16 +68,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "tnl/t_vp_build.h" #include "radeon_reg.h" #include "radeon_macros.h" -#include "radeon_ioctl.h" -#include "radeon_state.h" #include "r300_context.h" #include "r300_ioctl.h" #include "r300_state.h" #include "r300_reg.h" #include "r300_tex.h" #include "r300_emit.h" -#include "r300_fragprog.h" -extern int future_hw_tcl_on; +#include "r300_fragprog_common.h" +#include "r300_swtcl.h" /** * \brief Convert a OpenGL primitive type into a R300 primitive type. @@ -172,96 +172,187 @@ int r300NumVerts(r300ContextPtr rmesa, int num_verts, int prim) return num_verts - verts_off; } -static void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts) +static void r300EmitElts(GLcontext * ctx, unsigned long n_elts) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct r300_dma_region *rvb = &rmesa->state.elt_dma; void *out; + GLuint size; - if (r300IsGartMemory(rmesa, elts, n_elts * 4)) { - rvb->address = rmesa->radeon.radeonScreen->gartTextures.map; - rvb->start = ((char *)elts) - rvb->address; - rvb->aos_offset = - rmesa->radeon.radeonScreen->gart_texture_offset + - rvb->start; - return; - } else if (r300IsGartMemory(rmesa, elts, 1)) { - WARN_ONCE("Pointer not within GART memory!\n"); - _mesa_exit(-1); - } + size = ((rmesa->ind_buf.is_32bit ? 4 : 2) * n_elts + 3) & ~3; - r300AllocDmaRegion(rmesa, rvb, n_elts * 4, 4); - rvb->aos_offset = GET_START(rvb); - - out = rvb->address + rvb->start; - memcpy(out, elts, n_elts * 4); + radeonAllocDmaRegion(&rmesa->radeon, &rmesa->radeon.tcl.elt_dma_bo, + &rmesa->radeon.tcl.elt_dma_offset, size, 4); + radeon_bo_map(rmesa->radeon.tcl.elt_dma_bo, 1); + out = rmesa->radeon.tcl.elt_dma_bo->ptr + rmesa->radeon.tcl.elt_dma_offset; + memcpy(out, rmesa->ind_buf.ptr, size); + radeon_bo_unmap(rmesa->radeon.tcl.elt_dma_bo); } -static void r300FireEB(r300ContextPtr rmesa, unsigned long addr, - int vertex_count, int type) +static void r300FireEB(r300ContextPtr rmesa, int vertex_count, int type) { - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; - - start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0), 0); - e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count << 16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit); - - start_packet3(CP_PACKET3(R300_PACKET3_INDX_BUFFER, 2), 2); - e32(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | - (R300_VAP_PORT_IDX0 >> 2)); - e32(addr); - e32(vertex_count); + BATCH_LOCALS(&rmesa->radeon); + + r300_emit_scissor(rmesa->radeon.glCtx); + if (vertex_count > 0) { + int size; + + BEGIN_BATCH(10); + OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_INDX_2, 0); + if (rmesa->ind_buf.is_32bit) { + size = vertex_count; + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + ((vertex_count + 0) << 16) | type | + R300_VAP_VF_CNTL__INDEX_SIZE_32bit); + } else { + size = (vertex_count + 1) >> 1; + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | + ((vertex_count + 0) << 16) | type); + } + + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); + OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | + (R300_VAP_PORT_IDX0 >> 2)); + OUT_BATCH_RELOC(rmesa->radeon.tcl.elt_dma_offset, + rmesa->radeon.tcl.elt_dma_bo, + rmesa->radeon.tcl.elt_dma_offset, + RADEON_GEM_DOMAIN_GTT, 0, 0); + OUT_BATCH(size); + } else { + OUT_BATCH_PACKET3(R300_PACKET3_INDX_BUFFER, 2); + OUT_BATCH(R300_INDX_BUFFER_ONE_REG_WR | (0 << R300_INDX_BUFFER_SKIP_SHIFT) | + (R300_VAP_PORT_IDX0 >> 2)); + OUT_BATCH(rmesa->radeon.tcl.elt_dma_offset); + OUT_BATCH(size); + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.elt_dma_bo, + RADEON_GEM_DOMAIN_GTT, 0, 0); + } + END_BATCH(); + } } static void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset) { + BATCH_LOCALS(&rmesa->radeon); + uint32_t voffset; int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2; int i; - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; if (RADEON_DEBUG & DEBUG_VERTS) fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __FUNCTION__, nr, offset); - start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1), sz - 1); - e32(nr); - for (i = 0; i + 1 < nr; i += 2) { - e32((rmesa->state.aos[i].aos_size << 0) | - (rmesa->state.aos[i].aos_stride << 8) | - (rmesa->state.aos[i + 1].aos_size << 16) | - (rmesa->state.aos[i + 1].aos_stride << 24)); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH(sz+2+(nr * 2)); + OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1); + OUT_BATCH(nr); + + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) | + (rmesa->radeon.tcl.aos[i].stride << 8) | + (rmesa->radeon.tcl.aos[i + 1].components << 16) | + (rmesa->radeon.tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[i].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[i+1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } - e32(rmesa->state.aos[i].aos_offset + offset * 4 * rmesa->state.aos[i].aos_stride); - e32(rmesa->state.aos[i + 1].aos_offset + offset * 4 * rmesa->state.aos[i + 1].aos_stride); - } + if (nr & 1) { + OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) | + (rmesa->radeon.tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->radeon.tcl.aos[nr - 1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + END_BATCH(); + } else { + + BEGIN_BATCH(sz+2+(nr * 2)); + OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, sz - 1); + OUT_BATCH(nr); + + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->radeon.tcl.aos[i].components << 0) | + (rmesa->radeon.tcl.aos[i].stride << 8) | + (rmesa->radeon.tcl.aos[i + 1].components << 16) | + (rmesa->radeon.tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + OUT_BATCH(voffset); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + OUT_BATCH(voffset); + } - if (nr & 1) { - e32((rmesa->state.aos[nr - 1].aos_size << 0) | - (rmesa->state.aos[nr - 1].aos_stride << 8)); - e32(rmesa->state.aos[nr - 1].aos_offset + offset * 4 * rmesa->state.aos[nr - 1].aos_stride); + if (nr & 1) { + OUT_BATCH((rmesa->radeon.tcl.aos[nr - 1].components << 0) | + (rmesa->radeon.tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + OUT_BATCH(voffset); + } + for (i = 0; i + 1 < nr; i += 2) { + voffset = rmesa->radeon.tcl.aos[i + 0].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 0].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[i+0].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->radeon.tcl.aos[i + 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[i + 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[i+1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + if (nr & 1) { + voffset = rmesa->radeon.tcl.aos[nr - 1].offset + + offset * 4 * rmesa->radeon.tcl.aos[nr - 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->radeon.tcl.aos[nr-1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + END_BATCH(); } + } static void r300FireAOS(r300ContextPtr rmesa, int vertex_count, int type) { - int cmd_reserved = 0; - int cmd_written = 0; - drm_radeon_cmd_header_t *cmd = NULL; + BATCH_LOCALS(&rmesa->radeon); - start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0); - e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type); + r300_emit_scissor(rmesa->radeon.glCtx); + BEGIN_BATCH(3); + OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0); + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (vertex_count << 16) | type); + END_BATCH(); } -static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx, - int start, int end, int prim) +void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim) { + r300ContextPtr rmesa = R300_CONTEXT(ctx); + BATCH_LOCALS(&rmesa->radeon); int type, num_verts; - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *vb = &tnl->vb; type = r300PrimitiveType(rmesa, prim); num_verts = r300NumVerts(rmesa, end - start, prim); @@ -269,7 +360,13 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx, if (type < 0 || num_verts <= 0) return; - if (vb->Elts) { + /* Make space for at least 64 dwords. + * This is supposed to ensure that we can get all rendering + * commands into a single command buffer. + */ + rcommonEnsureCmdBufSpace(&rmesa->radeon, 128, __FUNCTION__); + + if (rmesa->ind_buf.ptr) { if (num_verts > 65535) { /* not implemented yet */ WARN_ONCE("Too many elts\n"); @@ -286,114 +383,142 @@ static void r300RunRenderPrimitive(r300ContextPtr rmesa, GLcontext * ctx, * allocating the index array might actually evict the vertex * arrays. *sigh* */ - r300EmitElts(ctx, vb->Elts, num_verts); - r300EmitAOS(rmesa, rmesa->state.aos_count, start); - r300FireEB(rmesa, rmesa->state.elt_dma.aos_offset, num_verts, type); + r300EmitElts(ctx, num_verts); + r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start); + if (rmesa->radeon.radeonScreen->kernel_mm) { + BEGIN_BATCH_NO_AUTOSTATE(2); + OUT_BATCH_REGSEQ(R300_VAP_VF_MAX_VTX_INDX, 1); + OUT_BATCH(rmesa->radeon.tcl.aos[0].count); + END_BATCH(); + } + r300FireEB(rmesa, num_verts, type); } else { - r300EmitAOS(rmesa, rmesa->state.aos_count, start); + r300EmitAOS(rmesa, rmesa->radeon.tcl.aos_count, start); r300FireAOS(rmesa, num_verts, type); } + COMMIT_BATCH(); } -static GLboolean r300RunRender(GLcontext * ctx, - struct tnl_pipeline_stage *stage) +static void r300RunRender(GLcontext * ctx, struct tnl_pipeline_stage *stage) { r300ContextPtr rmesa = R300_CONTEXT(ctx); int i; TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *vb = &tnl->vb; - if (RADEON_DEBUG & DEBUG_PRIMS) fprintf(stderr, "%s\n", __FUNCTION__); r300UpdateShaders(rmesa); - if (r300EmitArrays(ctx)) - return GL_TRUE; + r300EmitArrays(ctx); r300UpdateShaderStates(rmesa); r300EmitCacheFlush(rmesa); - r300EmitState(rmesa); + radeonEmitState(&rmesa->radeon); for (i = 0; i < vb->PrimitiveCount; i++) { GLuint prim = _tnl_translate_prim(&vb->Primitive[i]); GLuint start = vb->Primitive[i].start; GLuint end = vb->Primitive[i].start + vb->Primitive[i].count; - r300RunRenderPrimitive(rmesa, ctx, start, end, prim); + r300RunRenderPrimitive(ctx, start, end, prim); } r300EmitCacheFlush(rmesa); -#ifdef USER_BUFFERS - r300UseArrays(ctx); -#endif + radeonReleaseArrays(ctx, ~0); +} - r300ReleaseArrays(ctx); - return GL_FALSE; +static const char *getFallbackString(uint32_t bit) +{ + switch (bit) { + case R300_FALLBACK_VERTEX_PROGRAM : + return "vertex program"; + case R300_FALLBACK_LINE_SMOOTH: + return "smooth lines"; + case R300_FALLBACK_POINT_SMOOTH: + return "smooth points"; + case R300_FALLBACK_POLYGON_SMOOTH: + return "smooth polygons"; + case R300_FALLBACK_LINE_STIPPLE: + return "line stipple"; + case R300_FALLBACK_POLYGON_STIPPLE: + return "polygon stipple"; + case R300_FALLBACK_STENCIL_TWOSIDE: + return "two-sided stencil"; + case R300_FALLBACK_RENDER_MODE: + return "render mode != GL_RENDER"; + case R300_FALLBACK_FRAGMENT_PROGRAM: + return "fragment program"; + case R300_FALLBACK_AOS_LIMIT: + return "aos limit"; + case R300_FALLBACK_INVALID_BUFFERS: + return "invalid buffers"; + default: + return "unknown"; + } } -#define FALLBACK_IF(expr) \ - do { \ - if (expr) { \ - if (1 || RADEON_DEBUG & DEBUG_FALLBACKS) \ - WARN_ONCE("Software fallback:%s\n", \ - #expr); \ - return R300_FALLBACK_RAST; \ - } \ - } while(0) - -static int r300Fallback(GLcontext * ctx) +void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode) { - r300ContextPtr r300 = R300_CONTEXT(ctx); - const unsigned back = ctx->Stencil._BackFace; - - /* Do we need to use new-style shaders? - * Also is there a better way to do this? */ - if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - struct r500_fragment_program *fp = (struct r500_fragment_program *) - (char *)ctx->FragmentProgram._Current; - if (fp) { - if (!fp->translated) { - r500TranslateFragmentShader(r300, fp); - FALLBACK_IF(!fp->translated); + TNLcontext *tnl = TNL_CONTEXT(ctx); + r300ContextPtr rmesa = R300_CONTEXT(ctx); + uint32_t old_fallback = rmesa->fallback; + static uint32_t fallback_warn = 0; + + if (mode) { + if ((fallback_warn & bit) == 0) { + _mesa_fprintf(stderr, "WARNING! Falling back to software for %s\n", getFallbackString(bit)); + fallback_warn |= bit; + } + rmesa->fallback |= bit; + + /* update only if we change from no tcl fallbacks to some tcl fallbacks */ + if (rmesa->options.hw_tcl_enabled) { + if (((old_fallback & R300_TCL_FALLBACK_MASK) == 0) && + ((bit & R300_TCL_FALLBACK_MASK) > 0)) { + R300_STATECHANGE(rmesa, vap_cntl_status); + rmesa->hw.vap_cntl_status.cmd[1] |= R300_VAP_TCL_BYPASS; } } + + /* update only if we change from no raster fallbacks to some raster fallbacks */ + if (((old_fallback & R300_RASTER_FALLBACK_MASK) == 0) && + ((bit & R300_RASTER_FALLBACK_MASK) > 0)) { + + radeon_firevertices(&rmesa->radeon); + rmesa->radeon.swtcl.RenderIndex = ~0; + _swsetup_Wakeup( ctx ); + } } else { - struct r300_fragment_program *fp = (struct r300_fragment_program *) - (char *)ctx->FragmentProgram._Current; - if (fp) { - if (!fp->translated) { - r300TranslateFragmentShader(r300, fp); - FALLBACK_IF(!fp->translated); + rmesa->fallback &= ~bit; + + /* update only if we have disabled all tcl fallbacks */ + if (rmesa->options.hw_tcl_enabled) { + if ((old_fallback & R300_RASTER_FALLBACK_MASK) == bit) { + R300_STATECHANGE(rmesa, vap_cntl_status); + rmesa->hw.vap_cntl_status.cmd[1] &= ~R300_VAP_TCL_BYPASS; } } - } - FALLBACK_IF(ctx->RenderMode != GL_RENDER); - - /* If GL_EXT_stencil_two_side is disabled, this fallback check can - * be removed. - */ - FALLBACK_IF(ctx->Stencil.Ref[0] != ctx->Stencil.Ref[back] - || ctx->Stencil.ValueMask[0] != - ctx->Stencil.ValueMask[back] - || ctx->Stencil.WriteMask[0] != - ctx->Stencil.WriteMask[back]); - - if (ctx->Extensions.NV_point_sprite || ctx->Extensions.ARB_point_sprite) - FALLBACK_IF(ctx->Point.PointSprite); - - if (!r300->disable_lowimpact_fallback) { - FALLBACK_IF(ctx->Polygon.StippleFlag); - FALLBACK_IF(ctx->Multisample._Enabled); - FALLBACK_IF(ctx->Line.StippleFlag); - FALLBACK_IF(ctx->Line.SmoothFlag); - FALLBACK_IF(ctx->Point.SmoothFlag); + /* update only if we have disabled all raster fallbacks */ + if ((old_fallback & R300_RASTER_FALLBACK_MASK) == bit) { + _swrast_flush( ctx ); + + tnl->Driver.Render.Start = r300RenderStart; + tnl->Driver.Render.Finish = r300RenderFinish; + tnl->Driver.Render.PrimitiveNotify = r300RenderPrimitive; + tnl->Driver.Render.ResetLineStipple = r300ResetLineStipple; + tnl->Driver.Render.BuildVertices = _tnl_build_vertices; + tnl->Driver.Render.CopyPV = _tnl_copy_pv; + tnl->Driver.Render.Interp = _tnl_interp; + + _tnl_invalidate_vertex_state( ctx, ~0 ); + _tnl_invalidate_vertices( ctx, ~0 ); + } } - - return R300_FALLBACK_NONE; + } static GLboolean r300RunNonTCLRender(GLcontext * ctx, @@ -404,43 +529,15 @@ static GLboolean r300RunNonTCLRender(GLcontext * ctx, if (RADEON_DEBUG & DEBUG_PRIMS) fprintf(stderr, "%s\n", __FUNCTION__); - if (r300Fallback(ctx) >= R300_FALLBACK_RAST) - return GL_TRUE; - - if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) - return GL_TRUE; - - return r300RunRender(ctx, stage); -} - -static GLboolean r300RunTCLRender(GLcontext * ctx, - struct tnl_pipeline_stage *stage) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct r300_vertex_program *vp; - - hw_tcl_on = future_hw_tcl_on; - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (hw_tcl_on == GL_FALSE) + if (rmesa->fallback & R300_RASTER_FALLBACK_MASK) return GL_TRUE; - if (r300Fallback(ctx) >= R300_FALLBACK_TCL) { - hw_tcl_on = GL_FALSE; + if (rmesa->options.hw_tcl_enabled == GL_FALSE) return GL_TRUE; - } - - r300UpdateShaders(rmesa); - vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx); - if (vp->native == GL_FALSE) { - hw_tcl_on = GL_FALSE; - return GL_TRUE; - } + r300RunRender(ctx, stage); - return r300RunRender(ctx, stage); + return GL_FALSE; } const struct tnl_pipeline_stage _r300_render_stage = { @@ -451,12 +548,3 @@ const struct tnl_pipeline_stage _r300_render_stage = { NULL, r300RunNonTCLRender }; - -const struct tnl_pipeline_stage _r300_tcl_stage = { - "r300 Hardware Transform, Clipping and Lighting", - NULL, - NULL, - NULL, - NULL, - r300RunTCLRender -}; diff --git a/src/mesa/drivers/dri/r300/r300_render.h b/src/mesa/drivers/dri/r300/r300_render.h new file mode 100644 index 0000000000..ec785474a6 --- /dev/null +++ b/src/mesa/drivers/dri/r300/r300_render.h @@ -0,0 +1,69 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef __R300_RENDER_H__ +#define __R300_RENDER_H__ + +#include "main/mtypes.h" + +#define R300_FALLBACK_VERTEX_PROGRAM (1 << 0) +#define R300_TCL_FALLBACK_MASK 0x0000ffff + +#define R300_FALLBACK_LINE_SMOOTH (1 << 16) +#define R300_FALLBACK_POINT_SMOOTH (1 << 17) +#define R300_FALLBACK_POLYGON_SMOOTH (1 << 18) +#define R300_FALLBACK_LINE_STIPPLE (1 << 19) +#define R300_FALLBACK_POLYGON_STIPPLE (1 << 20) +#define R300_FALLBACK_STENCIL_TWOSIDE (1 << 21) +#define R300_FALLBACK_RENDER_MODE (1 << 22) +#define R300_FALLBACK_FRAGMENT_PROGRAM (1 << 23) +#define R300_FALLBACK_AOS_LIMIT (1 << 30) +#define R300_FALLBACK_INVALID_BUFFERS (1 << 31) +#define R300_RASTER_FALLBACK_MASK 0xffff0000 + +#define MASK_XYZW (R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W) +#define MASK_X R300_WRITE_ENA_X +#define MASK_Y R300_WRITE_ENA_Y +#define MASK_Z R300_WRITE_ENA_Z +#define MASK_W R300_WRITE_ENA_W + +#if SWIZZLE_X != R300_INPUT_ROUTE_SELECT_X || \ + SWIZZLE_Y != R300_INPUT_ROUTE_SELECT_Y || \ + SWIZZLE_Z != R300_INPUT_ROUTE_SELECT_Z || \ + SWIZZLE_W != R300_INPUT_ROUTE_SELECT_W || \ + SWIZZLE_ZERO != R300_INPUT_ROUTE_SELECT_ZERO || \ + SWIZZLE_ONE != R300_INPUT_ROUTE_SELECT_ONE +#error Cannot change these! +#endif + +extern const struct tnl_pipeline_stage _r300_render_stage; + +extern void r300SwitchFallback(GLcontext *ctx, uint32_t bit, GLboolean mode); + +extern void r300RunRenderPrimitive(GLcontext * ctx, int start, int end, int prim); + +#endif diff --git a/src/mesa/drivers/dri/r300/r300_shader.c b/src/mesa/drivers/dri/r300/r300_shader.c index f30fd986e0..0133b83796 100644 --- a/src/mesa/drivers/dri/r300/r300_shader.c +++ b/src/mesa/drivers/dri/r300/r300_shader.c @@ -1,18 +1,42 @@ +/* + * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> + * + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sublicense, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial + * portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + */ #include "main/glheader.h" #include "shader/program.h" #include "tnl/tnl.h" #include "r300_context.h" -#include "r300_fragprog.h" +#include "r300_fragprog_common.h" static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target, GLuint id) { - r300ContextPtr rmesa = R300_CONTEXT(ctx); struct r300_vertex_program_cont *vp; - struct r300_fragment_program *r300_fp; - struct r500_fragment_program *r500_fp; + struct r300_fragment_program *fp; switch (target) { case GL_VERTEX_STATE_PROGRAM_NV: @@ -20,28 +44,12 @@ static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target, vp = CALLOC_STRUCT(r300_vertex_program_cont); return _mesa_init_vertex_program(ctx, &vp->mesa_program, target, id); - case GL_FRAGMENT_PROGRAM_ARB: - if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - r500_fp = CALLOC_STRUCT(r500_fragment_program); - r500_fp->ctx = ctx; - return _mesa_init_fragment_program(ctx, &r500_fp->mesa_program, - target, id); - } else { - r300_fp = CALLOC_STRUCT(r300_fragment_program); - return _mesa_init_fragment_program(ctx, &r300_fp->mesa_program, - target, id); - } case GL_FRAGMENT_PROGRAM_NV: - if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - r500_fp = CALLOC_STRUCT(r500_fragment_program); - return _mesa_init_fragment_program(ctx, &r500_fp->mesa_program, - target, id); - } else { - r300_fp = CALLOC_STRUCT(r300_fragment_program); - return _mesa_init_fragment_program(ctx, &r300_fp->mesa_program, - target, id); - } + case GL_FRAGMENT_PROGRAM_ARB: + fp = CALLOC_STRUCT(r300_fragment_program); + return _mesa_init_fragment_program(ctx, &fp->Base, target, id); + default: _mesa_problem(ctx, "Bad target in r300NewProgram"); } @@ -57,20 +65,15 @@ static void r300DeleteProgram(GLcontext * ctx, struct gl_program *prog) static void r300ProgramStringNotify(GLcontext * ctx, GLenum target, struct gl_program *prog) { - r300ContextPtr rmesa = R300_CONTEXT(ctx); struct r300_vertex_program_cont *vp = (void *)prog; struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)prog; - struct r500_fragment_program *r500_fp = (struct r500_fragment_program *)prog; switch (target) { case GL_VERTEX_PROGRAM_ARB: vp->progs = NULL; break; case GL_FRAGMENT_PROGRAM_ARB: - if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) - r500_fp->translated = GL_FALSE; - else - r300_fp->translated = GL_FALSE; + r300_fp->translated = GL_FALSE; break; } @@ -81,7 +84,14 @@ r300ProgramStringNotify(GLcontext * ctx, GLenum target, struct gl_program *prog) static GLboolean r300IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog) { - return GL_TRUE; + if (target == GL_FRAGMENT_PROGRAM_ARB) { + struct r300_fragment_program *fp = (struct r300_fragment_program *)prog; + if (!fp->translated) + r300TranslateFragmentShader(ctx, &fp->Base); + + return !fp->error; + } else + return GL_TRUE; } void r300InitShaderFuncs(struct dd_function_table *functions) diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c index 07299f3b36..c0eda977db 100644 --- a/src/mesa/drivers/dri/r300/r300_state.c +++ b/src/mesa/drivers/dri/r300/r300_state.c @@ -42,6 +42,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/macros.h" #include "main/context.h" #include "main/dd.h" +#include "main/framebuffer.h" #include "main/simple_list.h" #include "main/api_arrayelt.h" #include "main/texformat.h" @@ -52,22 +53,22 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "shader/prog_statevars.h" #include "vbo/vbo.h" #include "tnl/tnl.h" +#include "tnl/t_vp_build.h" -#include "radeon_ioctl.h" -#include "radeon_state.h" #include "r300_context.h" #include "r300_ioctl.h" #include "r300_state.h" #include "r300_reg.h" #include "r300_emit.h" -#include "r300_fragprog.h" #include "r300_tex.h" +#include "r300_fragprog_common.h" +#include "r300_fragprog.h" +#include "r500_fragprog.h" +#include "r300_render.h" +#include "r300_vertprog.h" #include "drirenderbuffer.h" -extern int future_hw_tcl_on; -extern void _tnl_UpdateFixedFunctionProgram(GLcontext * ctx); - static void r300BlendColor(GLcontext * ctx, const GLfloat cf[4]) { r300ContextPtr rmesa = R300_CONTEXT(ctx); @@ -366,7 +367,7 @@ static void r300ClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq ) GLint *ip; /* no VAP UCP on non-TCL chipsets */ - if (!(rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) + if (!rmesa->options.hw_tcl_enabled) return; p = (GLint) plane - (GLint) GL_CLIP_PLANE0; @@ -385,7 +386,7 @@ static void r300SetClipPlaneState(GLcontext * ctx, GLenum cap, GLboolean state) GLuint p; /* no VAP UCP on non-TCL chipsets */ - if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) + if (!r300->options.hw_tcl_enabled) return; p = cap - GL_CLIP_PLANE0; @@ -451,24 +452,16 @@ static void r300SetPolygonOffsetState(GLcontext * ctx, GLboolean state) static GLboolean current_fragment_program_writes_depth(GLcontext* ctx) { - r300ContextPtr r300 = R300_CONTEXT(ctx); + struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; - if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) { - struct r300_fragment_program *fp = (struct r300_fragment_program *) - (char *)ctx->FragmentProgram._Current; - return (fp && fp->WritesDepth); - } else { - struct r500_fragment_program* fp = - (struct r500_fragment_program*)(char*) - ctx->FragmentProgram._Current; - return (fp && fp->writes_depth); - } + return (fp && fp->writes_depth); } static void r300SetEarlyZState(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); GLuint topZ = R300_ZTOP_ENABLE; + GLuint w_fmt, fgdepthsrc; if (ctx->Color.AlphaEnabled && ctx->Color.AlphaFunc != GL_ALWAYS) topZ = R300_ZTOP_DISABLE; @@ -485,6 +478,26 @@ static void r300SetEarlyZState(GLcontext * ctx) R300_STATECHANGE(r300, zstencil_format); r300->hw.zstencil_format.cmd[2] = topZ; } + + /* w_fmt value is set to get best performance + * see p.130 R5xx 3D acceleration guide v1.3 */ + if (current_fragment_program_writes_depth(ctx)) { + fgdepthsrc = R300_FG_DEPTH_SRC_SHADER; + w_fmt = R300_W_FMT_W24 | R300_W_SRC_US; + } else { + fgdepthsrc = R300_FG_DEPTH_SRC_SCAN; + w_fmt = R300_W_FMT_W0 | R300_W_SRC_US; + } + + if (w_fmt != r300->hw.us_out_fmt.cmd[5]) { + R300_STATECHANGE(r300, us_out_fmt); + r300->hw.us_out_fmt.cmd[5] = w_fmt; + } + + if (fgdepthsrc != r300->hw.fg_depth_src.cmd[1]) { + R300_STATECHANGE(r300, fg_depth_src); + r300->hw.fg_depth_src.cmd[1] = fgdepthsrc; + } } static void r300SetAlphaState(GLcontext * ctx) @@ -535,8 +548,6 @@ static void r300SetAlphaState(GLcontext * ctx) R300_STATECHANGE(r300, at); r300->hw.at.cmd[R300_AT_ALPHA_TEST] = pp_misc; r300->hw.at.cmd[R300_AT_UNKNOWN] = 0; - - r300SetEarlyZState(ctx); } static void r300AlphaFunc(GLcontext * ctx, GLenum func, GLfloat ref) @@ -584,15 +595,35 @@ static void r300SetDepthState(GLcontext * ctx) r300->hw.zs.cmd[R300_ZS_CNTL_1] |= translate_func(ctx->Depth.Func) << R300_Z_FUNC_SHIFT; } +} - r300SetEarlyZState(ctx); +static void r300CatchStencilFallback(GLcontext *ctx) +{ + const unsigned back = ctx->Stencil._BackFace; + + if (ctx->Stencil._Enabled && (ctx->Stencil.Ref[0] != ctx->Stencil.Ref[back] + || ctx->Stencil.ValueMask[0] != ctx->Stencil.ValueMask[back] + || ctx->Stencil.WriteMask[0] != ctx->Stencil.WriteMask[back])) { + r300SwitchFallback(ctx, R300_FALLBACK_STENCIL_TWOSIDE, GL_TRUE); + } else { + r300SwitchFallback(ctx, R300_FALLBACK_STENCIL_TWOSIDE, GL_FALSE); + } } static void r300SetStencilState(GLcontext * ctx, GLboolean state) { r300ContextPtr r300 = R300_CONTEXT(ctx); + GLboolean hw_stencil = GL_FALSE; - if (r300->state.stencil.hw_stencil) { + r300CatchStencilFallback(ctx); + + if (ctx->DrawBuffer) { + struct radeon_renderbuffer *rrbStencil + = radeon_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL); + hw_stencil = (rrbStencil && rrbStencil->bo); + } + + if (hw_stencil) { R300_STATECHANGE(r300, zs); if (state) { r300->hw.zs.cmd[R300_ZS_CNTL_0] |= @@ -601,10 +632,6 @@ static void r300SetStencilState(GLcontext * ctx, GLboolean state) r300->hw.zs.cmd[R300_ZS_CNTL_0] &= ~R300_STENCIL_ENABLE; } - } else { -#if R200_MERGED - FALLBACK(&r300->radeon, RADEON_FALLBACK_STENCIL, state); -#endif } } @@ -737,7 +764,12 @@ static void r300ColorMask(GLcontext * ctx, static void r300PointSize(GLcontext * ctx, GLfloat size) { r300ContextPtr r300 = R300_CONTEXT(ctx); - /* same size limits for AA, non-AA points */ + + /* We need to clamp to user defined range here, because + * the HW clamping happens only for per vertex point size. */ + size = CLAMP(size, ctx->Point.MinSize, ctx->Point.MaxSize); + + /* same size limits for AA, non-AA points */ size = CLAMP(size, ctx->Const.MinPointSize, ctx->Const.MaxPointSize); R300_STATECHANGE(r300, ps); @@ -830,29 +862,33 @@ static void r300ShadeModel(GLcontext * ctx, GLenum mode) R300_STATECHANGE(rmesa, shade); rmesa->hw.shade.cmd[1] = 0x00000002; + R300_STATECHANGE(rmesa, shade2); switch (mode) { case GL_FLAT: - rmesa->hw.shade.cmd[2] = R300_RE_SHADE_MODEL_FLAT; + rmesa->hw.shade2.cmd[1] = R300_RE_SHADE_MODEL_FLAT; break; case GL_SMOOTH: - rmesa->hw.shade.cmd[2] = R300_RE_SHADE_MODEL_SMOOTH; + rmesa->hw.shade2.cmd[1] = R300_RE_SHADE_MODEL_SMOOTH; break; default: return; } - rmesa->hw.shade.cmd[3] = 0x00000000; - rmesa->hw.shade.cmd[4] = 0x00000000; + rmesa->hw.shade2.cmd[2] = 0x00000000; + rmesa->hw.shade2.cmd[3] = 0x00000000; } static void r300StencilFuncSeparate(GLcontext * ctx, GLenum face, GLenum func, GLint ref, GLuint mask) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - GLuint refmask = - ((ctx->Stencil.Ref[0] & 0xff) << R300_STENCILREF_SHIFT) - | ((ctx->Stencil.ValueMask[0] & 0xff) << R300_STENCILMASK_SHIFT); - const unsigned back = ctx->Stencil._BackFace; + GLuint refmask; GLuint flag; + const unsigned back = ctx->Stencil._BackFace; + + r300CatchStencilFallback(ctx); + + refmask = ((ctx->Stencil.Ref[0] & 0xff) << R300_STENCILREF_SHIFT) + | ((ctx->Stencil.ValueMask[0] & 0xff) << R300_STENCILMASK_SHIFT); R300_STATECHANGE(rmesa, zs); rmesa->hw.zs.cmd[R300_ZS_CNTL_0] |= R300_STENCIL_FRONT_BACK; @@ -880,6 +916,8 @@ static void r300StencilMaskSeparate(GLcontext * ctx, GLenum face, GLuint mask) { r300ContextPtr rmesa = R300_CONTEXT(ctx); + r300CatchStencilFallback(ctx); + R300_STATECHANGE(rmesa, zs); rmesa->hw.zs.cmd[R300_ZS_CNTL_2] &= ~(R300_STENCILREF_MASK << @@ -896,6 +934,8 @@ static void r300StencilOpSeparate(GLcontext * ctx, GLenum face, r300ContextPtr rmesa = R300_CONTEXT(ctx); const unsigned back = ctx->Stencil._BackFace; + r300CatchStencilFallback(ctx); + R300_STATECHANGE(rmesa, zs); /* It is easier to mask what's left.. */ rmesa->hw.zs.cmd[R300_ZS_CNTL_1] &= @@ -924,28 +964,32 @@ static void r300StencilOpSeparate(GLcontext * ctx, GLenum face, * Window position and viewport transformation */ -/* - * To correctly position primitives: - */ -#define SUBPIXEL_X 0.125 -#define SUBPIXEL_Y 0.125 - static void r300UpdateWindow(GLcontext * ctx) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = rmesa->radeon.dri.drawable; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); GLfloat xoffset = dPriv ? (GLfloat) dPriv->x : 0; GLfloat yoffset = dPriv ? (GLfloat) dPriv->y + dPriv->h : 0; const GLfloat *v = ctx->Viewport._WindowMap.m; + const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF; + const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0); + GLfloat y_scale, y_bias; + + if (render_to_fbo) { + y_scale = 1.0; + y_bias = 0; + } else { + y_scale = -1.0; + y_bias = yoffset; + } GLfloat sx = v[MAT_SX]; - GLfloat tx = v[MAT_TX] + xoffset + SUBPIXEL_X; - GLfloat sy = -v[MAT_SY]; - GLfloat ty = (-v[MAT_TY]) + yoffset + SUBPIXEL_Y; - GLfloat sz = v[MAT_SZ] * rmesa->state.depth.scale; - GLfloat tz = v[MAT_TZ] * rmesa->state.depth.scale; + GLfloat tx = v[MAT_TX] + xoffset; + GLfloat sy = v[MAT_SY] * y_scale; + GLfloat ty = (v[MAT_TY] * y_scale) + y_bias; + GLfloat sz = v[MAT_SZ] * depthScale; + GLfloat tz = v[MAT_TZ] * depthScale; - R300_FIREVERTICES(rmesa); R300_STATECHANGE(rmesa, vpt); rmesa->hw.vpt.cmd[R300_VPT_XSCALE] = r300PackFloat32(sx); @@ -964,6 +1008,8 @@ static void r300Viewport(GLcontext * ctx, GLint x, GLint y, * values, or keep the originals hanging around. */ r300UpdateWindow(ctx); + + radeon_viewport(ctx, x, y, width, height); } static void r300DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval) @@ -974,13 +1020,13 @@ static void r300DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval) void r300UpdateViewportOffset(GLcontext * ctx) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = ((radeonContextPtr) rmesa)->dri.drawable; + __DRIdrawablePrivate *dPriv = radeon_get_drawable(&rmesa->radeon); GLfloat xoffset = (GLfloat) dPriv->x; GLfloat yoffset = (GLfloat) dPriv->y + dPriv->h; const GLfloat *v = ctx->Viewport._WindowMap.m; - GLfloat tx = v[MAT_TX] + xoffset + SUBPIXEL_X; - GLfloat ty = (-v[MAT_TY]) + yoffset + SUBPIXEL_Y; + GLfloat tx = v[MAT_TX] + xoffset; + GLfloat ty = (-v[MAT_TY]) + yoffset; if (rmesa->hw.vpt.cmd[R300_VPT_XOFFSET] != r300PackFloat32(tx) || rmesa->hw.vpt.cmd[R300_VPT_YOFFSET] != r300PackFloat32(ty)) { @@ -996,64 +1042,6 @@ void r300UpdateViewportOffset(GLcontext * ctx) radeonUpdateScissor(ctx); } -/** - * Tell the card where to render (offset, pitch). - * Effected by glDrawBuffer, etc - */ -void r300UpdateDrawBuffer(GLcontext * ctx) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - r300ContextPtr r300 = rmesa; - struct gl_framebuffer *fb = ctx->DrawBuffer; - driRenderbuffer *drb; - - if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) { - /* draw to front */ - drb = - (driRenderbuffer *) fb->Attachment[BUFFER_FRONT_LEFT]. - Renderbuffer; - } else if (fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) { - /* draw to back */ - drb = - (driRenderbuffer *) fb->Attachment[BUFFER_BACK_LEFT]. - Renderbuffer; - } else { - /* drawing to multiple buffers, or none */ - return; - } - - assert(drb); - assert(drb->flippedPitch); - - R300_STATECHANGE(rmesa, cb); - - r300->hw.cb.cmd[R300_CB_OFFSET] = drb->flippedOffset + //r300->radeon.state.color.drawOffset + - r300->radeon.radeonScreen->fbLocation; - r300->hw.cb.cmd[R300_CB_PITCH] = drb->flippedPitch; //r300->radeon.state.color.drawPitch; - - if (r300->radeon.radeonScreen->cpp == 4) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888; - else - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565; - - if (r300->radeon.sarea->tiling_enabled) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE; -#if 0 - R200_STATECHANGE(rmesa, ctx); - - /* Note: we used the (possibly) page-flipped values */ - rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] - = ((drb->flippedOffset + rmesa->r200Screen->fbLocation) - & R200_COLOROFFSET_MASK); - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = drb->flippedPitch; - - if (rmesa->sarea->tiling_enabled) { - rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] |= - R200_COLOR_TILE_ENABLE; - } -#endif -} - static void r300FetchStateParameter(GLcontext * ctx, const gl_state_index state[STATE_LENGTH], @@ -1064,12 +1052,14 @@ r300FetchStateParameter(GLcontext * ctx, switch (state[0]) { case STATE_INTERNAL: switch (state[1]) { - case STATE_R300_WINDOW_DIMENSION: - value[0] = r300->radeon.dri.drawable->w * 0.5f; /* width*0.5 */ - value[1] = r300->radeon.dri.drawable->h * 0.5f; /* height*0.5 */ - value[2] = 0.5F; /* for moving range [-1 1] -> [0 1] */ - value[3] = 1.0F; /* not used */ - break; + case STATE_R300_WINDOW_DIMENSION: { + __DRIdrawablePrivate * drawable = radeon_get_drawable(&r300->radeon); + value[0] = drawable->w * 0.5f; /* width*0.5 */ + value[1] = drawable->h * 0.5f; /* height*0.5 */ + value[2] = 0.5F; /* for moving range [-1 1] -> [0 1] */ + value[3] = 1.0F; /* not used */ + break; + } case STATE_R300_TEXRECT_FACTOR:{ struct gl_texture_object *t = @@ -1116,7 +1106,7 @@ void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state) if (!fp) return; - paramList = fp->mesa_program.Base.Parameters; + paramList = fp->Base.Base.Parameters; if (!paramList) return; @@ -1235,9 +1225,8 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings) { r300ContextPtr r300 = R300_CONTEXT(ctx); int i; - struct r300_fragment_program *fp = (struct r300_fragment_program *) - (char *)ctx->FragmentProgram._Current; - struct r300_fragment_program_code *code = &fp->code; + struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; + struct r300_fragment_program_code *code = &fp->code.r300; R300_STATECHANGE(r300, fpt); @@ -1271,15 +1260,15 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings) } r300->hw.fpt.cmd[R300_FPT_CMD_0] = - cmdpacket0(R300_US_TEX_INST_0, code->tex.length); + cmdpacket0(r300->radeon.radeonScreen, + R300_US_TEX_INST_0, code->tex.length); } static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings) { int i; - struct r500_fragment_program *fp = (struct r500_fragment_program *) - (char *)ctx->FragmentProgram._Current; - struct r500_fragment_program_code *code = &fp->code; + struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; + struct r500_fragment_program_code *code = &fp->code.r500; /* find all the texture instructions and relocate the texture units */ for (i = 0; i < code->inst_end + 1; i++) { @@ -1322,7 +1311,7 @@ static GLuint translate_lod_bias(GLfloat bias) static void r300SetupTextures(GLcontext * ctx) { int i, mtu; - struct r300_tex_obj *t; + struct radeon_tex_obj *t; r300ContextPtr r300 = R300_CONTEXT(ctx); int hw_tmu = 0; int last_hw_tmu = -1; /* -1 translates into no setup costs for fields */ @@ -1356,21 +1345,16 @@ static void r300SetupTextures(GLcontext * ctx) /* We cannot let disabled tmu offsets pass DRM */ for (i = 0; i < mtu; i++) { if (ctx->Texture.Unit[i]._ReallyEnabled) { - -#if 0 /* Enables old behaviour */ - hw_tmu = i; -#endif tmu_mappings[i] = hw_tmu; - t = (r300TexObjPtr) r300->state.texture.unit[i].texobj->DriverData; - /* XXX questionable fix for bug 9170: */ + t = radeon_tex_obj(ctx->Texture.Unit[i]._Current); if (!t) continue; - if ((t->format & 0xffffff00) == 0xffffff00) { + if ((t->pp_txformat & 0xffffff00) == 0xffffff00) { WARN_ONCE ("unknown texture format (entry %x) encountered. Help me !\n", - t->format & 0xff); + t->pp_txformat & 0xff); } if (RADEON_DEBUG & DEBUG_STATE) @@ -1381,29 +1365,28 @@ static void r300SetupTextures(GLcontext * ctx) r300->hw.tex.filter.cmd[R300_TEX_VALUE_0 + hw_tmu] = - gen_fixed_filter(t->filter) | (hw_tmu << 28); + gen_fixed_filter(t->pp_txfilter) | (hw_tmu << 28); /* Note: There is a LOD bias per texture unit and a LOD bias * per texture object. We add them here to get the correct behaviour. * (The per-texture object LOD bias was introduced in OpenGL 1.4 * and is not present in the EXT_texture_object extension). */ r300->hw.tex.filter_1.cmd[R300_TEX_VALUE_0 + hw_tmu] = - t->filter_1 | - translate_lod_bias(ctx->Texture.Unit[i].LodBias + t->base.tObj->LodBias); + t->pp_txfilter_1 | + translate_lod_bias(ctx->Texture.Unit[i].LodBias + t->base.LodBias); r300->hw.tex.size.cmd[R300_TEX_VALUE_0 + hw_tmu] = - t->size; + t->pp_txsize; r300->hw.tex.format.cmd[R300_TEX_VALUE_0 + - hw_tmu] = t->format; + hw_tmu] = t->pp_txformat; r300->hw.tex.pitch.cmd[R300_TEX_VALUE_0 + hw_tmu] = - t->pitch_reg; - r300->hw.tex.offset.cmd[R300_TEX_VALUE_0 + - hw_tmu] = t->offset; + t->pp_txpitch; + r300->hw.textures[hw_tmu] = t; - if (t->offset & R300_TXO_MACRO_TILE) { + if (t->tile_bits & R300_TXO_MACRO_TILE) { WARN_ONCE("macro tiling enabled!\n"); } - if (t->offset & R300_TXO_MICRO_TILE) { + if (t->tile_bits & R300_TXO_MICRO_TILE) { WARN_ONCE("micro tiling enabled!\n"); } @@ -1420,37 +1403,33 @@ static void r300SetupTextures(GLcontext * ctx) } r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FILTER0_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, last_hw_tmu + 1); r300->hw.tex.filter_1.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FILTER1_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER1_0, last_hw_tmu + 1); r300->hw.tex.size.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_SIZE_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_SIZE_0, last_hw_tmu + 1); r300->hw.tex.format.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FORMAT_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT_0, last_hw_tmu + 1); r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FORMAT2_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FORMAT2_0, last_hw_tmu + 1); r300->hw.tex.offset.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_OFFSET_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_OFFSET_0, last_hw_tmu + 1); r300->hw.tex.chroma_key.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_CHROMA_KEY_0, last_hw_tmu + 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_CHROMA_KEY_0, last_hw_tmu + 1); r300->hw.tex.border_color.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_BORDER_COLOR_0, last_hw_tmu + 1); - - if (!fp) /* should only happenen once, just after context is created */ - return; + cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, last_hw_tmu + 1); if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) { - if (fp->mesa_program.UsesKill && last_hw_tmu < 0) { + if (fp->Base.UsesKill && last_hw_tmu < 0) { // The KILL operation requires the first texture unit // to be enabled. r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1; r300->hw.tex.filter.cmd[R300_TEX_VALUE_0] = 0; r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = - cmdpacket0(R300_TX_FILTER0_0, 1); + cmdpacket0(r300->radeon.radeonScreen, R300_TX_FILTER0_0, 1); } - r300SetupFragmentShaderTextures(ctx, tmu_mappings); - } else - r500SetupFragmentShaderTextures(ctx, tmu_mappings); + } + r300->vtbl.SetupFragmentShaderTextures(ctx, tmu_mappings); if (RADEON_DEBUG & DEBUG_STATE) fprintf(stderr, "TX_ENABLE: %08x last_hw_tmu=%d\n", @@ -1469,26 +1448,21 @@ union r300_outputs_written { static void r300SetupRSUnit(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *VB = &tnl->vb; union r300_outputs_written OutputsWritten; GLuint InputsRead; int fp_reg, high_rr; int col_ip, tex_ip; int rs_tex_count = 0; - int i, count, col_fmt; + int i, col_fmt, hw_tcl_on; + + hw_tcl_on = r300->options.hw_tcl_enabled; if (hw_tcl_on) - OutputsWritten.vp_outputs = CURRENT_VERTEX_SHADER(ctx)->key.OutputsWritten; + OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten; else - RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->state.render_inputs_bitset); + RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset); - if (ctx->FragmentProgram._Current) - InputsRead = ctx->FragmentProgram._Current->Base.InputsRead; - else { - fprintf(stderr, "No ctx->FragmentProgram._Current!!\n"); - return; /* This should only ever happen once.. */ - } + InputsRead = ctx->FragmentProgram._Current->Base.InputsRead; R300_STATECHANGE(r300, ri); R300_STATECHANGE(r300, rc); @@ -1507,15 +1481,7 @@ static void r300SetupRSUnit(GLcontext * ctx) if (InputsRead & FRAG_BIT_COL0) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL0, _TNL_ATTRIB_COLOR0)) { - count = VB->AttribPtr[_TNL_ATTRIB_COLOR0]->size; - if (count == 4) - col_fmt = R300_RS_COL_FMT_RGBA; - else if (count == 3) - col_fmt = R300_RS_COL_FMT_RGB1; - else - col_fmt = R300_RS_COL_FMT_0001; - - r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(col_fmt); + r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R300_RS_INST_COL_ID(col_ip) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_COL0; ++col_ip; @@ -1527,15 +1493,7 @@ static void r300SetupRSUnit(GLcontext * ctx) if (InputsRead & FRAG_BIT_COL1) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL1, _TNL_ATTRIB_COLOR1)) { - count = VB->AttribPtr[_TNL_ATTRIB_COLOR1]->size; - if (count == 4) - col_fmt = R300_RS_COL_FMT_RGBA; - else if (count == 3) - col_fmt = R300_RS_COL_FMT_RGB1; - else - col_fmt = R300_RS_COL_FMT_0001; - - r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(col_fmt); + r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R300_RS_COL_PTR(col_ip) | R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R300_RS_INST_COL_ID(col_ip) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_COL1; ++col_ip; @@ -1545,6 +1503,7 @@ static void r300SetupRSUnit(GLcontext * ctx) } } + /* We always route 4 texcoord components */ for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { if (! ( InputsRead & FRAG_BIT_TEX(i) ) ) continue; @@ -1554,33 +1513,27 @@ static void r300SetupRSUnit(GLcontext * ctx) continue; } - int swiz; - - /* with TCL we always seem to route 4 components */ - if (hw_tcl_on) - count = 4; - else - count = VB->AttribPtr[_TNL_ATTRIB_TEX(i)]->size; - - switch(count) { - case 4: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3); break; - case 3: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(R300_RS_SEL_K1); break; - default: - case 1: - case 2: swiz = R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); break; - }; - - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= swiz | R300_RS_TEX_PTR(rs_tex_count); + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count); r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg); InputsRead &= ~(FRAG_BIT_TEX0 << i); - rs_tex_count += count; + rs_tex_count += 4; + ++tex_ip; + ++fp_reg; + } + + if (InputsRead & FRAG_BIT_WPOS) { + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count); + r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg); + InputsRead &= ~FRAG_BIT_WPOS; + rs_tex_count += 4; ++tex_ip; ++fp_reg; } if (InputsRead & FRAG_BIT_FOGC) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) { - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count); + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(R300_RS_SEL_K0) | R300_RS_SEL_R(R300_RS_SEL_K0); + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_Q(R300_RS_SEL_K1) | R300_RS_TEX_PTR(rs_tex_count); r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_FOGC; rs_tex_count += 4; @@ -1591,27 +1544,19 @@ static void r300SetupRSUnit(GLcontext * ctx) } } - if (InputsRead & FRAG_BIT_WPOS) { - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count); - r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg); - InputsRead &= ~FRAG_BIT_WPOS; - rs_tex_count += 4; - ++tex_ip; - ++fp_reg; - } - InputsRead &= ~FRAG_BIT_WPOS; - /* Setup default color if no color or tex was set */ if (rs_tex_count == 0 && col_ip == 0) { - r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(0) | R300_RS_COL_FMT(R300_RS_COL_FMT_0001); + r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_ADDR(0); + r300->hw.ri.cmd[R300_RI_INTERP_0] = R300_RS_COL_PTR(0) | R300_RS_COL_FMT(R300_RS_COL_FMT_0001); ++col_ip; } high_rr = (col_ip > tex_ip) ? col_ip : tex_ip; - r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; + r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; r300->hw.rc.cmd[2] |= high_rr - 1; - r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R300_RS_INST_0, high_rr); + r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, high_rr); + r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_IP_0, high_rr); if (InputsRead) WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead); @@ -1620,26 +1565,21 @@ static void r300SetupRSUnit(GLcontext * ctx) static void r500SetupRSUnit(GLcontext * ctx) { r300ContextPtr r300 = R300_CONTEXT(ctx); - TNLcontext *tnl = TNL_CONTEXT(ctx); - struct vertex_buffer *VB = &tnl->vb; union r300_outputs_written OutputsWritten; GLuint InputsRead; int fp_reg, high_rr; int col_ip, tex_ip; int rs_tex_count = 0; - int i, count, col_fmt; + int i, col_fmt, hw_tcl_on; + + hw_tcl_on = r300->options.hw_tcl_enabled; if (hw_tcl_on) - OutputsWritten.vp_outputs = CURRENT_VERTEX_SHADER(ctx)->key.OutputsWritten; + OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten; else - RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->state.render_inputs_bitset); + RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset); - if (ctx->FragmentProgram._Current) - InputsRead = ctx->FragmentProgram._Current->Base.InputsRead; - else { - fprintf(stderr, "No ctx->FragmentProgram._Current!!\n"); - return; /* This should only ever happen once.. */ - } + InputsRead = ctx->FragmentProgram._Current->Base.InputsRead; R300_STATECHANGE(r300, ri); R300_STATECHANGE(r300, rc); @@ -1658,15 +1598,7 @@ static void r500SetupRSUnit(GLcontext * ctx) if (InputsRead & FRAG_BIT_COL0) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL0, _TNL_ATTRIB_COLOR0)) { - count = VB->AttribPtr[_TNL_ATTRIB_COLOR0]->size; - if (count == 4) - col_fmt = R300_RS_COL_FMT_RGBA; - else if (count == 3) - col_fmt = R300_RS_COL_FMT_RGB1; - else - col_fmt = R300_RS_COL_FMT_0001; - - r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(col_fmt); + r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R500_RS_INST_COL_ID(col_ip) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_COL0; ++col_ip; @@ -1678,15 +1610,7 @@ static void r500SetupRSUnit(GLcontext * ctx) if (InputsRead & FRAG_BIT_COL1) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL1, _TNL_ATTRIB_COLOR1)) { - count = VB->AttribPtr[_TNL_ATTRIB_COLOR1]->size; - if (count == 4) - col_fmt = R300_RS_COL_FMT_RGBA; - else if (count == 3) - col_fmt = R300_RS_COL_FMT_RGB1; - else - col_fmt = R300_RS_COL_FMT_0001; - - r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(col_fmt); + r300->hw.ri.cmd[R300_RI_INTERP_0 + col_ip] = R500_RS_COL_PTR(col_ip) | R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); r300->hw.rr.cmd[R300_RR_INST_0 + col_ip] = R500_RS_INST_COL_ID(col_ip) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_COL1; ++col_ip; @@ -1696,7 +1620,7 @@ static void r500SetupRSUnit(GLcontext * ctx) } } - + /* We always route 4 texcoord components */ for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { if (! ( InputsRead & FRAG_BIT_TEX(i) ) ) continue; @@ -1706,55 +1630,37 @@ static void r500SetupRSUnit(GLcontext * ctx) continue; } - int swiz = 0; - - /* with TCL we always seem to route 4 components */ - if (hw_tcl_on) - count = 4; - else - count = VB->AttribPtr[_TNL_ATTRIB_TEX(i)]->size; - - if (count == 4) { - swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT; - swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT; - swiz |= (rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT; - swiz |= (rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT; - } else if (count == 3) { - swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT; - swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT; - swiz |= (rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT; - swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT; - } else if (count == 2) { - swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT; - swiz |= (rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT; - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT; - swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT; - } else if (count == 1) { - swiz |= (rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT; - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT; - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT; - swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT; - } else { - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT; - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT; - swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT; - swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT; - } + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) | + ((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) | + ((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) | + ((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT); - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= swiz; r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg); InputsRead &= ~(FRAG_BIT_TEX0 << i); - rs_tex_count += count; + rs_tex_count += 4; + ++tex_ip; + ++fp_reg; + } + + if (InputsRead & FRAG_BIT_WPOS) { + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) | + ((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) | + ((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) | + ((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT); + + r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg); + InputsRead &= ~FRAG_BIT_WPOS; + rs_tex_count += 4; ++tex_ip; ++fp_reg; } if (InputsRead & FRAG_BIT_FOGC) { if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) { - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) | - ((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) | - ((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) | - ((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT); + r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= (rs_tex_count << R500_RS_IP_TEX_PTR_S_SHIFT) | + (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) | + (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) | + (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT); r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg); InputsRead &= ~FRAG_BIT_FOGC; @@ -1766,87 +1672,27 @@ static void r500SetupRSUnit(GLcontext * ctx) } } - if (InputsRead & FRAG_BIT_WPOS) { - r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) | - ((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) | - ((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) | - ((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT); - - r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg); - InputsRead &= ~FRAG_BIT_WPOS; - rs_tex_count += 4; - ++tex_ip; - ++fp_reg; - } - /* Setup default color if no color or tex was set */ if (rs_tex_count == 0 && col_ip == 0) { - r300->hw.rr.cmd[R300_RR_INST_0] |= R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(0) | R500_RS_COL_FMT(R300_RS_COL_FMT_0001); + r300->hw.rr.cmd[R300_RR_INST_0] = R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_ADDR(0); + r300->hw.ri.cmd[R300_RI_INTERP_0] = R500_RS_COL_PTR(0) | R500_RS_COL_FMT(R300_RS_COL_FMT_0001); ++col_ip; } high_rr = (col_ip > tex_ip) ? col_ip : tex_ip; - r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; - r300->hw.rc.cmd[2] |= 0xC0 | (high_rr - 1); + r300->hw.rc.cmd[1] = (rs_tex_count << R300_IT_COUNT_SHIFT) | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; + r300->hw.rc.cmd[2] = 0xC0 | (high_rr - 1); - r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R500_RS_INST_0, high_rr); + r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, high_rr); + r300->hw.ri.cmd[R300_RI_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_IP_0, high_rr); if (InputsRead) WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead); } - - - -#define bump_vpu_count(ptr, new_count) do{\ - drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr));\ - int _nc=(new_count)/4; \ - assert(_nc < 256); \ - if(_nc>_p->vpu.count)_p->vpu.count=_nc;\ - }while(0) - -static INLINE void r300SetupVertexProgramFragment(r300ContextPtr r300, int dest, struct r300_vertex_shader_fragment *vsf) -{ - int i; - - if (vsf->length == 0) - return; - - if (vsf->length & 0x3) { - fprintf(stderr, "VERTEX_SHADER_FRAGMENT must have length divisible by 4\n"); - _mesa_exit(-1); - } - - switch ((dest >> 8) & 0xf) { - case 0: - R300_STATECHANGE(r300, vpi); - for (i = 0; i < vsf->length; i++) - r300->hw.vpi.cmd[R300_VPI_INSTR_0 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]); - bump_vpu_count(r300->hw.vpi.cmd, vsf->length + 4 * (dest & 0xff)); - break; - - case 2: - R300_STATECHANGE(r300, vpp); - for (i = 0; i < vsf->length; i++) - r300->hw.vpp.cmd[R300_VPP_PARAM_0 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]); - bump_vpu_count(r300->hw.vpp.cmd, vsf->length + 4 * (dest & 0xff)); - break; - case 4: - R300_STATECHANGE(r300, vps); - for (i = 0; i < vsf->length; i++) - r300->hw.vps.cmd[1 + i + 4 * (dest & 0xff)] = (vsf->body.d[i]); - bump_vpu_count(r300->hw.vps.cmd, vsf->length + 4 * (dest & 0xff)); - break; - default: - fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest); - _mesa_exit(-1); - } -} - #define MIN3(a, b, c) ((a) < (b) ? MIN2(a, c) : MIN2(b, c)) - -static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, +void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, GLuint output_count, GLuint temp_count) { int vtx_mem_size; @@ -1870,7 +1716,7 @@ static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, pvs_num_cntrls = MIN2(6, vtx_mem_size/temp_count); R300_STATECHANGE(rmesa, vap_cntl); - if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) { + if (rmesa->options.hw_tcl_enabled) { rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] = (pvs_num_slots << R300_PVS_NUM_SLOTS_SHIFT) | (pvs_num_cntrls << R300_PVS_NUM_CNTLRS_SHIFT) | @@ -1900,114 +1746,6 @@ static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, } -static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa) -{ - struct r300_vertex_shader_state *prog = &(rmesa->state.vertex_shader); - GLuint o_reg = 0; - GLuint i_reg = 0; - int i; - int inst_count = 0; - int param_count = 0; - int program_end = 0; - - for (i = VERT_ATTRIB_POS; i < VERT_ATTRIB_MAX; i++) { - if (rmesa->state.sw_tcl_inputs[i] != -1) { - prog->program.body.i[program_end + 0] = PVS_OP_DST_OPERAND(VE_MULTIPLY, GL_FALSE, GL_FALSE, o_reg++, VSF_FLAG_ALL, PVS_DST_REG_OUT); - prog->program.body.i[program_end + 1] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_X, PVS_SRC_SELECT_Y, PVS_SRC_SELECT_Z, PVS_SRC_SELECT_W, PVS_SRC_REG_INPUT, VSF_FLAG_NONE); - prog->program.body.i[program_end + 2] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE); - prog->program.body.i[program_end + 3] = PVS_SRC_OPERAND(rmesa->state.sw_tcl_inputs[i], PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_SELECT_FORCE_1, PVS_SRC_REG_INPUT, VSF_FLAG_NONE); - program_end += 4; - i_reg++; - } - } - - prog->program.length = program_end; - - r300SetupVertexProgramFragment(rmesa, R300_PVS_CODE_START, - &(prog->program)); - inst_count = (prog->program.length / 4) - 1; - - r300VapCntl(rmesa, i_reg, o_reg, 0); - - R300_STATECHANGE(rmesa, pvs); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = - (0 << R300_PVS_FIRST_INST_SHIFT) | - (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) | - (inst_count << R300_PVS_LAST_INST_SHIFT); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] = - (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | - (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] = - (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT); -} - -static int bit_count (int x) -{ - x = ((x & 0xaaaaaaaaU) >> 1) + (x & 0x55555555U); - x = ((x & 0xccccccccU) >> 2) + (x & 0x33333333U); - x = (x >> 16) + (x & 0xffff); - x = ((x & 0xf0f0) >> 4) + (x & 0x0f0f); - return (x >> 8) + (x & 0x00ff); -} - -static void r300SetupRealVertexProgram(r300ContextPtr rmesa) -{ - GLcontext *ctx = rmesa->radeon.glCtx; - struct r300_vertex_program *prog = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx); - int inst_count = 0; - int param_count = 0; - - /* FIXME: r300SetupVertexProgramFragment */ - R300_STATECHANGE(rmesa, vpp); - param_count = - r300VertexProgUpdateParams(ctx, - (struct r300_vertex_program_cont *) - ctx->VertexProgram._Current, - (float *)&rmesa->hw.vpp. - cmd[R300_VPP_PARAM_0]); - bump_vpu_count(rmesa->hw.vpp.cmd, param_count); - param_count /= 4; - - r300SetupVertexProgramFragment(rmesa, R300_PVS_CODE_START, &(prog->program)); - inst_count = (prog->program.length / 4) - 1; - - r300VapCntl(rmesa, bit_count(prog->key.InputsRead), - bit_count(prog->key.OutputsWritten), prog->num_temporaries); - - R300_STATECHANGE(rmesa, pvs); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = - (0 << R300_PVS_FIRST_INST_SHIFT) | - (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) | - (inst_count << R300_PVS_LAST_INST_SHIFT); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] = - (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | - (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT); - rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] = - (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT); -} - -static void r300SetupVertexProgram(r300ContextPtr rmesa) -{ - GLcontext *ctx = rmesa->radeon.glCtx; - - /* Reset state, in case we don't use something */ - ((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0; - ((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0; - ((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0; - - /* Not sure why this doesnt work... - 0x400 area might have something to do with pixel shaders as it appears right after pfs programming. - 0x406 is set to { 0.0, 0.0, 1.0, 0.0 } most of the time but should change with smooth points and in other rare cases. */ - //setup_vertex_shader_fragment(rmesa, 0x406, &unk4); - if (hw_tcl_on && ((struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx))->translated) { - r300SetupRealVertexProgram(rmesa); - } else { - /* FIXME: This needs to be replaced by vertex shader generation code. */ - r300SetupDefaultVertexProgram(rmesa); - } - -} - /** * Enable/Disable states. * @@ -2015,20 +1753,13 @@ static void r300SetupVertexProgram(r300ContextPtr rmesa) */ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state) { + r300ContextPtr rmesa = R300_CONTEXT(ctx); if (RADEON_DEBUG & DEBUG_STATE) fprintf(stderr, "%s( %s = %s )\n", __FUNCTION__, _mesa_lookup_enum_by_nr(cap), state ? "GL_TRUE" : "GL_FALSE"); switch (cap) { - case GL_TEXTURE_1D: - case GL_TEXTURE_2D: - case GL_TEXTURE_3D: - /* empty */ - break; - case GL_FOG: - /* empty */ - break; case GL_ALPHA_TEST: r300SetAlphaState(ctx); break; @@ -2046,22 +1777,46 @@ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state) case GL_CLIP_PLANE5: r300SetClipPlaneState(ctx, cap, state); break; + case GL_CULL_FACE: + r300UpdateCulling(ctx); + break; case GL_DEPTH_TEST: r300SetDepthState(ctx); break; - case GL_STENCIL_TEST: - r300SetStencilState(ctx, state); + case GL_LINE_SMOOTH: + if (rmesa->options.conformance_mode) + r300SwitchFallback(ctx, R300_FALLBACK_LINE_SMOOTH, ctx->Line.SmoothFlag); break; - case GL_CULL_FACE: - r300UpdateCulling(ctx); + case GL_LINE_STIPPLE: + if (rmesa->options.conformance_mode) + r300SwitchFallback(ctx, R300_FALLBACK_LINE_STIPPLE, ctx->Line.StippleFlag); + break; + case GL_POINT_SMOOTH: + if (rmesa->options.conformance_mode) + r300SwitchFallback(ctx, R300_FALLBACK_POINT_SMOOTH, ctx->Point.SmoothFlag); + break; + case GL_POLYGON_SMOOTH: + if (rmesa->options.conformance_mode) + r300SwitchFallback(ctx, R300_FALLBACK_POLYGON_SMOOTH, ctx->Polygon.SmoothFlag); + break; + case GL_POLYGON_STIPPLE: + if (rmesa->options.conformance_mode) + r300SwitchFallback(ctx, R300_FALLBACK_POLYGON_STIPPLE, ctx->Polygon.StippleFlag); break; case GL_POLYGON_OFFSET_POINT: case GL_POLYGON_OFFSET_LINE: case GL_POLYGON_OFFSET_FILL: r300SetPolygonOffsetState(ctx, state); break; + case GL_SCISSOR_TEST: + radeon_firevertices(&rmesa->radeon); + rmesa->radeon.state.scissor.enabled = state; + radeonUpdateScissor( ctx ); + break; + case GL_STENCIL_TEST: + r300SetStencilState(ctx, state); + break; default: - radeonEnable(ctx, cap, state); break; } } @@ -2072,15 +1827,14 @@ static void r300Enable(GLcontext * ctx, GLenum cap, GLboolean state) static void r300ResetHwState(r300ContextPtr r300) { GLcontext *ctx = r300->radeon.glCtx; - int has_tcl = 1; + int has_tcl; - if (!(r300->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) - has_tcl = 0; + has_tcl = r300->options.hw_tcl_enabled; if (RADEON_DEBUG & DEBUG_STATE) fprintf(stderr, "%s\n", __FUNCTION__); - r300UpdateWindow(ctx); + radeon_firevertices(&r300->radeon); r300ColorMask(ctx, ctx->Color.ColorMask[RCOMP], @@ -2102,8 +1856,6 @@ static void r300ResetHwState(r300ContextPtr r300) r300UpdateCulling(ctx); - r300UpdateTextureState(ctx); - r300SetBlendState(ctx); r300SetLogicOpState(ctx); @@ -2182,8 +1934,8 @@ static void r300ResetHwState(r300ContextPtr r300) } /* XXX: Enable anti-aliasing? */ - r300->hw.gb_misc.cmd[R300_GB_MISC_AA_CONFIG] = GB_AA_CONFIG_AA_DISABLE; - r300->hw.gb_misc.cmd[R300_GB_MISC_SELECT] = 0; + r300->hw.gb_misc2.cmd[R300_GB_MISC2_AA_CONFIG] = GB_AA_CONFIG_AA_DISABLE; + r300->hw.gb_misc2.cmd[R300_GB_MISC2_SELECT] = 0; r300->hw.ga_point_s0.cmd[1] = r300PackFloat32(0.0); r300->hw.ga_point_s0.cmd[2] = r300PackFloat32(0.0); @@ -2242,20 +1994,6 @@ static void r300ResetHwState(r300ContextPtr r300) r300BlendColor(ctx, ctx->Color.BlendColor); - /* Again, r300ClearBuffer uses this */ - r300->hw.cb.cmd[R300_CB_OFFSET] = - r300->radeon.state.color.drawOffset + - r300->radeon.radeonScreen->fbLocation; - r300->hw.cb.cmd[R300_CB_PITCH] = r300->radeon.state.color.drawPitch; - - if (r300->radeon.radeonScreen->cpp == 4) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888; - else - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565; - - if (r300->radeon.sarea->tiling_enabled) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE; - r300->hw.rb3d_dither_ctl.cmd[1] = 0; r300->hw.rb3d_dither_ctl.cmd[2] = 0; r300->hw.rb3d_dither_ctl.cmd[3] = 0; @@ -2268,44 +2006,18 @@ static void r300ResetHwState(r300ContextPtr r300) r300->hw.rb3d_aaresolve_ctl.cmd[1] = 0; - r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[1] = 0x00000000; - r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[2] = 0xffffffff; - - r300->hw.zb.cmd[R300_ZB_OFFSET] = - r300->radeon.radeonScreen->depthOffset + - r300->radeon.radeonScreen->fbLocation; - r300->hw.zb.cmd[R300_ZB_PITCH] = r300->radeon.radeonScreen->depthPitch; - - if (r300->radeon.sarea->tiling_enabled) { - /* XXX: Turn off when clearing buffers ? */ - r300->hw.zb.cmd[R300_ZB_PITCH] |= R300_DEPTHMACROTILE_ENABLE; - - if (ctx->Visual.depthBits == 24) - r300->hw.zb.cmd[R300_ZB_PITCH] |= - R300_DEPTHMICROTILE_TILED; - } + r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[1] = 0x00000000; + r300->hw.rb3d_discard_src_pixel_lte_threshold.cmd[2] = 0xffffffff; r300->hw.zb_depthclearvalue.cmd[1] = 0; - switch (ctx->Visual.depthBits) { - case 16: - r300->hw.zstencil_format.cmd[1] = R300_DEPTHFORMAT_16BIT_INT_Z; - break; - case 24: - r300->hw.zstencil_format.cmd[1] = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; - break; - default: - fprintf(stderr, "Error: Unsupported depth %d... exiting\n", ctx->Visual.depthBits); - _mesa_exit(-1); - } - r300->hw.zstencil_format.cmd[2] = R300_ZTOP_DISABLE; r300->hw.zstencil_format.cmd[3] = 0x00000003; r300->hw.zstencil_format.cmd[4] = 0x00000000; r300SetEarlyZState(ctx); - r300->hw.unk4F30.cmd[1] = 0; - r300->hw.unk4F30.cmd[2] = 0; + r300->hw.zb_zmask.cmd[1] = 0; + r300->hw.zb_zmask.cmd[2] = 0; r300->hw.zb_hiz_offset.cmd[1] = 0; @@ -2319,20 +2031,26 @@ static void r300ResetHwState(r300ContextPtr r300) r300->hw.vps.cmd[R300_VPS_ZERO_3] = 0; } - r300->hw.all_dirty = GL_TRUE; + r300->radeon.hw.all_dirty = GL_TRUE; } void r300UpdateShaders(r300ContextPtr rmesa) { GLcontext *ctx; - struct r300_vertex_program *vp; + struct r300_fragment_program *fp; int i; ctx = rmesa->radeon.glCtx; + fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; - if (rmesa->NewGLState && hw_tcl_on) { - rmesa->NewGLState = 0; + /* should only happenen once, just after context is created */ + /* TODO: shouldn't we fallback to sw here? */ + if (!fp) { + _mesa_fprintf(stderr, "No ctx->FragmentProgram._Current!!\n"); + return; + } + if (rmesa->radeon.NewGLState && rmesa->options.hw_tcl_enabled) { for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) { rmesa->temp_attrib[i] = TNL_CONTEXT(ctx)->vb.AttribPtr[i]; @@ -2348,21 +2066,16 @@ void r300UpdateShaders(r300ContextPtr rmesa) } r300SelectVertexShader(rmesa); - vp = (struct r300_vertex_program *) - CURRENT_VERTEX_SHADER(ctx); - /*if (vp->translated == GL_FALSE) - r300TranslateVertexShader(vp); */ - if (vp->translated == GL_FALSE) { - fprintf(stderr, "Failing back to sw-tcl\n"); - hw_tcl_on = future_hw_tcl_on = 0; - r300ResetHwState(rmesa); - - r300UpdateStateParameters(ctx, _NEW_PROGRAM | - _NEW_PROGRAM_CONSTANTS); - return; - } + r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, rmesa->selected_vp->error); } + + if (!fp->translated || rmesa->radeon.NewGLState) + r300TranslateFragmentShader(ctx, ctx->FragmentProgram._Current); + + r300SwitchFallback(ctx, R300_FALLBACK_FRAGMENT_PROGRAM, fp->error); + r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS); + rmesa->radeon.NewGLState = 0; } static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, @@ -2386,35 +2099,23 @@ static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, } -static void r300SetupPixelShader(r300ContextPtr rmesa) +static void r300SetupPixelShader(GLcontext *ctx) { - GLcontext *ctx = rmesa->radeon.glCtx; - struct r300_fragment_program *fp = (struct r300_fragment_program *) - (char *)ctx->FragmentProgram._Current; + r300ContextPtr rmesa = R300_CONTEXT(ctx); + struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; struct r300_fragment_program_code *code; int i, k; - if (!fp) /* should only happenen once, just after context is created */ - return; - - r300TranslateFragmentShader(rmesa, fp); - if (!fp->translated) { - fprintf(stderr, "%s: No valid fragment shader, exiting\n", - __FUNCTION__); - return; - } - code = &fp->code; - - r300SetupTextures(ctx); + code = &fp->code.r300; R300_STATECHANGE(rmesa, fpi[0]); R300_STATECHANGE(rmesa, fpi[1]); R300_STATECHANGE(rmesa, fpi[2]); R300_STATECHANGE(rmesa, fpi[3]); - rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, code->alu.length); - rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, code->alu.length); - rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, code->alu.length); - rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); + rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_RGB_INST_0, code->alu.length); + rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_RGB_ADDR_0, code->alu.length); + rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_INST_0, code->alu.length); + rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_US_ALU_ALPHA_ADDR_0, code->alu.length); for (i = 0; i < code->alu.length; i++) { rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst0; rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst1; @@ -2445,10 +2146,10 @@ static void r300SetupPixelShader(r300ContextPtr rmesa) } R300_STATECHANGE(rmesa, fpp); - rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(R300_PFS_PARAM_0_X, code->const_nr * 4); + rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_PFS_PARAM_0_X, code->const_nr * 4); for (i = 0; i < code->const_nr; i++) { const GLfloat *constant = get_fragmentprogram_constant(ctx, - &fp->mesa_program.Base, code->constant[i]); + &fp->Base.Base, code->constant[i]); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat24(constant[0]); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat24(constant[1]); rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat24(constant[2]); @@ -2470,29 +2171,17 @@ static void r300SetupPixelShader(r300ContextPtr rmesa) if(_nc>_p->r500fp.count)_p->r500fp.count=_nc;\ } while(0) -static void r500SetupPixelShader(r300ContextPtr rmesa) +static void r500SetupPixelShader(GLcontext *ctx) { - GLcontext *ctx = rmesa->radeon.glCtx; - struct r500_fragment_program *fp = (struct r500_fragment_program *) - (char *)ctx->FragmentProgram._Current; + r300ContextPtr rmesa = R300_CONTEXT(ctx); + struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; int i; struct r500_fragment_program_code *code; - if (!fp) /* should only happenen once, just after context is created */ - return; - ((drm_r300_cmd_header_t *) rmesa->hw.r500fp.cmd)->r500fp.count = 0; ((drm_r300_cmd_header_t *) rmesa->hw.r500fp_const.cmd)->r500fp.count = 0; - r500TranslateFragmentShader(rmesa, fp); - if (!fp->translated) { - fprintf(stderr, "%s: No valid fragment shader, exiting\n", - __FUNCTION__); - return; - } - code = &fp->code; - - r300SetupTextures(ctx); + code = &fp->code.r500; R300_STATECHANGE(rmesa, fp); rmesa->hw.fp.cmd[R500_FP_PIXSIZE] = code->max_temp_idx; @@ -2522,58 +2211,96 @@ static void r500SetupPixelShader(r300ContextPtr rmesa) R300_STATECHANGE(rmesa, r500fp_const); for (i = 0; i < code->const_nr; i++) { const GLfloat *constant = get_fragmentprogram_constant(ctx, - &fp->mesa_program.Base, code->constant[i]); + &fp->Base.Base, code->constant[i]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat32(constant[0]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat32(constant[1]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat32(constant[2]); rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 3] = r300PackFloat32(constant[3]); } bump_r500fp_const_count(rmesa->hw.r500fp_const.cmd, code->const_nr * 4); +} + +void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten) +{ + r300ContextPtr rmesa = R300_CONTEXT( ctx ); + struct vertex_attribute *attrs = rmesa->vbuf.attribs; + int i, j, reg_count; + uint32_t *vir0 = &rmesa->hw.vir[0].cmd[1]; + uint32_t *vir1 = &rmesa->hw.vir[1].cmd[1]; + + for (i = 0; i < R300_VIR_CMDSIZE-1; ++i) + vir0[i] = vir1[i] = 0; + + for (i = 0, j = 0; i < rmesa->vbuf.num_attribs; ++i) { + int tmp; + + tmp = attrs[i].data_type | (attrs[i].dst_loc << R300_DST_VEC_LOC_SHIFT); + if (attrs[i]._signed) + tmp |= R300_SIGNED; + if (attrs[i].normalize) + tmp |= R300_NORMALIZE; + + if (i % 2 == 0) { + vir0[j] = tmp << R300_DATA_TYPE_0_SHIFT; + vir1[j] = attrs[i].swizzle | (attrs[i].write_mask << R300_WRITE_ENA_SHIFT); + } else { + vir0[j] |= tmp << R300_DATA_TYPE_1_SHIFT; + vir1[j] |= (attrs[i].swizzle | (attrs[i].write_mask << R300_WRITE_ENA_SHIFT)) << R300_SWIZZLE1_SHIFT; + ++j; + } + } + + reg_count = (rmesa->vbuf.num_attribs + 1) >> 1; + if (rmesa->vbuf.num_attribs % 2 != 0) { + vir0[reg_count-1] |= R300_LAST_VEC << R300_DATA_TYPE_0_SHIFT; + } else { + vir0[reg_count-1] |= R300_LAST_VEC << R300_DATA_TYPE_1_SHIFT; + } + + R300_STATECHANGE(rmesa, vir[0]); + R300_STATECHANGE(rmesa, vir[1]); + R300_STATECHANGE(rmesa, vof); + R300_STATECHANGE(rmesa, vic); + + if (rmesa->radeon.radeonScreen->kernel_mm) { + rmesa->hw.vir[0].cmd[0] &= 0xC000FFFF; + rmesa->hw.vir[1].cmd[0] &= 0xC000FFFF; + rmesa->hw.vir[0].cmd[0] |= (reg_count & 0x3FFF) << 16; + rmesa->hw.vir[1].cmd[0] |= (reg_count & 0x3FFF) << 16; + } else { + ((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count = reg_count; + ((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count = reg_count; + } + rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead); + rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead); + rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead); + rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead); } void r300UpdateShaderStates(r300ContextPtr rmesa) { GLcontext *ctx; ctx = rmesa->radeon.glCtx; + struct r300_fragment_program *r300_fp; - r300UpdateTextureState(ctx); - r300SetEarlyZState(ctx); + r300_fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current; - /* w_fmt value is set to get best performance - * see p.130 R5xx 3D acceleration guide v1.3 */ - GLuint w_fmt, fgdepthsrc; - if (current_fragment_program_writes_depth(ctx)) { - fgdepthsrc = R300_FG_DEPTH_SRC_SHADER; - w_fmt = R300_W_FMT_W24 | R300_W_SRC_US; - } else { - fgdepthsrc = R300_FG_DEPTH_SRC_SCAN; - w_fmt = R300_W_FMT_W0 | R300_W_SRC_US; - } + /* should only happenen once, just after context is created */ + if (!r300_fp) + return; - if (w_fmt != rmesa->hw.us_out_fmt.cmd[5]) { - R300_STATECHANGE(rmesa, us_out_fmt); - rmesa->hw.us_out_fmt.cmd[5] = w_fmt; - } + r300SetEarlyZState(ctx); - if (fgdepthsrc != rmesa->hw.fg_depth_src.cmd[1]) { - R300_STATECHANGE(rmesa, fg_depth_src); - rmesa->hw.fg_depth_src.cmd[1] = fgdepthsrc; - } + r300SetupTextures(ctx); - if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) - r500SetupPixelShader(rmesa); - else - r300SetupPixelShader(rmesa); + rmesa->vtbl.SetupPixelShader(ctx); - if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) - r500SetupRSUnit(ctx); - else - r300SetupRSUnit(ctx); + rmesa->vtbl.SetupRSUnit(ctx); - if ((rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL)) + if (rmesa->options.hw_tcl_enabled) { r300SetupVertexProgram(rmesa); - + } } /** @@ -2587,15 +2314,18 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state) _swsetup_InvalidateState(ctx, new_state); _vbo_InvalidateState(ctx, new_state); _tnl_InvalidateState(ctx, new_state); - _ae_invalidate_state(ctx, new_state); if (new_state & _NEW_BUFFERS) { - r300UpdateDrawBuffer(ctx); + _mesa_update_framebuffer(ctx); + /* this updates the DrawBuffer's Width/Height if it's a FBO */ + _mesa_update_draw_buffer_bounds(ctx); + + R300_STATECHANGE(r300, cb); } r300UpdateStateParameters(ctx, new_state); - r300->NewGLState |= new_state; + r300->radeon.NewGLState |= new_state; } /** @@ -2605,58 +2335,12 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state) */ void r300InitState(r300ContextPtr r300) { - GLcontext *ctx = r300->radeon.glCtx; - GLuint depth_fmt; - - radeonInitState(&r300->radeon); - - switch (ctx->Visual.depthBits) { - case 16: - r300->state.depth.scale = 1.0 / (GLfloat) 0xffff; - depth_fmt = R300_DEPTHFORMAT_16BIT_INT_Z; - break; - case 24: - r300->state.depth.scale = 1.0 / (GLfloat) 0xffffff; - depth_fmt = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; - break; - default: - fprintf(stderr, "Error: Unsupported depth %d... exiting\n", - ctx->Visual.depthBits); - _mesa_exit(-1); - } - - /* Only have hw stencil when depth buffer is 24 bits deep */ - r300->state.stencil.hw_stencil = (ctx->Visual.stencilBits > 0 && - ctx->Visual.depthBits == 24); - - memset(&(r300->state.texture), 0, sizeof(r300->state.texture)); - r300ResetHwState(r300); } static void r300RenderMode(GLcontext * ctx, GLenum mode) { - r300ContextPtr rmesa = R300_CONTEXT(ctx); - (void)rmesa; - (void)mode; -} - -void r300UpdateClipPlanes( GLcontext *ctx ) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - GLuint p; - - for (p = 0; p < ctx->Const.MaxClipPlanes; p++) { - if (ctx->Transform.ClipPlanesEnabled & (1 << p)) { - GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p]; - - R300_STATECHANGE( rmesa, vpucp[p] ); - rmesa->hw.vpucp[p].cmd[R300_VPUCP_X] = ip[0]; - rmesa->hw.vpucp[p].cmd[R300_VPUCP_Y] = ip[1]; - rmesa->hw.vpucp[p].cmd[R300_VPUCP_Z] = ip[2]; - rmesa->hw.vpucp[p].cmd[R300_VPUCP_W] = ip[3]; - } - } + r300SwitchFallback(ctx, R300_FALLBACK_RENDER_MODE, ctx->RenderMode != GL_RENDER); } /** @@ -2664,7 +2348,6 @@ void r300UpdateClipPlanes( GLcontext *ctx ) */ void r300InitStateFuncs(struct dd_function_table *functions) { - radeonInitStateFuncs(functions); functions->UpdateState = r300InvalidateState; functions->AlphaFunc = r300AlphaFunc; @@ -2700,4 +2383,25 @@ void r300InitStateFuncs(struct dd_function_table *functions) functions->RenderMode = r300RenderMode; functions->ClipPlane = r300ClipPlane; + functions->Scissor = radeonScissor; + + functions->DrawBuffer = radeonDrawBuffer; + functions->ReadBuffer = radeonReadBuffer; +} + +void r300InitShaderFunctions(r300ContextPtr r300) +{ + if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + r300->vtbl.SetupRSUnit = r500SetupRSUnit; + r300->vtbl.SetupPixelShader = r500SetupPixelShader; + r300->vtbl.SetupFragmentShaderTextures = r500SetupFragmentShaderTextures; + r300->vtbl.BuildFragmentProgramHwCode = r500BuildFragmentProgramHwCode; + r300->vtbl.FragmentProgramDump = r500FragmentProgramDump; + } else { + r300->vtbl.SetupRSUnit = r300SetupRSUnit; + r300->vtbl.SetupPixelShader = r300SetupPixelShader; + r300->vtbl.SetupFragmentShaderTextures = r300SetupFragmentShaderTextures; + r300->vtbl.BuildFragmentProgramHwCode = r300BuildFragmentProgramHwCode; + r300->vtbl.FragmentProgramDump = r300FragmentProgramDump; + } } diff --git a/src/mesa/drivers/dri/r300/r300_state.h b/src/mesa/drivers/dri/r300/r300_state.h index 0589ab7cad..2328289420 100644 --- a/src/mesa/drivers/dri/r300/r300_state.h +++ b/src/mesa/drivers/dri/r300/r300_state.h @@ -39,42 +39,25 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define R300_NEWPRIM( rmesa ) \ do { \ - if ( rmesa->dma.flush ) \ - rmesa->dma.flush( rmesa ); \ + if ( rmesa->radeon.dma.flush ) \ + rmesa->radeon.dma.flush( rmesa->radeon.glCtx ); \ } while (0) #define R300_STATECHANGE(r300, atom) \ do { \ R300_NEWPRIM(r300); \ r300->hw.atom.dirty = GL_TRUE; \ - r300->hw.is_dirty = GL_TRUE; \ + r300->radeon.hw.is_dirty = GL_TRUE; \ } while(0) -#define R300_PRINT_STATE(r300, atom) \ - r300PrintStateAtom(r300, &r300->hw.atom) - -/* Fire the buffered vertices no matter what. - TODO: This has not been implemented yet - */ -#define R300_FIREVERTICES( r300 ) \ -do { \ - \ - if ( (r300)->cmdbuf.count_used || (r300)->dma.flush ) { \ - r300Flush( (r300)->radeon.glCtx ); \ - } \ - \ -} while (0) - -// r300_state.c -extern int future_hw_tcl_on; -void _tnl_UpdateFixedFunctionProgram (GLcontext * ctx); void r300UpdateViewportOffset (GLcontext * ctx); void r300UpdateDrawBuffer (GLcontext * ctx); void r300UpdateStateParameters (GLcontext * ctx, GLuint new_state); void r300UpdateShaders (r300ContextPtr rmesa); void r300UpdateShaderStates (r300ContextPtr rmesa); void r300InitState (r300ContextPtr r300); -void r300UpdateClipPlanes (GLcontext * ctx); void r300InitStateFuncs (struct dd_function_table *functions); +void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, GLuint output_count, GLuint temp_count); +void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten); #endif /* __R300_STATE_H__ */ diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c index ba3621b16b..ce4179208e 100644 --- a/src/mesa/drivers/dri/r300/r300_swtcl.c +++ b/src/mesa/drivers/dri/r300/r300_swtcl.c @@ -28,362 +28,237 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* * Authors: * Dave Airlie <airlied@linux.ie> + * Maciej Cencora <m.cencora@gmail.com> */ -/* derived from r200 swtcl path */ - - - -#include "main/glheader.h" -#include "main/mtypes.h" -#include "main/colormac.h" -#include "main/enums.h" -#include "main/image.h" -#include "main/imports.h" -#include "main/light.h" -#include "main/macros.h" - -#include "swrast/s_context.h" -#include "swrast/s_fog.h" -#include "swrast_setup/swrast_setup.h" -#include "math/m_translate.h" #include "tnl/tnl.h" -#include "tnl/t_context.h" #include "tnl/t_pipeline.h" -#include "r300_context.h" -#include "r300_swtcl.h" #include "r300_state.h" -#include "r300_ioctl.h" +#include "r300_swtcl.h" #include "r300_emit.h" -#include "r300_mem.h" - -static void flush_last_swtcl_prim( r300ContextPtr rmesa ); +#include "r300_tex.h" +#include "r300_render.h" - -void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset); -void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr); #define EMIT_ATTR( ATTR, STYLE ) \ do { \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = (ATTR); \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = (STYLE); \ - rmesa->swtcl.vertex_attr_count++; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = (ATTR); \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = (STYLE); \ + rmesa->radeon.swtcl.vertex_attr_count++; \ } while (0) #define EMIT_PAD( N ) \ do { \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].attrib = 0; \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].format = EMIT_PAD; \ - rmesa->swtcl.vertex_attrs[rmesa->swtcl.vertex_attr_count].offset = (N); \ - rmesa->swtcl.vertex_attr_count++; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].attrib = 0; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].format = EMIT_PAD; \ + rmesa->radeon.swtcl.vertex_attrs[rmesa->radeon.swtcl.vertex_attr_count].offset = (N); \ + rmesa->radeon.swtcl.vertex_attr_count++; \ +} while (0) + +#define ADD_ATTR(_attr, _format, _dst_loc, _swizzle, _write_mask, _normalize) \ +do { \ + attrs[num_attrs].element = (_attr); \ + attrs[num_attrs].data_type = (_format); \ + attrs[num_attrs].dst_loc = (_dst_loc); \ + attrs[num_attrs].swizzle = (_swizzle); \ + attrs[num_attrs].write_mask = (_write_mask); \ + attrs[num_attrs]._signed = 0; \ + attrs[num_attrs].normalize = (_normalize); \ + ++num_attrs; \ } while (0) -static void r300SetVertexFormat( GLcontext *ctx ) +void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead, GLuint *_OutputsWritten) { r300ContextPtr rmesa = R300_CONTEXT( ctx ); TNLcontext *tnl = TNL_CONTEXT(ctx); struct vertex_buffer *VB = &tnl->vb; - DECLARE_RENDERINPUTS(index_bitset); - GLuint InputsRead = 0, OutputsWritten = 0; - int vap_fmt_1 = 0; - int offset = 0; - int vte = 0; - int fog_id; - GLint inputs[VERT_ATTRIB_MAX]; - GLint tab[VERT_ATTRIB_MAX]; - int swizzle[VERT_ATTRIB_MAX][4]; - GLuint i, nr; - GLuint sz; - - DECLARE_RENDERINPUTS(render_inputs_bitset); - RENDERINPUTS_COPY(render_inputs_bitset, tnl->render_inputs_bitset); - RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset ); - RENDERINPUTS_COPY(rmesa->state.render_inputs_bitset, render_inputs_bitset); - - vte = rmesa->hw.vte.cmd[1]; - vte &= ~(R300_VTX_XY_FMT | R300_VTX_Z_FMT | R300_VTX_W0_FMT); - /* Important: - */ - if ( VB->NdcPtr != NULL ) { - VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr; - vte |= R300_VTX_XY_FMT | R300_VTX_Z_FMT; - } - else { - VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr; - vte |= R300_VTX_W0_FMT; - } - - assert( VB->AttribPtr[VERT_ATTRIB_POS] != NULL ); - rmesa->swtcl.vertex_attr_count = 0; - - /* EMIT_ATTR's must be in order as they tell t_vertex.c how to - * build up a hardware vertex. - */ - if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS)) { - sz = VB->AttribPtr[VERT_ATTRIB_POS]->size; - InputsRead |= 1 << VERT_ATTRIB_POS; - OutputsWritten |= 1 << VERT_RESULT_HPOS; - EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 ); - offset = sz; - } else { - offset = 4; - EMIT_PAD(4 * sizeof(float)); - } -/* - if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POINTSIZE )) { - EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F ); - offset += 1; - } -*/ - if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_COLOR0)) { - sz = VB->AttribPtr[VERT_ATTRIB_COLOR0]->size; - rmesa->swtcl.coloroffset = offset; + int first_free_tex = 0; + GLuint InputsRead = 0; + GLuint OutputsWritten = 0; + int num_attrs = 0; + GLuint fp_reads = ctx->FragmentProgram._Current->Base.InputsRead; + struct vertex_attribute *attrs = rmesa->vbuf.attribs; + + rmesa->swtcl.coloroffset = rmesa->swtcl.specoffset = 0; + rmesa->radeon.swtcl.vertex_attr_count = 0; + + /* We always want non Ndc coords format */ + VB->AttribPtr[VERT_ATTRIB_POS] = VB->ClipPtr; + + /* Always write position vector */ + InputsRead |= 1 << VERT_ATTRIB_POS; + OutputsWritten |= 1 << VERT_RESULT_HPOS; + EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F ); + ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_POS, SWIZZLE_XYZW, MASK_XYZW, 0); + rmesa->swtcl.coloroffset = 4; + + if (fp_reads & FRAG_BIT_COL0) { InputsRead |= 1 << VERT_ATTRIB_COLOR0; OutputsWritten |= 1 << VERT_RESULT_COL0; - EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_1F + sz - 1 ); - offset += sz; +#if MESA_LITTLE_ENDIAN + EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_RGBA ); + ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1); +#else + EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_ABGR ); + ADD_ATTR(VERT_ATTRIB_COLOR0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR0, SWIZZLE_XYZW, MASK_XYZW, 1); +#endif } - rmesa->swtcl.specoffset = 0; - if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) { - sz = VB->AttribPtr[VERT_ATTRIB_COLOR1]->size; - rmesa->swtcl.specoffset = offset; - EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_1F + sz - 1 ); + if (fp_reads & FRAG_BIT_COL1) { + GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); InputsRead |= 1 << VERT_ATTRIB_COLOR1; OutputsWritten |= 1 << VERT_RESULT_COL1; +#if MESA_LITTLE_ENDIAN + EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_RGBA ); + ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1); +#else + EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_4UB_4F_ABGR ); + ADD_ATTR(VERT_ATTRIB_COLOR1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR1, swiz, MASK_XYZW, 1); +#endif + rmesa->swtcl.specoffset = rmesa->swtcl.coloroffset + 1; } - fog_id = -1; - if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_FOG)) { - /* find first free tex coord slot */ - if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) { - int i; - for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { - if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) { - fog_id = i; - break; - } - } - } else { - fog_id = 0; - } - - if (fog_id == -1) { - fprintf(stderr, "\tout of free texcoords to do fog\n"); - _mesa_exit(-1); + if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) { + VB->AttribPtr[VERT_ATTRIB_GENERIC0] = VB->ColorPtr[1]; + OutputsWritten |= 1 << VERT_RESULT_BFC0; +#if MESA_LITTLE_ENDIAN + EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_RGBA ); + ADD_ATTR(VERT_ATTRIB_GENERIC0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR2, SWIZZLE_XYZW, MASK_XYZW, 1); +#else + EMIT_ATTR( _TNL_ATTRIB_GENERIC0, EMIT_4UB_4F_ABGR ); + ADD_ATTR(VERT_ATTRIB_GENERIC0, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR2, SWIZZLE_XYZW, MASK_XYZW, 1); +#endif + if (fp_reads & FRAG_BIT_COL1) { + VB->AttribPtr[VERT_ATTRIB_GENERIC1] = VB->SecondaryColorPtr[1]; + GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); + OutputsWritten |= 1 << VERT_RESULT_BFC1; +#if MESA_LITTLE_ENDIAN + EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_RGBA ); + ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1); +#else + EMIT_ATTR( _TNL_ATTRIB_GENERIC1, EMIT_4UB_4F_ABGR ); + ADD_ATTR(VERT_ATTRIB_GENERIC1, R300_DATA_TYPE_BYTE, SWTCL_OVM_COLOR3, swiz, MASK_XYZW, 1); +#endif } + } - sz = VB->AttribPtr[VERT_ATTRIB_FOG]->size; - EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F + sz - 1); - InputsRead |= 1 << VERT_ATTRIB_FOG; - OutputsWritten |= 1 << VERT_RESULT_FOGC; - vap_fmt_1 |= sz << (3 * fog_id); + if (RENDERINPUTS_TEST(tnl->render_inputs_bitset, _TNL_ATTRIB_POINTSIZE )) { + GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO); + InputsRead |= 1 << VERT_ATTRIB_POINT_SIZE; + OutputsWritten |= 1 << VERT_RESULT_PSIZ; + EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F ); + ADD_ATTR(VERT_ATTRIB_POINT_SIZE, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_POINT_SIZE, swiz, MASK_X, 0); } - if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) { + /** + * Sending only one texcoord component may lead to lock up, + * so for all textures always output 4 texcoord components to RS. + */ + { int i; - + GLuint swiz, format, hw_format; for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { - if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) { - sz = VB->TexCoordPtr[i]->size; + if (fp_reads & FRAG_BIT_TEX(i)) { + switch (VB->TexCoordPtr[i]->size) { + case 1: + format = EMIT_1F; + hw_format = R300_DATA_TYPE_FLOAT_1; + swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE); + break; + case 2: + format = EMIT_2F; + hw_format = R300_DATA_TYPE_FLOAT_2; + swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ONE); + break; + case 3: + format = EMIT_3F; + hw_format = R300_DATA_TYPE_FLOAT_3; + swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); + break; + case 4: + format = EMIT_4F; + hw_format = R300_DATA_TYPE_FLOAT_4; + swiz = SWIZZLE_XYZW; + break; + default: + continue; + } InputsRead |= 1 << (VERT_ATTRIB_TEX0 + i); OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i); - EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_1F + sz - 1 ); - vap_fmt_1 |= sz << (3 * i); + EMIT_ATTR(_TNL_ATTRIB_TEX(i), format); + ADD_ATTR(VERT_ATTRIB_TEX0 + i, hw_format, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0); + ++first_free_tex; } } } /* RS can't put fragment position on the pixel stack, so stuff it in texcoord if needed */ - if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POS) && (ctx->FragmentProgram._Current->Base.InputsRead & FRAG_BIT_WPOS)) { - int first_free_tex = -1; - if (fog_id >= 0) { - first_free_tex = fog_id+1; - } else { - if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) { - int i; - for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { - if (!RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) { - first_free_tex = i; - break; - } - } - } else { - first_free_tex = 0; - } - } - - if (first_free_tex == -1) { + if (fp_reads & FRAG_BIT_WPOS) { + if (first_free_tex >= ctx->Const.MaxTextureUnits) { fprintf(stderr, "\tout of free texcoords to write w pos\n"); _mesa_exit(-1); } - sz = VB->AttribPtr[VERT_ATTRIB_POS]->size; InputsRead |= 1 << (VERT_ATTRIB_TEX0 + first_free_tex); OutputsWritten |= 1 << (VERT_RESULT_TEX0 + first_free_tex); - EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_1F + sz - 1 ); - vap_fmt_1 |= sz << (3 * first_free_tex); - } - - for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) { - if (InputsRead & (1 << i)) { - inputs[i] = nr++; - } else { - inputs[i] = -1; - } + EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F ); + ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_TEX(first_free_tex), SWIZZLE_XYZW, MASK_XYZW, 0); + ++first_free_tex; } - /* Fixed, apply to vir0 only */ - if (InputsRead & (1 << VERT_ATTRIB_POS)) - inputs[VERT_ATTRIB_POS] = 0; - if (InputsRead & (1 << VERT_ATTRIB_COLOR0)) - inputs[VERT_ATTRIB_COLOR0] = 2; - if (InputsRead & (1 << VERT_ATTRIB_COLOR1)) - inputs[VERT_ATTRIB_COLOR1] = 3; - if (InputsRead & (1 << VERT_ATTRIB_FOG)) - inputs[VERT_ATTRIB_FOG] = 6 + fog_id; - for (i = VERT_ATTRIB_TEX0; i <= VERT_ATTRIB_TEX7; i++) - if (InputsRead & (1 << i)) - inputs[i] = 6 + (i - VERT_ATTRIB_TEX0); - - for (i = 0, nr = 0; i < VERT_ATTRIB_MAX; i++) { - if (InputsRead & (1 << i)) { - tab[nr++] = i; + if (fp_reads & FRAG_BIT_FOGC) { + if (first_free_tex >= ctx->Const.MaxTextureUnits) { + fprintf(stderr, "\tout of free texcoords to write fog coordinate\n"); + _mesa_exit(-1); } - } - - for (i = 0; i < nr; i++) { - int ci; - swizzle[i][0] = SWIZZLE_ZERO; - swizzle[i][1] = SWIZZLE_ZERO; - swizzle[i][2] = SWIZZLE_ZERO; - swizzle[i][3] = SWIZZLE_ONE; - - for (ci = 0; ci < VB->AttribPtr[tab[i]]->size; ci++) { - swizzle[i][ci] = ci; - } + InputsRead |= 1 << VERT_ATTRIB_FOG; + OutputsWritten |= 1 << VERT_RESULT_FOGC; + GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO); + EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F ); + ADD_ATTR(VERT_ATTRIB_FOG, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0); } R300_NEWPRIM(rmesa); - R300_STATECHANGE(rmesa, vir[0]); - ((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count = - r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0], - VB->AttribPtr, inputs, tab, nr); - R300_STATECHANGE(rmesa, vir[1]); - ((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count = - r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle, - nr); - - R300_STATECHANGE(rmesa, vic); - rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead); - rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead); - - R300_STATECHANGE(rmesa, vof); - rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten); - rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = vap_fmt_1; - - rmesa->swtcl.vertex_size = - _tnl_install_attrs( ctx, - rmesa->swtcl.vertex_attrs, - rmesa->swtcl.vertex_attr_count, - NULL, 0 ); - - rmesa->swtcl.vertex_size /= 4; + rmesa->vbuf.num_attribs = num_attrs; + *_InputsRead = InputsRead; + *_OutputsWritten = OutputsWritten; - RENDERINPUTS_COPY( rmesa->tnl_index_bitset, index_bitset ); - - - R300_STATECHANGE(rmesa, vte); - rmesa->hw.vte.cmd[1] = vte; - rmesa->hw.vte.cmd[2] = rmesa->swtcl.vertex_size; + RENDERINPUTS_COPY(rmesa->render_inputs_bitset, tnl->render_inputs_bitset); } - -/* Flush vertices in the current dma region. - */ -static void flush_last_swtcl_prim( r300ContextPtr rmesa ) +static void r300PrepareVertices(GLcontext *ctx) { - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); - - rmesa->dma.flush = NULL; - - if (rmesa->dma.current.buf) { - struct r300_dma_region *current = &rmesa->dma.current; - GLuint current_offset = GET_START(current); - - assert (current->start + - rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == - current->ptr); - - if (rmesa->dma.current.start != rmesa->dma.current.ptr) { - - r300EnsureCmdBufSpace( rmesa, rmesa->hw.max_state_size + (12*sizeof(int)), __FUNCTION__); - - r300EmitState(rmesa); - - r300EmitVertexAOS( rmesa, - rmesa->swtcl.vertex_size, - current_offset); + r300ContextPtr rmesa = R300_CONTEXT(ctx); + GLuint InputsRead, OutputsWritten; - r300EmitVbufPrim( rmesa, - rmesa->swtcl.hw_primitive, - rmesa->swtcl.numverts); + r300ChooseSwtclVertexFormat(ctx, &InputsRead, &OutputsWritten); + r300SetupVAP(ctx, InputsRead, OutputsWritten); - r300EmitCacheFlush(rmesa); - } + rmesa->radeon.swtcl.vertex_size = + _tnl_install_attrs( ctx, + rmesa->radeon.swtcl.vertex_attrs, + rmesa->radeon.swtcl.vertex_attr_count, + NULL, 0 ); - rmesa->swtcl.numverts = 0; - current->start = current->ptr; - } + rmesa->radeon.swtcl.vertex_size /= 4; } -/* Alloc space in the current dma region. - */ -static void * -r300AllocDmaLowVerts( r300ContextPtr rmesa, int nverts, int vsize ) -{ - GLuint bytes = vsize * nverts; - - if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) - r300RefillCurrentDmaRegion( rmesa, bytes); - - if (!rmesa->dma.flush) { - rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; - rmesa->dma.flush = flush_last_swtcl_prim; - } - - ASSERT( vsize == rmesa->swtcl.vertex_size * 4 ); - ASSERT( rmesa->dma.flush == flush_last_swtcl_prim ); - ASSERT( rmesa->dma.current.start + - rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 == - rmesa->dma.current.ptr ); - - { - GLubyte *head = (GLubyte *) (rmesa->dma.current.address + rmesa->dma.current.ptr); - rmesa->dma.current.ptr += bytes; - rmesa->swtcl.numverts += nverts; - return head; - } -} static GLuint reduced_prim[] = { - GL_POINTS, - GL_LINES, - GL_LINES, - GL_LINES, - GL_TRIANGLES, - GL_TRIANGLES, - GL_TRIANGLES, - GL_TRIANGLES, - GL_TRIANGLES, - GL_TRIANGLES, + GL_POINTS, + GL_LINES, + GL_LINES, + GL_LINES, + GL_TRIANGLES, + GL_TRIANGLES, + GL_TRIANGLES, + GL_TRIANGLES, + GL_TRIANGLES, + GL_TRIANGLES, }; static void r300RasterPrimitive( GLcontext *ctx, GLuint prim ); -static void r300RenderPrimitive( GLcontext *ctx, GLenum prim ); -//static void r300ResetLineStipple( GLcontext *ctx ); /*********************************************************************** * Emit primitives as inline vertices * @@ -405,15 +280,13 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim ); #undef LOCAL_VARS #undef ALLOC_VERTS #define CTX_ARG r300ContextPtr rmesa -#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size -#define ALLOC_VERTS( n, size ) r300AllocDmaLowVerts( rmesa, n, size * 4 ) +#define GET_VERTEX_DWORDS() rmesa->radeon.swtcl.vertex_size +#define ALLOC_VERTS( n, size ) rcommonAllocDmaLowVerts( &rmesa->radeon, n, size * 4 ) #define LOCAL_VARS \ r300ContextPtr rmesa = R300_CONTEXT(ctx); \ - const char *r300verts = (char *)rmesa->swtcl.verts; + const char *r300verts = (char *)rmesa->radeon.swtcl.verts; #define VERT(x) (r300Vertex *)(r300verts + ((x) * vertsize * sizeof(int))) #define VERTEX r300Vertex -#define DO_DEBUG_VERTS (1 && (RADEON_DEBUG & DEBUG_VERTS)) -#define PRINT_VERTEX(x) #undef TAG #define TAG(x) r300_##x #include "tnl_dd/t_dd_triemit.h" @@ -433,9 +306,8 @@ static void r300RenderPrimitive( GLcontext *ctx, GLenum prim ); * Build render functions from dd templates * ***********************************************************************/ -#define R300_TWOSIDE_BIT 0x01 -#define R300_UNFILLED_BIT 0x02 -#define R300_MAX_TRIFUNC 0x04 +#define R300_UNFILLED_BIT 0x01 +#define R300_MAX_TRIFUNC 0x02 static struct { tnl_points_func points; @@ -446,9 +318,9 @@ static struct { #define DO_FALLBACK 0 #define DO_UNFILLED (IND & R300_UNFILLED_BIT) -#define DO_TWOSIDE (IND & R300_TWOSIDE_BIT) +#define DO_TWOSIDE 0 #define DO_FLAT 0 -#define DO_OFFSET 0 +#define DO_OFFSET 0 #define DO_TRI 1 #define DO_QUAD 1 #define DO_LINE 1 @@ -468,33 +340,39 @@ static struct { #define VERT_Y(_v) _v->v.y #define VERT_Z(_v) _v->v.z #define AREA_IS_CCW( a ) (a < 0) -#define GET_VERTEX(e) (rmesa->swtcl.verts + (e*rmesa->swtcl.vertex_size*sizeof(int))) - -/* Only used to pull back colors into vertices (ie, we know color is - * floating point). - */ -#define R300_COLOR( dst, src ) \ -do { \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]); \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]); \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]); \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[3], (src)[3]); \ +#define GET_VERTEX(e) (rmesa->radeon.swtcl.verts + (e*rmesa->radeon.swtcl.vertex_size*sizeof(int))) + +#define VERT_SET_RGBA( v, c ) \ +do { \ + r300_color_t *color = (r300_color_t *)&((v)->ui[coloroffset]); \ + UNCLAMPED_FLOAT_TO_UBYTE(color->red, (c)[0]); \ + UNCLAMPED_FLOAT_TO_UBYTE(color->green, (c)[1]); \ + UNCLAMPED_FLOAT_TO_UBYTE(color->blue, (c)[2]); \ + UNCLAMPED_FLOAT_TO_UBYTE(color->alpha, (c)[3]); \ } while (0) -#define VERT_SET_RGBA( v, c ) if (coloroffset) R300_COLOR( v->ub4[coloroffset], c ) -#define VERT_COPY_RGBA( v0, v1 ) if (coloroffset) v0->ui[coloroffset] = v1->ui[coloroffset] -#define VERT_SAVE_RGBA( idx ) if (coloroffset) color[idx] = v[idx]->ui[coloroffset] -#define VERT_RESTORE_RGBA( idx ) if (coloroffset) v[idx]->ui[coloroffset] = color[idx] +#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset] + +#define VERT_SET_SPEC( v0, c ) \ +do { \ + if (specoffset) { \ + UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.red, (c)[0]); \ + UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.green, (c)[1]); \ + UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.blue, (c)[2]); \ + } \ +} while (0) -#define R300_SPEC( dst, src ) \ -do { \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[0], (src)[2]); \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[1], (src)[1]); \ - UNCLAMPED_FLOAT_TO_UBYTE((dst)[2], (src)[0]); \ +#define VERT_COPY_SPEC( v0, v1 ) \ +do { \ + if (specoffset) { \ + v0->v.specular.red = v1->v.specular.red; \ + v0->v.specular.green = v1->v.specular.green; \ + v0->v.specular.blue = v1->v.specular.blue; \ + } \ } while (0) -#define VERT_SET_SPEC( v, c ) if (specoffset) R300_SPEC( v->ub4[specoffset], c ) -#define VERT_COPY_SPEC( v0, v1 ) if (specoffset) COPY_3V(v0->ub4[specoffset], v1->ub4[specoffset]) +#define VERT_SAVE_RGBA( idx ) color[idx] = v[idx]->ui[coloroffset] +#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx] #define VERT_SAVE_SPEC( idx ) if (specoffset) spec[idx] = v[idx]->ui[specoffset] #define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx] @@ -514,7 +392,7 @@ do { \ ***********************************************************************/ #define RASTERIZE(x) r300RasterPrimitive( ctx, reduced_prim[x] ) -#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive +#define RENDER_PRIMITIVE rmesa->radeon.swtcl.render_primitive #undef TAG #define TAG(x) x #include "tnl_dd/t_dd_unfilled.h" @@ -530,26 +408,15 @@ do { \ #define TAG(x) x #include "tnl_dd/t_dd_tritmp.h" -#define IND (R300_TWOSIDE_BIT) -#define TAG(x) x##_twoside -#include "tnl_dd/t_dd_tritmp.h" - #define IND (R300_UNFILLED_BIT) #define TAG(x) x##_unfilled #include "tnl_dd/t_dd_tritmp.h" -#define IND (R300_TWOSIDE_BIT|R300_UNFILLED_BIT) -#define TAG(x) x##_twoside_unfilled -#include "tnl_dd/t_dd_tritmp.h" - - static void init_rast_tab( void ) { init(); - init_twoside(); init_unfilled(); - init_twoside_unfilled(); } /**********************************************************************/ @@ -571,8 +438,8 @@ static void init_rast_tab( void ) #undef LOCAL_VARS #define LOCAL_VARS \ r300ContextPtr rmesa = R300_CONTEXT(ctx); \ - const GLuint vertsize = rmesa->swtcl.vertex_size; \ - const char *r300verts = (char *)rmesa->swtcl.verts; \ + const GLuint vertsize = rmesa->radeon.swtcl.vertex_size; \ + const char *r300verts = (char *)rmesa->radeon.swtcl.verts; \ const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts; \ const GLboolean stipple = ctx->Line.StippleFlag; \ (void) elt; (void) stipple; @@ -601,10 +468,9 @@ static void r300ChooseRenderState( GLcontext *ctx ) GLuint index = 0; GLuint flags = ctx->_TriangleCaps; - if (flags & DD_TRI_LIGHT_TWOSIDE) index |= R300_TWOSIDE_BIT; if (flags & DD_TRI_UNFILLED) index |= R300_UNFILLED_BIT; - if (index != rmesa->swtcl.RenderIndex) { + if (index != rmesa->radeon.swtcl.RenderIndex) { tnl->Driver.Render.Points = rast_tab[index].points; tnl->Driver.Render.Line = rast_tab[index].line; tnl->Driver.Render.ClippedLine = rast_tab[index].line; @@ -621,30 +487,32 @@ static void r300ChooseRenderState( GLcontext *ctx ) tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon; } - rmesa->swtcl.RenderIndex = index; + rmesa->radeon.swtcl.RenderIndex = index; } } -static void r300RenderStart(GLcontext *ctx) +void r300RenderStart(GLcontext *ctx) { - r300ContextPtr rmesa = R300_CONTEXT( ctx ); + r300ContextPtr rmesa = R300_CONTEXT( ctx ); r300ChooseRenderState(ctx); - r300SetVertexFormat(ctx); + r300PrepareVertices(ctx); + + r300ValidateBuffers(ctx); r300UpdateShaders(rmesa); r300UpdateShaderStates(rmesa); r300EmitCacheFlush(rmesa); - if (rmesa->dma.flush != 0 && - rmesa->dma.flush != flush_last_swtcl_prim) - rmesa->dma.flush( rmesa ); - + /* investigate if we can put back flush optimisation if needed */ + if (rmesa->radeon.dma.flush != NULL) { + rmesa->radeon.dma.flush(ctx); + } } -static void r300RenderFinish(GLcontext *ctx) +void r300RenderFinish(GLcontext *ctx) { } @@ -652,28 +520,26 @@ static void r300RasterPrimitive( GLcontext *ctx, GLuint hwprim ) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - if (rmesa->swtcl.hw_primitive != hwprim) { - R300_NEWPRIM( rmesa ); - rmesa->swtcl.hw_primitive = hwprim; + if (rmesa->radeon.swtcl.hw_primitive != hwprim) { + R300_NEWPRIM( rmesa ); + rmesa->radeon.swtcl.hw_primitive = hwprim; } } -static void r300RenderPrimitive(GLcontext *ctx, GLenum prim) +void r300RenderPrimitive(GLcontext *ctx, GLenum prim) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - rmesa->swtcl.render_primitive = prim; + rmesa->radeon.swtcl.render_primitive = prim; if ((prim == GL_TRIANGLES) && (ctx->_TriangleCaps & DD_TRI_UNFILLED)) - return; + return; r300RasterPrimitive( ctx, reduced_prim[prim] ); } -static void r300ResetLineStipple(GLcontext *ctx) +void r300ResetLineStipple(GLcontext *ctx) { - - } void r300InitSwtcl(GLcontext *ctx) @@ -699,50 +565,68 @@ void r300InitSwtcl(GLcontext *ctx) _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 48 * sizeof(GLfloat) ); - rmesa->swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf; - rmesa->swtcl.RenderIndex = ~0; - rmesa->swtcl.render_primitive = GL_TRIANGLES; - rmesa->swtcl.hw_primitive = 0; + rmesa->radeon.swtcl.verts = (GLubyte *)tnl->clipspace.vertex_buf; + rmesa->radeon.swtcl.RenderIndex = ~0; + rmesa->radeon.swtcl.render_primitive = GL_TRIANGLES; + rmesa->radeon.swtcl.hw_primitive = 0; _tnl_invalidate_vertex_state( ctx, ~0 ); _tnl_invalidate_vertices( ctx, ~0 ); - RENDERINPUTS_ZERO( rmesa->tnl_index_bitset ); _tnl_need_projected_coords( ctx, GL_FALSE ); - r300ChooseRenderState(ctx); } void r300DestroySwtcl(GLcontext *ctx) { } -void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, GLuint offset) +static void r300EmitVertexAOS(r300ContextPtr rmesa, GLuint vertex_size, struct radeon_bo *bo, GLuint offset) { - int cmd_reserved = 0; - int cmd_written = 0; + BATCH_LOCALS(&rmesa->radeon); - drm_radeon_cmd_header_t *cmd = NULL; if (RADEON_DEBUG & DEBUG_VERTS) - fprintf(stderr, "%s: vertex_size %d, offset 0x%x \n", - __FUNCTION__, vertex_size, offset); - - start_packet3(CP_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2), 2); - e32(1); - e32(vertex_size | (vertex_size << 8)); - e32(offset); + fprintf(stderr, "%s: vertex_size %d, offset 0x%x \n", + __FUNCTION__, vertex_size, offset); + + BEGIN_BATCH(7); + OUT_BATCH_PACKET3(R300_PACKET3_3D_LOAD_VBPNTR, 2); + OUT_BATCH(1); + OUT_BATCH(vertex_size | (vertex_size << 8)); + OUT_BATCH_RELOC(offset, bo, offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + END_BATCH(); } -void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr) +static void r300EmitVbufPrim(r300ContextPtr rmesa, GLuint primitive, GLuint vertex_nr) { - - int cmd_reserved = 0; - int cmd_written = 0; + BATCH_LOCALS(&rmesa->radeon); int type, num_verts; - drm_radeon_cmd_header_t *cmd = NULL; type = r300PrimitiveType(rmesa, primitive); num_verts = r300NumVerts(rmesa, vertex_nr, primitive); - start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0); - e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type); + BEGIN_BATCH(3); + OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0); + OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type); + END_BATCH(); +} + +void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset) +{ + r300ContextPtr rmesa = R300_CONTEXT(ctx); + + rcommonEnsureCmdBufSpace(&rmesa->radeon, + rmesa->radeon.hw.max_state_size + (12*sizeof(int)), + __FUNCTION__); + radeonEmitState(&rmesa->radeon); + r300_emit_scissor(ctx); + r300EmitVertexAOS(rmesa, + rmesa->radeon.swtcl.vertex_size, + rmesa->radeon.dma.current, + current_offset); + + r300EmitVbufPrim(rmesa, + rmesa->radeon.swtcl.hw_primitive, + rmesa->radeon.swtcl.numverts); + r300EmitCacheFlush(rmesa); + COMMIT_BATCH(); } diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.h b/src/mesa/drivers/dri/r300/r300_swtcl.h index 55df53c1ad..c271d26546 100644 --- a/src/mesa/drivers/dri/r300/r300_swtcl.h +++ b/src/mesa/drivers/dri/r300/r300_swtcl.h @@ -39,7 +39,27 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "swrast/swrast.h" #include "r300_context.h" +/* + * Here are definitions of OVM locations of vertex attributes for non TCL hw + */ +#define SWTCL_OVM_POS 0 +#define SWTCL_OVM_COLOR0 2 +#define SWTCL_OVM_COLOR1 3 +#define SWTCL_OVM_COLOR2 4 +#define SWTCL_OVM_COLOR3 5 +#define SWTCL_OVM_TEX(n) ((n) + 6) +#define SWTCL_OVM_POINT_SIZE 15 + +extern void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *InputsRead, GLuint *OutputsWritten); + extern void r300InitSwtcl( GLcontext *ctx ); extern void r300DestroySwtcl( GLcontext *ctx ); +extern void r300RenderStart(GLcontext *ctx); +extern void r300RenderFinish(GLcontext *ctx); +extern void r300RenderPrimitive(GLcontext *ctx, GLenum prim); +extern void r300ResetLineStipple(GLcontext *ctx); + +extern void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset); + #endif diff --git a/src/mesa/drivers/dri/r300/r300_tex.c b/src/mesa/drivers/dri/r300/r300_tex.c index 7c699ec572..0af5bb4f46 100644 --- a/src/mesa/drivers/dri/r300/r300_tex.c +++ b/src/mesa/drivers/dri/r300/r300_tex.c @@ -38,6 +38,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "main/context.h" #include "main/enums.h" #include "main/image.h" +#include "main/mipmap.h" #include "main/simple_list.h" #include "main/texformat.h" #include "main/texstore.h" @@ -49,6 +50,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_context.h" #include "r300_state.h" #include "r300_ioctl.h" +#include "radeon_mipmap_tree.h" #include "r300_tex.h" #include "xmlpool.h" @@ -77,20 +79,20 @@ static unsigned int translate_wrap_mode(GLenum wrapmode) * * \param t Texture object whose wrap modes are to be set */ -static void r300UpdateTexWrap(r300TexObjPtr t) +static void r300UpdateTexWrap(radeonTexObjPtr t) { - struct gl_texture_object *tObj = t->base.tObj; + struct gl_texture_object *tObj = &t->base; - t->filter &= + t->pp_txfilter &= ~(R300_TX_WRAP_S_MASK | R300_TX_WRAP_T_MASK | R300_TX_WRAP_R_MASK); - t->filter |= translate_wrap_mode(tObj->WrapS) << R300_TX_WRAP_S_SHIFT; + t->pp_txfilter |= translate_wrap_mode(tObj->WrapS) << R300_TX_WRAP_S_SHIFT; if (tObj->Target != GL_TEXTURE_1D) { - t->filter |= translate_wrap_mode(tObj->WrapT) << R300_TX_WRAP_T_SHIFT; + t->pp_txfilter |= translate_wrap_mode(tObj->WrapT) << R300_TX_WRAP_T_SHIFT; if (tObj->Target == GL_TEXTURE_3D) - t->filter |= translate_wrap_mode(tObj->WrapR) << R300_TX_WRAP_R_SHIFT; + t->pp_txfilter |= translate_wrap_mode(tObj->WrapR) << R300_TX_WRAP_R_SHIFT; } } @@ -117,10 +119,13 @@ static GLuint aniso_filter(GLfloat anisotropy) * \param magf Texture magnification mode * \param anisotropy Maximum anisotropy level */ -static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat anisotropy) +static void r300SetTexFilter(radeonTexObjPtr t, GLenum minf, GLenum magf, GLfloat anisotropy) { - t->filter &= ~(R300_TX_MIN_FILTER_MASK | R300_TX_MIN_FILTER_MIP_MASK | R300_TX_MAG_FILTER_MASK | R300_TX_MAX_ANISO_MASK); - t->filter_1 &= ~R300_EDGE_ANISO_EDGE_ONLY; + /* Force revalidation to account for switches from/to mipmapping. */ + t->validated = GL_FALSE; + + t->pp_txfilter &= ~(R300_TX_MIN_FILTER_MASK | R300_TX_MIN_FILTER_MIP_MASK | R300_TX_MAG_FILTER_MASK | R300_TX_MAX_ANISO_MASK); + t->pp_txfilter_1 &= ~R300_EDGE_ANISO_EDGE_ONLY; /* Note that EXT_texture_filter_anisotropic is extremely vague about * how anisotropic filtering interacts with the "normal" filter modes. @@ -128,7 +133,7 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat * filter settings completely. This includes driconf's settings. */ if (anisotropy >= 2.0 && (minf != GL_NEAREST) && (magf != GL_NEAREST)) { - t->filter |= R300_TX_MAG_FILTER_ANISO + t->pp_txfilter |= R300_TX_MAG_FILTER_ANISO | R300_TX_MIN_FILTER_ANISO | R300_TX_MIN_FILTER_MIP_LINEAR | aniso_filter(anisotropy); @@ -139,22 +144,22 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat switch (minf) { case GL_NEAREST: - t->filter |= R300_TX_MIN_FILTER_NEAREST; + t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST; break; case GL_LINEAR: - t->filter |= R300_TX_MIN_FILTER_LINEAR; + t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR; break; case GL_NEAREST_MIPMAP_NEAREST: - t->filter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_NEAREST; + t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_NEAREST; break; case GL_NEAREST_MIPMAP_LINEAR: - t->filter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_LINEAR; + t->pp_txfilter |= R300_TX_MIN_FILTER_NEAREST|R300_TX_MIN_FILTER_MIP_LINEAR; break; case GL_LINEAR_MIPMAP_NEAREST: - t->filter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_NEAREST; + t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_NEAREST; break; case GL_LINEAR_MIPMAP_LINEAR: - t->filter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_LINEAR; + t->pp_txfilter |= R300_TX_MIN_FILTER_LINEAR|R300_TX_MIN_FILTER_MIP_LINEAR; break; } @@ -163,15 +168,15 @@ static void r300SetTexFilter(r300TexObjPtr t, GLenum minf, GLenum magf, GLfloat */ switch (magf) { case GL_NEAREST: - t->filter |= R300_TX_MAG_FILTER_NEAREST; + t->pp_txfilter |= R300_TX_MAG_FILTER_NEAREST; break; case GL_LINEAR: - t->filter |= R300_TX_MAG_FILTER_LINEAR; + t->pp_txfilter |= R300_TX_MAG_FILTER_LINEAR; break; } } -static void r300SetTexBorderColor(r300TexObjPtr t, const GLfloat color[4]) +static void r300SetTexBorderColor(radeonTexObjPtr t, const GLfloat color[4]) { GLubyte c[4]; CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]); @@ -182,729 +187,6 @@ static void r300SetTexBorderColor(r300TexObjPtr t, const GLfloat color[4]) } /** - * Allocate space for and load the mesa images into the texture memory block. - * This will happen before drawing with a new texture, or drawing with a - * texture after it was swapped out or teximaged again. - */ - -static r300TexObjPtr r300AllocTexObj(struct gl_texture_object *texObj) -{ - r300TexObjPtr t; - - t = CALLOC_STRUCT(r300_tex_obj); - texObj->DriverData = t; - if (t != NULL) { - if (RADEON_DEBUG & DEBUG_TEXTURE) { - fprintf(stderr, "%s( %p, %p )\n", __FUNCTION__, - (void *)texObj, (void *)t); - } - - /* Initialize non-image-dependent parts of the state: - */ - t->base.tObj = texObj; - t->border_fallback = GL_FALSE; - - make_empty_list(&t->base); - - r300UpdateTexWrap(t); - r300SetTexFilter(t, texObj->MinFilter, texObj->MagFilter, texObj->MaxAnisotropy); - r300SetTexBorderColor(t, texObj->BorderColor); - } - - return t; -} - -/* try to find a format which will only need a memcopy */ -static const struct gl_texture_format *r300Choose8888TexFormat(GLenum srcFormat, - GLenum srcType) -{ - const GLuint ui = 1; - const GLubyte littleEndian = *((const GLubyte *)&ui); - - if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) || - (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) || - (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) || - (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian)) { - return &_mesa_texformat_rgba8888; - } else if ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) || - (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) || - (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) || - (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian)) { - return &_mesa_texformat_rgba8888_rev; - } else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && !littleEndian) || - srcType == GL_UNSIGNED_INT_8_8_8_8)) { - return &_mesa_texformat_argb8888_rev; - } else if (srcFormat == GL_BGRA && ((srcType == GL_UNSIGNED_BYTE && littleEndian) || - srcType == GL_UNSIGNED_INT_8_8_8_8_REV)) { - return &_mesa_texformat_argb8888; - } else - return _dri_texformat_argb8888; -} - -static const struct gl_texture_format *r300ChooseTextureFormat(GLcontext * ctx, - GLint - internalFormat, - GLenum format, - GLenum type) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - const GLboolean do32bpt = - (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_32); - const GLboolean force16bpt = - (rmesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FORCE_16); - (void)format; - -#if 0 - fprintf(stderr, "InternalFormat=%s(%d) type=%s format=%s\n", - _mesa_lookup_enum_by_nr(internalFormat), internalFormat, - _mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format)); - fprintf(stderr, "do32bpt=%d force16bpt=%d\n", do32bpt, force16bpt); -#endif - - switch (internalFormat) { - case 4: - case GL_RGBA: - case GL_COMPRESSED_RGBA: - switch (type) { - case GL_UNSIGNED_INT_10_10_10_2: - case GL_UNSIGNED_INT_2_10_10_10_REV: - return do32bpt ? _dri_texformat_argb8888 : - _dri_texformat_argb1555; - case GL_UNSIGNED_SHORT_4_4_4_4: - case GL_UNSIGNED_SHORT_4_4_4_4_REV: - return _dri_texformat_argb4444; - case GL_UNSIGNED_SHORT_5_5_5_1: - case GL_UNSIGNED_SHORT_1_5_5_5_REV: - return _dri_texformat_argb1555; - default: - return do32bpt ? r300Choose8888TexFormat(format, type) : - _dri_texformat_argb4444; - } - - case 3: - case GL_RGB: - case GL_COMPRESSED_RGB: - switch (type) { - case GL_UNSIGNED_SHORT_4_4_4_4: - case GL_UNSIGNED_SHORT_4_4_4_4_REV: - return _dri_texformat_argb4444; - case GL_UNSIGNED_SHORT_5_5_5_1: - case GL_UNSIGNED_SHORT_1_5_5_5_REV: - return _dri_texformat_argb1555; - case GL_UNSIGNED_SHORT_5_6_5: - case GL_UNSIGNED_SHORT_5_6_5_REV: - return _dri_texformat_rgb565; - default: - return do32bpt ? _dri_texformat_argb8888 : - _dri_texformat_rgb565; - } - - case GL_RGBA8: - case GL_RGB10_A2: - case GL_RGBA12: - case GL_RGBA16: - return !force16bpt ? - r300Choose8888TexFormat(format, - type) : _dri_texformat_argb4444; - - case GL_RGBA4: - case GL_RGBA2: - return _dri_texformat_argb4444; - - case GL_RGB5_A1: - return _dri_texformat_argb1555; - - case GL_RGB8: - case GL_RGB10: - case GL_RGB12: - case GL_RGB16: - return !force16bpt ? _dri_texformat_argb8888 : - _dri_texformat_rgb565; - - case GL_RGB5: - case GL_RGB4: - case GL_R3_G3_B2: - return _dri_texformat_rgb565; - - case GL_ALPHA: - case GL_ALPHA4: - case GL_ALPHA8: - case GL_ALPHA12: - case GL_ALPHA16: - case GL_COMPRESSED_ALPHA: - return _dri_texformat_a8; - - case 1: - case GL_LUMINANCE: - case GL_LUMINANCE4: - case GL_LUMINANCE8: - case GL_LUMINANCE12: - case GL_LUMINANCE16: - case GL_COMPRESSED_LUMINANCE: - return _dri_texformat_l8; - - case 2: - case GL_LUMINANCE_ALPHA: - case GL_LUMINANCE4_ALPHA4: - case GL_LUMINANCE6_ALPHA2: - case GL_LUMINANCE8_ALPHA8: - case GL_LUMINANCE12_ALPHA4: - case GL_LUMINANCE12_ALPHA12: - case GL_LUMINANCE16_ALPHA16: - case GL_COMPRESSED_LUMINANCE_ALPHA: - return _dri_texformat_al88; - - case GL_INTENSITY: - case GL_INTENSITY4: - case GL_INTENSITY8: - case GL_INTENSITY12: - case GL_INTENSITY16: - case GL_COMPRESSED_INTENSITY: - return _dri_texformat_i8; - - case GL_YCBCR_MESA: - if (type == GL_UNSIGNED_SHORT_8_8_APPLE || - type == GL_UNSIGNED_BYTE) - return &_mesa_texformat_ycbcr; - else - return &_mesa_texformat_ycbcr_rev; - - case GL_RGB_S3TC: - case GL_RGB4_S3TC: - case GL_COMPRESSED_RGB_S3TC_DXT1_EXT: - return &_mesa_texformat_rgb_dxt1; - - case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT: - return &_mesa_texformat_rgba_dxt1; - - case GL_RGBA_S3TC: - case GL_RGBA4_S3TC: - case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT: - return &_mesa_texformat_rgba_dxt3; - - case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT: - return &_mesa_texformat_rgba_dxt5; - - case GL_ALPHA16F_ARB: - return &_mesa_texformat_alpha_float16; - case GL_ALPHA32F_ARB: - return &_mesa_texformat_alpha_float32; - case GL_LUMINANCE16F_ARB: - return &_mesa_texformat_luminance_float16; - case GL_LUMINANCE32F_ARB: - return &_mesa_texformat_luminance_float32; - case GL_LUMINANCE_ALPHA16F_ARB: - return &_mesa_texformat_luminance_alpha_float16; - case GL_LUMINANCE_ALPHA32F_ARB: - return &_mesa_texformat_luminance_alpha_float32; - case GL_INTENSITY16F_ARB: - return &_mesa_texformat_intensity_float16; - case GL_INTENSITY32F_ARB: - return &_mesa_texformat_intensity_float32; - case GL_RGB16F_ARB: - return &_mesa_texformat_rgba_float16; - case GL_RGB32F_ARB: - return &_mesa_texformat_rgba_float32; - case GL_RGBA16F_ARB: - return &_mesa_texformat_rgba_float16; - case GL_RGBA32F_ARB: - return &_mesa_texformat_rgba_float32; - - case GL_DEPTH_COMPONENT: - case GL_DEPTH_COMPONENT16: - case GL_DEPTH_COMPONENT24: - case GL_DEPTH_COMPONENT32: -#if 0 - switch (type) { - case GL_UNSIGNED_BYTE: - case GL_UNSIGNED_SHORT: - return &_mesa_texformat_z16; - case GL_UNSIGNED_INT: - return &_mesa_texformat_z32; - case GL_UNSIGNED_INT_24_8_EXT: - default: - return &_mesa_texformat_z24_s8; - } -#else - return &_mesa_texformat_z16; -#endif - - default: - _mesa_problem(ctx, - "unexpected internalFormat 0x%x in r300ChooseTextureFormat", - (int)internalFormat); - return NULL; - } - - return NULL; /* never get here */ -} - -static GLboolean -r300ValidateClientStorage(GLcontext * ctx, GLenum target, - GLint internalFormat, - GLint srcWidth, GLint srcHeight, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "intformat %s format %s type %s\n", - _mesa_lookup_enum_by_nr(internalFormat), - _mesa_lookup_enum_by_nr(format), - _mesa_lookup_enum_by_nr(type)); - - if (!ctx->Unpack.ClientStorage) - return 0; - - if (ctx->_ImageTransferState || - texImage->IsCompressed || texObj->GenerateMipmap) - return 0; - - /* This list is incomplete, may be different on ppc??? - */ - switch (internalFormat) { - case GL_RGBA: - if (format == GL_BGRA && type == GL_UNSIGNED_INT_8_8_8_8_REV) { - texImage->TexFormat = _dri_texformat_argb8888; - } else - return 0; - break; - - case GL_RGB: - if (format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5) { - texImage->TexFormat = _dri_texformat_rgb565; - } else - return 0; - break; - - case GL_YCBCR_MESA: - if (format == GL_YCBCR_MESA && - type == GL_UNSIGNED_SHORT_8_8_REV_APPLE) { - texImage->TexFormat = &_mesa_texformat_ycbcr_rev; - } else if (format == GL_YCBCR_MESA && - (type == GL_UNSIGNED_SHORT_8_8_APPLE || - type == GL_UNSIGNED_BYTE)) { - texImage->TexFormat = &_mesa_texformat_ycbcr; - } else - return 0; - break; - - default: - return 0; - } - - /* Could deal with these packing issues, but currently don't: - */ - if (packing->SkipPixels || - packing->SkipRows || packing->SwapBytes || packing->LsbFirst) { - return 0; - } - - GLint srcRowStride = _mesa_image_row_stride(packing, srcWidth, - format, type); - - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: srcRowStride %d/%x\n", - __FUNCTION__, srcRowStride, srcRowStride); - - /* Could check this later in upload, pitch restrictions could be - * relaxed, but would need to store the image pitch somewhere, - * as packing details might change before image is uploaded: - */ - if (!r300IsGartMemory(rmesa, pixels, srcHeight * srcRowStride) - || (srcRowStride & 63)) - return 0; - - /* Have validated that _mesa_transfer_teximage would be a straight - * memcpy at this point. NOTE: future calls to TexSubImage will - * overwrite the client data. This is explicitly mentioned in the - * extension spec. - */ - texImage->Data = (void *)pixels; - texImage->IsClientData = GL_TRUE; - texImage->RowStride = srcRowStride / texImage->TexFormat->TexelBytes; - - return 1; -} - -static void r300TexImage1D(GLcontext * ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint border, - GLenum format, GLenum type, const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D"); - return; - } - } - - /* Note, this will call ChooseTextureFormat */ - _mesa_store_teximage1d(ctx, target, level, internalFormat, - width, border, format, type, pixels, - &ctx->Unpack, texObj, texImage); - - t->dirty_images[0] |= (1 << level); -} - -static void r300TexSubImage1D(GLcontext * ctx, GLenum target, GLint level, - GLint xoffset, - GLsizei width, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - - assert(t); /* this _should_ be true */ - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D"); - return; - } - } - - _mesa_store_texsubimage1d(ctx, target, level, xoffset, width, - format, type, pixels, packing, texObj, - texImage); - - t->dirty_images[0] |= (1 << level); -} - -static void r300TexImage2D(GLcontext * ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLenum format, GLenum type, const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = - (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - if (t != NULL) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D"); - return; - } - } - - texImage->IsClientData = GL_FALSE; - - if (r300ValidateClientStorage(ctx, target, - internalFormat, - width, height, - format, type, pixels, - packing, texObj, texImage)) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using client storage\n", - __FUNCTION__); - } else { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using normal storage\n", - __FUNCTION__); - - /* Normal path: copy (to cached memory) and eventually upload - * via another copy to GART memory and then a blit... Could - * eliminate one copy by going straight to (permanent) GART. - * - * Note, this will call r300ChooseTextureFormat. - */ - _mesa_store_teximage2d(ctx, target, level, internalFormat, - width, height, border, format, type, - pixels, &ctx->Unpack, texObj, texImage); - - t->dirty_images[face] |= (1 << level); - } -} - -static void r300TexSubImage2D(GLcontext * ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = - (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - assert(t); /* this _should_ be true */ - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D"); - return; - } - } - - _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width, - height, format, type, pixels, packing, texObj, - texImage); - - t->dirty_images[face] |= (1 << level); -} - -static void r300CompressedTexImage2D(GLcontext * ctx, GLenum target, - GLint level, GLint internalFormat, - GLint width, GLint height, GLint border, - GLsizei imageSize, const GLvoid * data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = - (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - if (t != NULL) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "glCompressedTexImage2D"); - return; - } - } - - texImage->IsClientData = GL_FALSE; - - /* can't call this, different parameters. Would never evaluate to true anyway currently */ -#if 0 - if (r300ValidateClientStorage(ctx, target, - internalFormat, - width, height, - format, type, pixels, - packing, texObj, texImage)) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using client storage\n", - __FUNCTION__); - } else -#endif - { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using normal storage\n", - __FUNCTION__); - - /* Normal path: copy (to cached memory) and eventually upload - * via another copy to GART memory and then a blit... Could - * eliminate one copy by going straight to (permanent) GART. - * - * Note, this will call r300ChooseTextureFormat. - */ - _mesa_store_compressed_teximage2d(ctx, target, level, - internalFormat, width, height, - border, imageSize, data, - texObj, texImage); - - t->dirty_images[face] |= (1 << level); - } -} - -static void r300CompressedTexSubImage2D(GLcontext * ctx, GLenum target, - GLint level, GLint xoffset, - GLint yoffset, GLsizei width, - GLsizei height, GLenum format, - GLsizei imageSize, const GLvoid * data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - GLuint face; - - /* which cube face or ordinary 2D image */ - switch (target) { - case GL_TEXTURE_CUBE_MAP_POSITIVE_X: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_X: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Y: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y: - case GL_TEXTURE_CUBE_MAP_POSITIVE_Z: - case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z: - face = - (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X; - ASSERT(face < 6); - break; - default: - face = 0; - } - - assert(t); /* this _should_ be true */ - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, - "glCompressedTexSubImage3D"); - return; - } - } - - _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, - yoffset, width, height, format, - imageSize, data, texObj, texImage); - - t->dirty_images[face] |= (1 << level); -} - -static void r300TexImage3D(GLcontext * ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint depth, - GLint border, - GLenum format, GLenum type, const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage3D"); - return; - } - } - - texImage->IsClientData = GL_FALSE; - -#if 0 - if (r300ValidateClientStorage(ctx, target, - internalFormat, - width, height, - format, type, pixels, - packing, texObj, texImage)) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using client storage\n", - __FUNCTION__); - } else -#endif - { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: Using normal storage\n", - __FUNCTION__); - - /* Normal path: copy (to cached memory) and eventually upload - * via another copy to GART memory and then a blit... Could - * eliminate one copy by going straight to (permanent) GART. - * - * Note, this will call r300ChooseTextureFormat. - */ - _mesa_store_teximage3d(ctx, target, level, internalFormat, - width, height, depth, border, - format, type, pixels, - &ctx->Unpack, texObj, texImage); - - t->dirty_images[0] |= (1 << level); - } -} - -static void -r300TexSubImage3D(GLcontext * ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLsizei width, GLsizei height, GLsizei depth, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage) -{ - driTextureObject *t = (driTextureObject *) texObj->DriverData; - -/* fprintf(stderr, "%s\n", __FUNCTION__); */ - - assert(t); /* this _should_ be true */ - if (t) { - driSwapOutTextureObject(t); - } else { - t = (driTextureObject *) r300AllocTexObj(texObj); - if (!t) { - _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage3D"); - return; - } - texObj->DriverData = t; - } - - _mesa_store_texsubimage3d(ctx, target, level, xoffset, yoffset, zoffset, - width, height, depth, - format, type, pixels, packing, texObj, - texImage); - - t->dirty_images[0] |= (1 << level); -} - -/** * Changes variables and flags for a state update, which will happen at the * next UpdateTextureState */ @@ -913,7 +195,7 @@ static void r300TexParameter(GLcontext * ctx, GLenum target, struct gl_texture_object *texObj, GLenum pname, const GLfloat * params) { - r300TexObjPtr t = (r300TexObjPtr) texObj->DriverData; + radeonTexObj* t = radeon_tex_obj(texObj); if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) { fprintf(stderr, "%s( %s )\n", __FUNCTION__, @@ -946,7 +228,11 @@ static void r300TexParameter(GLcontext * ctx, GLenum target, * we just have to rely on loading the right subset of mipmap levels * to simulate a clamped LOD. */ - driSwapOutTextureObject((driTextureObject *) t); + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = 0; + t->validated = GL_FALSE; + } break; case GL_DEPTH_TEXTURE_MODE: @@ -969,27 +255,10 @@ static void r300TexParameter(GLcontext * ctx, GLenum target, } } -static void r300BindTexture(GLcontext * ctx, GLenum target, - struct gl_texture_object *texObj) -{ - if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) { - fprintf(stderr, "%s( %p ) unit=%d\n", __FUNCTION__, - (void *)texObj, ctx->Texture.CurrentUnit); - } - - if ((target == GL_TEXTURE_1D) - || (target == GL_TEXTURE_2D) - || (target == GL_TEXTURE_3D) - || (target == GL_TEXTURE_CUBE_MAP) - || (target == GL_TEXTURE_RECTANGLE_NV)) { - assert(texObj->DriverData != NULL); - } -} - static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - driTextureObject *t = (driTextureObject *) texObj->DriverData; + radeonTexObj* t = radeon_tex_obj(texObj); if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) { fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__, @@ -997,14 +266,24 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj) _mesa_lookup_enum_by_nr(texObj->Target)); } - if (t != NULL) { - if (rmesa) { - R300_FIREVERTICES(rmesa); - } + if (rmesa) { + int i; + radeon_firevertices(&rmesa->radeon); + + for(i = 0; i < R300_MAX_TEXTURE_UNITS; ++i) + if (rmesa->hw.textures[i] == t) + rmesa->hw.textures[i] = 0; + } - driDestroyTextureObject(t); + if (t->bo) { + radeon_bo_unref(t->bo); + t->bo = NULL; + } + + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = 0; } - /* Free mipmap images and the texture object itself */ _mesa_delete_texture_object(ctx, texObj); } @@ -1013,8 +292,6 @@ static void r300DeleteTexture(GLcontext * ctx, struct gl_texture_object *texObj) * Called via ctx->Driver.NewTextureObject. * Note: this function will be called during context creation to * allocate the default texture objects. - * Note: we could use containment here to 'derive' the driver-specific - * texture object from the core mesa gl_texture_object. Not done at this time. * Fixup MaxAnisotropy according to user preference. */ static struct gl_texture_object *r300NewTextureObject(GLcontext * ctx, @@ -1022,14 +299,23 @@ static struct gl_texture_object *r300NewTextureObject(GLcontext * ctx, GLenum target) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_object *obj; - obj = _mesa_new_texture_object(ctx, name, target); - if (!obj) - return NULL; - obj->MaxAnisotropy = rmesa->initialMaxAnisotropy; + radeonTexObj* t = CALLOC_STRUCT(radeon_tex_obj); + - r300AllocTexObj(obj); - return obj; + if (RADEON_DEBUG & (DEBUG_STATE | DEBUG_TEXTURE)) { + fprintf(stderr, "%s( %p (target = %s) )\n", __FUNCTION__, + t, _mesa_lookup_enum_by_nr(target)); + } + + _mesa_initialize_texture_object(&t->base, name, target); + t->base.MaxAnisotropy = rmesa->radeon.initialMaxAnisotropy; + + /* Initialize hardware state */ + r300UpdateTexWrap(t); + r300SetTexFilter(t, t->base.MinFilter, t->base.MagFilter, t->base.MaxAnisotropy); + r300SetTexBorderColor(t, t->base.BorderColor); + + return &t->base; } void r300InitTextureFuncs(struct dd_function_table *functions) @@ -1037,22 +323,30 @@ void r300InitTextureFuncs(struct dd_function_table *functions) /* Note: we only plug in the functions we implement in the driver * since _mesa_init_driver_functions() was already called. */ - functions->ChooseTextureFormat = r300ChooseTextureFormat; - functions->TexImage1D = r300TexImage1D; - functions->TexImage2D = r300TexImage2D; - functions->TexImage3D = r300TexImage3D; - functions->TexSubImage1D = r300TexSubImage1D; - functions->TexSubImage2D = r300TexSubImage2D; - functions->TexSubImage3D = r300TexSubImage3D; + functions->NewTextureImage = radeonNewTextureImage; + functions->FreeTexImageData = radeonFreeTexImageData; + functions->MapTexture = radeonMapTexture; + functions->UnmapTexture = radeonUnmapTexture; + + functions->ChooseTextureFormat = radeonChooseTextureFormat_mesa; + functions->TexImage1D = radeonTexImage1D; + functions->TexImage2D = radeonTexImage2D; + functions->TexImage3D = radeonTexImage3D; + functions->TexSubImage1D = radeonTexSubImage1D; + functions->TexSubImage2D = radeonTexSubImage2D; + functions->TexSubImage3D = radeonTexSubImage3D; + functions->GetTexImage = radeonGetTexImage; + functions->GetCompressedTexImage = radeonGetCompressedTexImage; functions->NewTextureObject = r300NewTextureObject; - functions->BindTexture = r300BindTexture; functions->DeleteTexture = r300DeleteTexture; functions->IsTextureResident = driIsTextureResident; functions->TexParameter = r300TexParameter; - functions->CompressedTexImage2D = r300CompressedTexImage2D; - functions->CompressedTexSubImage2D = r300CompressedTexSubImage2D; + functions->CompressedTexImage2D = radeonCompressedTexImage2D; + functions->CompressedTexSubImage2D = radeonCompressedTexSubImage2D; + + functions->GenerateMipmap = radeonGenerateMipmap; driInitTextureFormats(); } diff --git a/src/mesa/drivers/dri/r300/r300_tex.h b/src/mesa/drivers/dri/r300/r300_tex.h index b86d45bfe0..8a653ea2d1 100644 --- a/src/mesa/drivers/dri/r300/r300_tex.h +++ b/src/mesa/drivers/dri/r300/r300_tex.h @@ -37,16 +37,17 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. extern void r300SetDepthTexMode(struct gl_texture_object *tObj); +extern void r300SetTexBuffer(__DRIcontext *pDRICtx, GLint target, + __DRIdrawable *dPriv); + +extern void r300SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, + GLint format, __DRIdrawable *dPriv); + extern void r300SetTexOffset(__DRIcontext *pDRICtx, GLint texname, unsigned long long offset, GLint depth, GLuint pitch); -extern void r300UpdateTextureState(GLcontext * ctx); - -extern int r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr t, - GLuint face); - -extern void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t); +extern GLboolean r300ValidateBuffers(GLcontext * ctx); extern void r300InitTextureFuncs(struct dd_function_table *functions); diff --git a/src/mesa/drivers/dri/r300/r300_texmem.c b/src/mesa/drivers/dri/r300/r300_texmem.c deleted file mode 100644 index a89ab83d94..0000000000 --- a/src/mesa/drivers/dri/r300/r300_texmem.c +++ /dev/null @@ -1,568 +0,0 @@ -/************************************************************************** - -Copyright (C) Tungsten Graphics 2002. All Rights Reserved. -The Weather Channel, Inc. funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 -license. This notice must be preserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation on the rights to use, copy, modify, merge, publish, -distribute, sub license, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -NON-INFRINGEMENT. IN NO EVENT SHALL ATI, VA LINUX SYSTEMS AND/OR THEIR -SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER -IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR -IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE -SOFTWARE. - -**************************************************************************/ - -/** - * \file - * - * \author Gareth Hughes <gareth@valinux.com> - * - * \author Kevin E. Martin <martin@valinux.com> - */ - -#include <errno.h> - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/context.h" -#include "main/colormac.h" -#include "main/macros.h" -#include "main/simple_list.h" -#include "main/texobj.h" -#include "radeon_reg.h" /* gets definition for usleep */ -#include "r300_context.h" -#include "r300_state.h" -#include "r300_cmdbuf.h" -#include "radeon_ioctl.h" -#include "r300_tex.h" -#include "r300_ioctl.h" -#include <unistd.h> /* for usleep() */ - -#ifdef USER_BUFFERS -#include "r300_mem.h" -#endif - -/** - * Destroy any device-dependent state associated with the texture. This may - * include NULLing out hardware state that points to the texture. - */ -void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t) -{ - int i; - - if (RADEON_DEBUG & DEBUG_TEXTURE) { - fprintf(stderr, "%s( %p, %p )\n", __FUNCTION__, - (void *)t, (void *)t->base.tObj); - } - - for (i = 0; i < rmesa->radeon.glCtx->Const.MaxTextureUnits; i++) { - if (rmesa->state.texture.unit[i].texobj == t->base.tObj) { - _mesa_reference_texobj(&rmesa->state.texture.unit[i].texobj, NULL); - } - } -} - -/* ------------------------------------------------------------ - * Texture image conversions - */ - -static void r300UploadGARTClientSubImage(r300ContextPtr rmesa, - r300TexObjPtr t, - struct gl_texture_image *texImage, - GLint hwlevel, - GLint x, GLint y, - GLint width, GLint height) -{ - const struct gl_texture_format *texFormat = texImage->TexFormat; - GLuint srcPitch, dstPitch; - int blit_format; - int srcOffset; - - /* - * XXX it appears that we always upload the full image, not a subimage. - * I.e. x==0, y==0, width=texWidth, height=texWidth. If this is ever - * changed, the src pitch will have to change. - */ - switch (texFormat->TexelBytes) { - case 1: - blit_format = R300_CP_COLOR_FORMAT_CI8; - srcPitch = t->image[0][0].width * texFormat->TexelBytes; - dstPitch = t->image[0][0].width * texFormat->TexelBytes; - break; - case 2: - blit_format = R300_CP_COLOR_FORMAT_RGB565; - srcPitch = t->image[0][0].width * texFormat->TexelBytes; - dstPitch = t->image[0][0].width * texFormat->TexelBytes; - break; - case 4: - blit_format = R300_CP_COLOR_FORMAT_ARGB8888; - srcPitch = t->image[0][0].width * texFormat->TexelBytes; - dstPitch = t->image[0][0].width * texFormat->TexelBytes; - break; - case 8: - case 16: - blit_format = R300_CP_COLOR_FORMAT_CI8; - srcPitch = t->image[0][0].width * texFormat->TexelBytes; - dstPitch = t->image[0][0].width * texFormat->TexelBytes; - break; - default: - return; - } - - t->image[0][hwlevel].data = texImage->Data; - srcOffset = r300GartOffsetFromVirtual(rmesa, texImage->Data); - - assert(srcOffset != ~0); - - /* Don't currently need to cope with small pitches? - */ - width = texImage->Width; - height = texImage->Height; - - if (texFormat->TexelBytes > 4) { - width *= texFormat->TexelBytes; - } - - r300EmitWait(rmesa, R300_WAIT_3D); - - r300EmitBlit(rmesa, blit_format, - srcPitch, - srcOffset, - dstPitch, - t->bufAddr, - x, - y, - t->image[0][hwlevel].x + x, - t->image[0][hwlevel].y + y, width, height); - - r300EmitWait(rmesa, R300_WAIT_2D); -} - -static void r300UploadRectSubImage(r300ContextPtr rmesa, - r300TexObjPtr t, - struct gl_texture_image *texImage, - GLint x, GLint y, GLint width, GLint height) -{ - const struct gl_texture_format *texFormat = texImage->TexFormat; - int blit_format, dstPitch, done; - - switch (texFormat->TexelBytes) { - case 1: - blit_format = R300_CP_COLOR_FORMAT_CI8; - break; - case 2: - blit_format = R300_CP_COLOR_FORMAT_RGB565; - break; - case 4: - blit_format = R300_CP_COLOR_FORMAT_ARGB8888; - break; - case 8: - case 16: - blit_format = R300_CP_COLOR_FORMAT_CI8; - break; - default: - return; - } - - t->image[0][0].data = texImage->Data; - - /* Currently don't need to cope with small pitches. - */ - width = texImage->Width; - height = texImage->Height; - dstPitch = t->pitch; - - if (texFormat->TexelBytes > 4) { - width *= texFormat->TexelBytes; - } - - if (rmesa->prefer_gart_client_texturing && texImage->IsClientData) { - /* In this case, could also use GART texturing. This is - * currently disabled, but has been tested & works. - */ - t->offset = r300GartOffsetFromVirtual(rmesa, texImage->Data); - t->pitch = texImage->RowStride * texFormat->TexelBytes - 32; - - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, - "Using GART texturing for rectangular client texture\n"); - - /* Release FB memory allocated for this image: - */ - /* FIXME This may not be correct as driSwapOutTextureObject sets - * FIXME dirty_images. It may be fine, though. - */ - if (t->base.memBlock) { - driSwapOutTextureObject((driTextureObject *) t); - } - } else if (texImage->IsClientData) { - /* Data already in GART memory, with usable pitch. - */ - GLuint srcPitch; - srcPitch = texImage->RowStride * texFormat->TexelBytes; - r300EmitBlit(rmesa, - blit_format, - srcPitch, - r300GartOffsetFromVirtual(rmesa, texImage->Data), - dstPitch, t->bufAddr, 0, 0, 0, 0, width, height); - } else { - /* Data not in GART memory, or bad pitch. - */ - for (done = 0; done < height;) { - struct r300_dma_region region; - int lines = - MIN2(height - done, RADEON_BUFFER_SIZE / dstPitch); - int src_pitch; - char *tex; - - src_pitch = texImage->RowStride * texFormat->TexelBytes; - - tex = (char *)texImage->Data + done * src_pitch; - - memset(®ion, 0, sizeof(region)); - r300AllocDmaRegion(rmesa, ®ion, lines * dstPitch, - 1024); - - /* Copy texdata to dma: - */ - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, - "%s: src_pitch %d dst_pitch %d\n", - __FUNCTION__, src_pitch, dstPitch); - - if (src_pitch == dstPitch) { - memcpy(region.address + region.start, tex, - lines * src_pitch); - } else { - char *buf = region.address + region.start; - int i; - for (i = 0; i < lines; i++) { - memcpy(buf, tex, src_pitch); - buf += dstPitch; - tex += src_pitch; - } - } - - r300EmitWait(rmesa, R300_WAIT_3D); - - /* Blit to framebuffer - */ - r300EmitBlit(rmesa, - blit_format, - dstPitch, GET_START(®ion), - dstPitch | (t->tile_bits >> 16), - t->bufAddr, 0, 0, 0, done, width, lines); - - r300EmitWait(rmesa, R300_WAIT_2D); -#ifdef USER_BUFFERS - r300_mem_use(rmesa, region.buf->id); -#endif - - r300ReleaseDmaRegion(rmesa, ®ion, __FUNCTION__); - done += lines; - } - } -} - -/** - * Upload the texture image associated with texture \a t at the specified - * level at the address relative to \a start. - */ -static void r300UploadSubImage(r300ContextPtr rmesa, r300TexObjPtr t, - GLint hwlevel, - GLint x, GLint y, GLint width, GLint height, - GLuint face) -{ - struct gl_texture_image *texImage = NULL; - GLuint offset; - GLint imageWidth, imageHeight; - GLint ret; - drm_radeon_texture_t tex; - drm_radeon_tex_image_t tmp; - const int level = hwlevel + t->base.firstLevel; - - if (RADEON_DEBUG & DEBUG_TEXTURE) { - fprintf(stderr, - "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", - __FUNCTION__, (void *)t, (void *)t->base.tObj, level, - width, height, face); - } - - ASSERT(face < 6); - - /* Ensure we have a valid texture to upload */ - if ((hwlevel < 0) || (hwlevel >= R300_MAX_TEXTURE_LEVELS)) { - _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__); - return; - } - - texImage = t->base.tObj->Image[face][level]; - - if (!texImage) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: texImage %d is NULL!\n", - __FUNCTION__, level); - return; - } - if (!texImage->Data) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: image data is NULL!\n", - __FUNCTION__); - return; - } - - if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - assert(level == 0); - assert(hwlevel == 0); - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: image data is rectangular\n", - __FUNCTION__); - r300UploadRectSubImage(rmesa, t, texImage, x, y, width, height); - return; - } else if (texImage->IsClientData) { - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, - "%s: image data is in GART client storage\n", - __FUNCTION__); - r300UploadGARTClientSubImage(rmesa, t, texImage, hwlevel, x, y, - width, height); - return; - } else if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "%s: image data is in normal memory\n", - __FUNCTION__); - - imageWidth = texImage->Width; - imageHeight = texImage->Height; - - offset = t->bufAddr; - - if (RADEON_DEBUG & (DEBUG_TEXTURE | DEBUG_IOCTL)) { - GLint imageX = 0; - GLint imageY = 0; - GLint blitX = t->image[face][hwlevel].x; - GLint blitY = t->image[face][hwlevel].y; - GLint blitWidth = t->image[face][hwlevel].width; - GLint blitHeight = t->image[face][hwlevel].height; - fprintf(stderr, " upload image: %d,%d at %d,%d\n", - imageWidth, imageHeight, imageX, imageY); - fprintf(stderr, " upload blit: %d,%d at %d,%d\n", - blitWidth, blitHeight, blitX, blitY); - fprintf(stderr, " blit ofs: 0x%07x level: %d/%d\n", - (GLuint) offset, hwlevel, level); - } - - t->image[face][hwlevel].data = texImage->Data; - - /* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct. - * NOTE: we're always use a 1KB-wide blit and I8 texture format. - * We used to use 1, 2 and 4-byte texels and used to use the texture - * width to dictate the blit width - but that won't work for compressed - * textures. (Brian) - * NOTE: can't do that with texture tiling. (sroland) - */ - tex.offset = offset; - tex.image = &tmp; - /* copy (x,y,width,height,data) */ - memcpy(&tmp, &t->image[face][hwlevel], sizeof(tmp)); - - if (texImage->TexFormat->TexelBytes > 4) { - const int log2TexelBytes = - (3 + (texImage->TexFormat->TexelBytes >> 4)); - tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */ - tex.pitch = - MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / - 64, 1); - tex.height = imageHeight; - tex.width = imageWidth << log2TexelBytes; - tex.offset += (tmp.x << log2TexelBytes) & ~1023; - tmp.x = tmp.x % (1024 >> log2TexelBytes); - tmp.width = tmp.width << log2TexelBytes; - } else if (texImage->TexFormat->TexelBytes) { - /* use multi-byte upload scheme */ - tex.height = imageHeight; - tex.width = imageWidth; - switch (texImage->TexFormat->TexelBytes) { - case 1: - tex.format = RADEON_TXFORMAT_I8; - break; - case 2: - tex.format = RADEON_TXFORMAT_AI88; - break; - case 4: - tex.format = RADEON_TXFORMAT_ARGB8888; - break; - } - tex.pitch = - MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / - 64, 1); - tex.offset += tmp.x & ~1023; - tmp.x = tmp.x % 1024; - - if (t->tile_bits & R300_TXO_MICRO_TILE) { - /* need something like "tiled coordinates" ? */ - tmp.y = tmp.x / (tex.pitch * 128) * 2; - tmp.x = - tmp.x % (tex.pitch * 128) / 2 / - texImage->TexFormat->TexelBytes; - tex.pitch |= RADEON_DST_TILE_MICRO >> 22; - } else { - tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1); - } -#if 1 - if ((t->tile_bits & R300_TXO_MACRO_TILE) && - (texImage->Width * texImage->TexFormat->TexelBytes >= 256) - && ((!(t->tile_bits & R300_TXO_MICRO_TILE) - && (texImage->Height >= 8)) - || (texImage->Height >= 16))) { - /* weird: R200 disables macro tiling if mip width is smaller than 256 bytes, - OR if height is smaller than 8 automatically, but if micro tiling is active - the limit is height 16 instead ? */ - tex.pitch |= RADEON_DST_TILE_MACRO >> 22; - } -#endif - } else { - /* In case of for instance 8x8 texture (2x2 dxt blocks), - padding after the first two blocks is needed (only - with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */ - /* set tex.height to 1/4 since 1 "macropixel" (dxt-block) - has 4 real pixels. Needed so the kernel module reads - the right amount of data. */ - tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */ - tex.pitch = (R300_BLIT_WIDTH_BYTES / 64); - tex.height = (imageHeight + 3) / 4; - tex.width = (imageWidth + 3) / 4; - if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1) { - tex.width *= 8; - } else { - tex.width *= 16; - } - } - - LOCK_HARDWARE(&rmesa->radeon); - do { - ret = - drmCommandWriteRead(rmesa->radeon.dri.fd, - DRM_RADEON_TEXTURE, &tex, - sizeof(drm_radeon_texture_t)); - if (ret) { - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, - "DRM_RADEON_TEXTURE: again!\n"); - usleep(1); - } - } while (ret == -EAGAIN); - - UNLOCK_HARDWARE(&rmesa->radeon); - - if (ret) { - fprintf(stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret); - fprintf(stderr, " offset=0x%08x\n", offset); - fprintf(stderr, " image width=%d height=%d\n", - imageWidth, imageHeight); - fprintf(stderr, " blit width=%d height=%d data=%p\n", - t->image[face][hwlevel].width, - t->image[face][hwlevel].height, - t->image[face][hwlevel].data); - _mesa_exit(-1); - } -} - -/** - * Upload the texture images associated with texture \a t. This might - * require the allocation of texture memory. - * - * \param rmesa Context pointer - * \param t Texture to be uploaded - * \param face Cube map face to be uploaded. Zero for non-cube maps. - */ - -int r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr t, GLuint face) -{ - const int numLevels = t->base.lastLevel - t->base.firstLevel + 1; - - if (t->image_override) - return 0; - - if (RADEON_DEBUG & (DEBUG_TEXTURE | DEBUG_IOCTL)) { - fprintf(stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__, - (void *)rmesa->radeon.glCtx, (void *)t->base.tObj, - t->base.totalSize, t->base.firstLevel, - t->base.lastLevel); - } - - if (t->base.totalSize == 0) - return 0; - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "%s: Syncing\n", __FUNCTION__); - radeonFinish(rmesa->radeon.glCtx); - } - - LOCK_HARDWARE(&rmesa->radeon); - - if (t->base.memBlock == NULL) { - int heap; - - heap = driAllocateTexture(rmesa->texture_heaps, rmesa->nr_heaps, - (driTextureObject *) t); - if (heap == -1) { - UNLOCK_HARDWARE(&rmesa->radeon); - return -1; - } - - /* Set the base offset of the texture image */ - t->bufAddr = rmesa->radeon.radeonScreen->texOffset[heap] - + t->base.memBlock->ofs; - t->offset = t->bufAddr; - - if (!(t->base.tObj->Image[0][0]->IsClientData)) { - /* hope it's safe to add that here... */ - t->offset |= t->tile_bits; - } - } - - /* Let the world know we've used this memory recently. - */ - driUpdateTextureLRU((driTextureObject *) t); - UNLOCK_HARDWARE(&rmesa->radeon); - - /* Upload any images that are new */ - if (t->base.dirty_images[face]) { - int i; - for (i = 0; i < numLevels; i++) { - if ((t->base. - dirty_images[face] & (1 << - (i + t->base.firstLevel))) != - 0) { - r300UploadSubImage(rmesa, t, i, 0, 0, - t->image[face][i].width, - t->image[face][i].height, - face); - } - } - t->base.dirty_images[face] = 0; - } - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "%s: Syncing\n", __FUNCTION__); - radeonFinish(rmesa->radeon.glCtx); - } - - return 0; -} diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c index f6ae4b675b..6e47321246 100644 --- a/src/mesa/drivers/dri/r300/r300_texstate.c +++ b/src/mesa/drivers/dri/r300/r300_texstate.c @@ -47,7 +47,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "r300_context.h" #include "r300_state.h" #include "r300_ioctl.h" -#include "radeon_ioctl.h" +#include "radeon_mipmap_tree.h" #include "r300_tex.h" #include "r300_reg.h" @@ -117,7 +117,12 @@ static const struct tx_table { _ASSIGN(INTENSITY_FLOAT16, R300_EASY_TX_FORMAT(X, X, X, X, FL_I16)), _ASSIGN(Z16, R300_EASY_TX_FORMAT(X, X, X, X, X16)), _ASSIGN(Z24_S8, R300_EASY_TX_FORMAT(X, X, X, X, X24_Y8)), + _ASSIGN(S8_Z24, R300_EASY_TX_FORMAT(Y, Y, Y, Y, X24_Y8)), _ASSIGN(Z32, R300_EASY_TX_FORMAT(X, X, X, X, X32)), + /* EXT_texture_sRGB */ + _ASSIGN(SRGBA8, R300_EASY_TX_FORMAT(Y, Z, W, X, W8Z8Y8X8) | R300_TX_FORMAT_GAMMA), + _ASSIGN(SLA8, R300_EASY_TX_FORMAT(X, X, X, Y, Y8X8) | R300_TX_FORMAT_GAMMA), + _ASSIGN(SL8, R300_EASY_TX_FORMAT(X, X, X, ONE, X8) | R300_TX_FORMAT_GAMMA), /* *INDENT-ON* */ }; @@ -143,13 +148,12 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj) }, }; const GLuint *format; - r300TexObjPtr t; + radeonTexObjPtr t; if (!tObj) return; - t = (r300TexObjPtr) tObj->DriverData; - + t = radeon_tex_obj(tObj); switch (tObj->Image[0][tObj->BaseLevel]->TexFormat->MesaFormat) { case MESA_FORMAT_Z16: @@ -171,13 +175,13 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj) switch (tObj->DepthMode) { case GL_LUMINANCE: - t->format = format[0]; + t->pp_txformat = format[0]; break; case GL_INTENSITY: - t->format = format[1]; + t->pp_txformat = format[1]; break; case GL_ALPHA: - t->format = format[2]; + t->pp_txformat = format[2]; break; default: /* Error...which should have already been caught by higher @@ -190,406 +194,134 @@ void r300SetDepthTexMode(struct gl_texture_object *tObj) /** - * Compute sizes and fill in offset and blit information for the given - * image (determined by \p face and \p level). - * - * \param curOffset points to the offset at which the image is to be stored - * and is updated by this function according to the size of the image. - */ -static void compute_tex_image_offset( - struct gl_texture_object *tObj, - GLuint face, - GLint level, - GLint* curOffset) -{ - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; - const struct gl_texture_image* texImage; - GLuint blitWidth = R300_BLIT_WIDTH_BYTES; - GLuint texelBytes; - GLuint size; - - texImage = tObj->Image[0][level + t->base.firstLevel]; - if (!texImage) - return; - - texelBytes = texImage->TexFormat->TexelBytes; - - /* find image size in bytes */ - if (texImage->IsCompressed) { - if ((t->format & R300_TX_FORMAT_DXT1) == - R300_TX_FORMAT_DXT1) { - // fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format); - if ((texImage->Width + 3) < 8) /* width one block */ - size = texImage->CompressedSize * 4; - else if ((texImage->Width + 3) < 16) - size = texImage->CompressedSize * 2; - else - size = texImage->CompressedSize; - } else { - /* DXT3/5, 16 bytes per block */ - WARN_ONCE - ("DXT 3/5 suffers from multitexturing problems!\n"); - // fprintf(stderr,"DXT 3/5 %d\n", texImage->Width); - if ((texImage->Width + 3) < 8) - size = texImage->CompressedSize * 2; - else - size = texImage->CompressedSize; - } - } else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - size = - ((texImage->Width * texelBytes + - 63) & ~63) * texImage->Height; - blitWidth = 64 / texelBytes; - } else if (t->tile_bits & R300_TXO_MICRO_TILE) { - /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned, - though the actual offset may be different (if texture is less than - 32 bytes width) to the untiled case */ - int w = (texImage->Width * texelBytes * 2 + 31) & ~31; - size = - (w * ((texImage->Height + 1) / 2)) * - texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } else { - int w = (texImage->Width * texelBytes + 31) & ~31; - size = w * texImage->Height * texImage->Depth; - blitWidth = MAX2(texImage->Width, 64 / texelBytes); - } - assert(size > 0); - - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n", - texImage->Width, texImage->Height, - texImage->Depth, - texImage->TexFormat->TexelBytes, - texImage->InternalFormat); - - /* All images are aligned to a 32-byte offset */ - *curOffset = (*curOffset + 0x1f) & ~0x1f; - - if (texelBytes) { - /* fix x and y coords up later together with offset */ - t->image[face][level].x = *curOffset; - t->image[face][level].y = 0; - t->image[face][level].width = - MIN2(size / texelBytes, blitWidth); - t->image[face][level].height = - (size / texelBytes) / t->image[face][level].width; - } else { - t->image[face][level].x = *curOffset % R300_BLIT_WIDTH_BYTES; - t->image[face][level].y = *curOffset / R300_BLIT_WIDTH_BYTES; - t->image[face][level].width = - MIN2(size, R300_BLIT_WIDTH_BYTES); - t->image[face][level].height = size / t->image[face][level].width; - } - - if (RADEON_DEBUG & DEBUG_TEXTURE) - fprintf(stderr, - "level %d, face %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n", - level, face, texImage->Width, texImage->Height, - t->image[face][level].x, t->image[face][level].y, - t->image[face][level].width, t->image[face][level].height, - size, *curOffset); - - *curOffset += size; -} - - - -/** - * This function computes the number of bytes of storage needed for - * the given texture object (all mipmap levels, all cube faces). - * The \c image[face][level].x/y/width/height parameters for upload/blitting - * are computed here. \c filter, \c format, etc. will be set here - * too. + * Compute the cached hardware register values for the given texture object. * * \param rmesa Context pointer - * \param tObj GL texture object whose images are to be posted to - * hardware state. + * \param t the r300 texture object */ -static void r300SetTexImages(r300ContextPtr rmesa, - struct gl_texture_object *tObj) +static void setup_hardware_state(r300ContextPtr rmesa, radeonTexObj *t) { - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; - const struct gl_texture_image *baseImage = - tObj->Image[0][tObj->BaseLevel]; - GLint curOffset; - GLint i, texelBytes; - GLint numLevels; - GLint log2Width, log2Height, log2Depth; - - /* Set the hardware texture format - */ + const struct gl_texture_image *firstImage; + int firstlevel = t->mt ? t->mt->firstLevel : 0; + + firstImage = t->base.Image[0][firstlevel]; + if (!t->image_override - && VALID_FORMAT(baseImage->TexFormat->MesaFormat)) { - if (baseImage->TexFormat->BaseFormat == GL_DEPTH_COMPONENT) { - r300SetDepthTexMode(tObj); + && VALID_FORMAT(firstImage->TexFormat->MesaFormat)) { + if (firstImage->TexFormat->BaseFormat == GL_DEPTH_COMPONENT) { + r300SetDepthTexMode(&t->base); } else { - t->format = tx_table[baseImage->TexFormat->MesaFormat].format; + t->pp_txformat = tx_table[firstImage->TexFormat->MesaFormat].format; } - t->filter |= tx_table[baseImage->TexFormat->MesaFormat].filter; + t->pp_txfilter |= tx_table[firstImage->TexFormat->MesaFormat].filter; } else if (!t->image_override) { _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__); return; } - texelBytes = baseImage->TexFormat->TexelBytes; - - /* Compute which mipmap levels we really want to send to the hardware. - */ - driCalculateTextureFirstLastLevel((driTextureObject *) t); - log2Width = tObj->Image[0][t->base.firstLevel]->WidthLog2; - log2Height = tObj->Image[0][t->base.firstLevel]->HeightLog2; - log2Depth = tObj->Image[0][t->base.firstLevel]->DepthLog2; - - numLevels = t->base.lastLevel - t->base.firstLevel + 1; + if (t->image_override && t->bo) + return; - assert(numLevels <= R300_MAX_TEXTURE_LEVELS); + t->pp_txsize = (((firstImage->Width - 1) << R300_TX_WIDTHMASK_SHIFT) + | ((firstImage->Height - 1) << R300_TX_HEIGHTMASK_SHIFT) + | ((firstImage->DepthLog2) << R300_TX_DEPTHMASK_SHIFT) + | ((t->mt->lastLevel - t->mt->firstLevel) << R300_TX_MAX_MIP_LEVEL_SHIFT)); - /* Calculate mipmap offsets and dimensions for blitting (uploading) - * The idea is that we lay out the mipmap levels within a block of - * memory organized as a rectangle of width BLIT_WIDTH_BYTES. - */ t->tile_bits = 0; - /* figure out if this texture is suitable for tiling. */ -#if 0 /* Disabled for now */ - if (texelBytes) { - if ((tObj->Target != GL_TEXTURE_RECTANGLE_NV) && - /* texrect might be able to use micro tiling too in theory? */ - (baseImage->Height > 1)) { - - /* allow 32 (bytes) x 1 mip (which will use two times the space - the non-tiled version would use) max if base texture is large enough */ - if ((numLevels == 1) || - (((baseImage->Width * texelBytes / - baseImage->Height) <= 32) - && (baseImage->Width * texelBytes > 64)) - || - ((baseImage->Width * texelBytes / - baseImage->Height) <= 16)) { - t->tile_bits |= R300_TXO_MICRO_TILE; - } - } - - if (tObj->Target != GL_TEXTURE_RECTANGLE_NV) { - /* we can set macro tiling even for small textures, they will be untiled anyway */ - t->tile_bits |= R300_TXO_MACRO_TILE; - } - } -#endif - - curOffset = 0; + if (t->base.Target == GL_TEXTURE_CUBE_MAP) + t->pp_txformat |= R300_TX_FORMAT_CUBIC_MAP; + if (t->base.Target == GL_TEXTURE_3D) + t->pp_txformat |= R300_TX_FORMAT_3D; - if (tObj->Target == GL_TEXTURE_CUBE_MAP) { - ASSERT(log2Width == log2Height); - t->format |= R300_TX_FORMAT_CUBIC_MAP; - for(i = 0; i < numLevels; i++) { - GLuint face; - for(face = 0; face < 6; face++) - compute_tex_image_offset(tObj, face, i, &curOffset); - } - } else { - if (tObj->Target == GL_TEXTURE_3D) - t->format |= R300_TX_FORMAT_3D; - - for (i = 0; i < numLevels; i++) - compute_tex_image_offset(tObj, 0, i, &curOffset); - } - - /* Align the total size of texture memory block. - */ - t->base.totalSize = - (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK; - - t->size = - (((tObj->Image[0][t->base.firstLevel]->Width - - 1) << R300_TX_WIDTHMASK_SHIFT) - | ((tObj->Image[0][t->base.firstLevel]->Height - 1) << - R300_TX_HEIGHTMASK_SHIFT) - | ((tObj->Image[0][t->base.firstLevel]->DepthLog2) << - R300_TX_DEPTHMASK_SHIFT)) - | ((numLevels - 1) << R300_TX_MAX_MIP_LEVEL_SHIFT); - - t->pitch = 0; - - /* Only need to round to nearest 32 for textures, but the blitter - * requires 64-byte aligned pitches, and we may/may not need the - * blitter. NPOT only! - */ - if (baseImage->IsCompressed) { - t->pitch |= - (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63); - } else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) { - unsigned int align = (64 / texelBytes) - 1; - t->pitch |= ((tObj->Image[0][t->base.firstLevel]->Width * - texelBytes) + 63) & ~(63); - t->size |= R300_TX_SIZE_TXPITCH_EN; + if (t->base.Target == GL_TEXTURE_RECTANGLE_NV) { + unsigned int align = (64 / t->mt->bpp) - 1; + t->pp_txsize |= R300_TX_SIZE_TXPITCH_EN; if (!t->image_override) - t->pitch_reg = - (((tObj->Image[0][t->base.firstLevel]->Width) + - align) & ~align) - 1; - } else { - t->pitch |= - ((tObj->Image[0][t->base.firstLevel]->Width * - texelBytes) + 63) & ~(63); + t->pp_txpitch = ((firstImage->Width + align) & ~align) - 1; } if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { - if (tObj->Image[0][t->base.firstLevel]->Width > 2048) - t->pitch_reg |= R500_TXWIDTH_BIT11; - if (tObj->Image[0][t->base.firstLevel]->Height > 2048) - t->pitch_reg |= R500_TXHEIGHT_BIT11; + if (firstImage->Width > 2048) + t->pp_txpitch |= R500_TXWIDTH_BIT11; + if (firstImage->Height > 2048) + t->pp_txpitch |= R500_TXHEIGHT_BIT11; } } -/* ================================================================ - * Texture unit state management +/** + * Ensure the given texture is ready for rendering. + * + * Mostly this means populating the texture object's mipmap tree. */ - -static GLboolean r300EnableTexture2D(GLcontext * ctx, int unit) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; - - ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D); - - if (t->base.dirty_images[0]) { - R300_FIREVERTICES(rmesa); - - r300SetTexImages(rmesa, tObj); - r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0); - if (!t->base.memBlock && !t->image_override) - return GL_FALSE; - } - - return GL_TRUE; -} - -static GLboolean r300EnableTexture3D(GLcontext * ctx, int unit) +static GLboolean r300_validate_texture(GLcontext * ctx, struct gl_texture_object *texObj) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; + radeonTexObj *t = radeon_tex_obj(texObj); - ASSERT(tObj->Target == GL_TEXTURE_3D); - - /* r300 does not support mipmaps for 3D textures. */ - if ((tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR)) { + if (!radeon_validate_texture_miptree(ctx, texObj)) return GL_FALSE; - } - if (t->base.dirty_images[0]) { - R300_FIREVERTICES(rmesa); - r300SetTexImages(rmesa, tObj); - r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0); - if (!t->base.memBlock) - return GL_FALSE; - } + /* Configure the hardware registers (more precisely, the cached version + * of the hardware registers). */ + setup_hardware_state(rmesa, t); + t->validated = GL_TRUE; return GL_TRUE; } -static GLboolean r300EnableTextureCube(GLcontext * ctx, int unit) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; - GLuint face; - - ASSERT(tObj->Target == GL_TEXTURE_CUBE_MAP); - - if (t->base.dirty_images[0] || t->base.dirty_images[1] || - t->base.dirty_images[2] || t->base.dirty_images[3] || - t->base.dirty_images[4] || t->base.dirty_images[5]) { - /* flush */ - R300_FIREVERTICES(rmesa); - /* layout memory space, once for all faces */ - r300SetTexImages(rmesa, tObj); - } - - /* upload (per face) */ - for (face = 0; face < 6; face++) { - if (t->base.dirty_images[face]) { - r300UploadTexImages(rmesa, - (r300TexObjPtr) tObj->DriverData, - face); - } - } - - if (!t->base.memBlock) { - /* texmem alloc failed, use s/w fallback */ - return GL_FALSE; - } - - return GL_TRUE; -} - -static GLboolean r300EnableTextureRect(GLcontext * ctx, int unit) +/** + * Ensure all enabled and complete textures are uploaded along with any buffers being used. + */ +GLboolean r300ValidateBuffers(GLcontext * ctx) { r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_Current; - r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData; - - ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV); + struct radeon_renderbuffer *rrb; + int i; - if (t->base.dirty_images[0]) { - R300_FIREVERTICES(rmesa); + radeon_validate_reset_bos(&rmesa->radeon); - r300SetTexImages(rmesa, tObj); - r300UploadTexImages(rmesa, (r300TexObjPtr) tObj->DriverData, 0); - if (!t->base.memBlock && !t->image_override && - !rmesa->prefer_gart_client_texturing) - return GL_FALSE; + rrb = radeon_get_colorbuffer(&rmesa->radeon); + /* color buffer */ + if (rrb && rrb->bo) { + radeon_validate_bo(&rmesa->radeon, rrb->bo, + 0, RADEON_GEM_DOMAIN_VRAM); } - return GL_TRUE; -} - -static GLboolean r300UpdateTexture(GLcontext * ctx, int unit) -{ - r300ContextPtr rmesa = R300_CONTEXT(ctx); - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - struct gl_texture_object *tObj = texUnit->_ReallyEnabled ? - texUnit->_Current : NULL; - r300TexObjPtr t = tObj ? (r300TexObjPtr) tObj->DriverData : NULL; - - /* Fallback if there's a texture border */ - if (tObj && tObj->Image[0][tObj->BaseLevel]->Border > 0) { - tObj = NULL; - t = NULL; + /* depth buffer */ + rrb = radeon_get_depthbuffer(&rmesa->radeon); + if (rrb && rrb->bo) { + radeon_validate_bo(&rmesa->radeon, rrb->bo, + 0, RADEON_GEM_DOMAIN_VRAM); } + + for (i = 0; i < ctx->Const.MaxTextureImageUnits; ++i) { + radeonTexObj *t; - /* Update state if this is a different texture object to last - * time. - */ - if (rmesa->state.texture.unit[unit].texobj != tObj) { - if (rmesa->state.texture.unit[unit].texobj != NULL) { - r300TexObjPtr t_old = (r300TexObjPtr) rmesa->state.texture.unit[unit].texobj->DriverData; - - /* The old texture is no longer bound to this texture unit. - * Mark it as such. - */ - - t_old->base.bound &= ~(1 << unit); - } - - _mesa_reference_texobj(&rmesa->state.texture.unit[unit].texobj, tObj); + if (!ctx->Texture.Unit[i]._ReallyEnabled) + continue; - if (t) { - t->base.bound |= (1 << unit); - driUpdateTextureLRU(&t->base); /* XXX: should be locked! */ + if (!r300_validate_texture(ctx, ctx->Texture.Unit[i]._Current)) { + _mesa_warning(ctx, + "failed to validate texture for unit %d.\n", + i); } + t = radeon_tex_obj(ctx->Texture.Unit[i]._Current); + if (t->image_override && t->bo) + radeon_validate_bo(&rmesa->radeon, t->bo, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0); + + else if (t->mt->bo) + radeon_validate_bo(&rmesa->radeon, t->mt->bo, + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0); } + if (rmesa->radeon.dma.current) + radeon_validate_bo(&rmesa->radeon, rmesa->radeon.dma.current, RADEON_GEM_DOMAIN_GTT, 0); - return !t || !t->border_fallback; + return radeon_revalidate_bos(ctx); } void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname, @@ -598,78 +330,166 @@ void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname, r300ContextPtr rmesa = pDRICtx->driverPrivate; struct gl_texture_object *tObj = _mesa_lookup_texture(rmesa->radeon.glCtx, texname); - r300TexObjPtr t; + radeonTexObjPtr t = radeon_tex_obj(tObj); uint32_t pitch_val; if (!tObj) return; - t = (r300TexObjPtr) tObj->DriverData; - t->image_override = GL_TRUE; if (!offset) return; - t->offset = offset; - t->pitch_reg &= (1 << 13) -1; + t->bo = NULL; + t->override_offset = offset; + t->pp_txpitch &= (1 << 13) -1; pitch_val = pitch; switch (depth) { case 32: - t->format = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8); - t->filter |= tx_table[2].filter; + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8); + t->pp_txfilter |= tx_table[2].filter; pitch_val /= 4; break; case 24: default: - t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8); - t->filter |= tx_table[4].filter; + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8); + t->pp_txfilter |= tx_table[4].filter; pitch_val /= 4; break; case 16: - t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5); - t->filter |= tx_table[5].filter; + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5); + t->pp_txfilter |= tx_table[5].filter; pitch_val /= 2; break; } pitch_val--; - t->pitch_reg |= pitch_val; + t->pp_txpitch |= pitch_val; } -static GLboolean r300UpdateTextureUnit(GLcontext * ctx, int unit) +void r300SetTexBuffer2(__DRIcontext *pDRICtx, GLint target, GLint glx_texture_format, __DRIdrawable *dPriv) { - struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit]; - - if (texUnit->_ReallyEnabled & (TEXTURE_RECT_BIT)) { - return (r300EnableTextureRect(ctx, unit) && - r300UpdateTexture(ctx, unit)); - } else if (texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT)) { - return (r300EnableTexture2D(ctx, unit) && - r300UpdateTexture(ctx, unit)); - } else if (texUnit->_ReallyEnabled & (TEXTURE_3D_BIT)) { - return (r300EnableTexture3D(ctx, unit) && - r300UpdateTexture(ctx, unit)); - } else if (texUnit->_ReallyEnabled & (TEXTURE_CUBE_BIT)) { - return (r300EnableTextureCube(ctx, unit) && - r300UpdateTexture(ctx, unit)); - } else if (texUnit->_ReallyEnabled) { - return GL_FALSE; - } else { - return r300UpdateTexture(ctx, unit); + struct gl_texture_unit *texUnit; + struct gl_texture_object *texObj; + struct gl_texture_image *texImage; + struct radeon_renderbuffer *rb; + radeon_texture_image *rImage; + radeonContextPtr radeon; + r300ContextPtr rmesa; + struct radeon_framebuffer *rfb; + radeonTexObjPtr t; + uint32_t pitch_val; + uint32_t internalFormat, type, format; + + type = GL_BGRA; + format = GL_UNSIGNED_BYTE; + internalFormat = (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT ? 3 : 4); + + radeon = pDRICtx->driverPrivate; + rmesa = pDRICtx->driverPrivate; + + rfb = dPriv->driverPrivate; + texUnit = &radeon->glCtx->Texture.Unit[radeon->glCtx->Texture.CurrentUnit]; + texObj = _mesa_select_tex_object(radeon->glCtx, texUnit, target); + texImage = _mesa_get_tex_image(radeon->glCtx, texObj, target, 0); + + rImage = get_radeon_texture_image(texImage); + t = radeon_tex_obj(texObj); + if (t == NULL) { + return; + } + + radeon_update_renderbuffers(pDRICtx, dPriv); + /* back & depth buffer are useless free them right away */ + rb = (void*)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; + } + rb = (void*)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer; + if (rb && rb->bo) { + radeon_bo_unref(rb->bo); + rb->bo = NULL; } + rb = rfb->color_rb[0]; + if (rb->bo == NULL) { + /* Failed to BO for the buffer */ + return; + } + + _mesa_lock_texture(radeon->glCtx, texObj); + if (t->bo) { + radeon_bo_unref(t->bo); + t->bo = NULL; + } + if (rImage->bo) { + radeon_bo_unref(rImage->bo); + rImage->bo = NULL; + } + if (t->mt) { + radeon_miptree_unreference(t->mt); + t->mt = NULL; + } + if (rImage->mt) { + radeon_miptree_unreference(rImage->mt); + rImage->mt = NULL; + } + _mesa_init_teximage_fields(radeon->glCtx, target, texImage, + rb->width, rb->height, 1, 0, rb->cpp); + texImage->RowStride = rb->pitch / rb->cpp; + texImage->TexFormat = radeonChooseTextureFormat(radeon->glCtx, + internalFormat, + type, format, 0); + rImage->bo = rb->bo; + radeon_bo_ref(rImage->bo); + t->bo = rb->bo; + radeon_bo_ref(t->bo); + t->tile_bits = 0; + t->image_override = GL_TRUE; + t->override_offset = 0; + t->pp_txpitch &= (1 << 13) -1; + pitch_val = rb->pitch; + switch (rb->cpp) { + case 4: + if (glx_texture_format == GLX_TEXTURE_FORMAT_RGB_EXT) + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8); + else + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8); + t->pp_txfilter |= tx_table[2].filter; + pitch_val /= 4; + break; + case 3: + default: + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8); + t->pp_txfilter |= tx_table[4].filter; + pitch_val /= 4; + break; + case 2: + t->pp_txformat = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5); + t->pp_txfilter |= tx_table[5].filter; + pitch_val /= 2; + break; + } + pitch_val--; + t->pp_txsize = ((rb->width - 1) << R300_TX_WIDTHMASK_SHIFT) | + ((rb->height - 1) << R300_TX_HEIGHTMASK_SHIFT); + t->pp_txsize |= R300_TX_SIZE_TXPITCH_EN; + t->pp_txpitch |= pitch_val; + + if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) { + if (rb->width > 2048) + t->pp_txpitch |= R500_TXWIDTH_BIT11; + if (rb->height > 2048) + t->pp_txpitch |= R500_TXHEIGHT_BIT11; + } + t->validated = GL_TRUE; + _mesa_unlock_texture(radeon->glCtx, texObj); + return; } -void r300UpdateTextureState(GLcontext * ctx) +void r300SetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv) { - int i; - - for (i = 0; i < 8; i++) { - if (!r300UpdateTextureUnit(ctx, i)) { - _mesa_warning(ctx, - "failed to update texture state for unit %d.\n", - i); - } - } + r300SetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv); } diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c index 146daa367c..c41a8fdd62 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.c +++ b/src/mesa/drivers/dri/r300/r300_vertprog.c @@ -34,10 +34,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "shader/program.h" #include "shader/prog_instruction.h" #include "shader/prog_parameter.h" +#include "shader/prog_print.h" #include "shader/prog_statevars.h" #include "tnl/tnl.h" #include "r300_context.h" +#include "r300_state.h" /* TODO: Get rid of t_src_class call */ #define CMP_SRCS(a, b) ((a.RelAddr != b.RelAddr) || (a.Index != b.Index && \ @@ -64,7 +66,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. int u_temp_used = (VSF_MAX_FRAGMENT_TEMPS - 1) - u_temp_i; \ if((vp->num_temporaries + u_temp_used) > VSF_MAX_FRAGMENT_TEMPS) { \ WARN_ONCE("Ran out of temps, num temps %d, us %d\n", vp->num_temporaries, u_temp_used); \ - vp->native = GL_FALSE; \ + vp->error = GL_TRUE; \ } \ u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \ } while (0) @@ -214,21 +216,8 @@ static void vp_dump_inputs(struct r300_vertex_program *vp, char *caller) static unsigned long t_src_index(struct r300_vertex_program *vp, struct prog_src_register *src) { - int i; - int max_reg = -1; - if (src->File == PROGRAM_INPUT) { - if (vp->inputs[src->Index] != -1) - return vp->inputs[src->Index]; - - for (i = 0; i < VERT_ATTRIB_MAX; i++) - if (vp->inputs[i] > max_reg) - max_reg = vp->inputs[i]; - - vp->inputs[src->Index] = max_reg + 1; - - //vp_dump_inputs(vp, __FUNCTION__); - + assert(vp->inputs[src->Index] != -1); return vp->inputs[src->Index]; } else { if (src->Index < 0) { @@ -943,11 +932,17 @@ static GLuint *r300TranslateOpcodeXPD(struct r300_vertex_program *vp, static void t_inputs_outputs(struct r300_vertex_program *vp) { int i; - int cur_reg = 0; + int cur_reg; - for (i = 0; i < VERT_ATTRIB_MAX; i++) - vp->inputs[i] = -1; + cur_reg = -1; + for (i = 0; i < VERT_ATTRIB_MAX; i++) { + if (vp->key.InputsRead & (1 << i)) + vp->inputs[i] = ++cur_reg; + else + vp->inputs[i] = -1; + } + cur_reg = 0; for (i = 0; i < VERT_RESULT_MAX; i++) vp->outputs[i] = -1; @@ -961,26 +956,36 @@ static void t_inputs_outputs(struct r300_vertex_program *vp) vp->outputs[VERT_RESULT_PSIZ] = cur_reg++; } + /* If we're writing back facing colors we need to send + * four colors to make front/back face colors selection work. + * If the vertex program doesn't write all 4 colors, lets + * pretend it does by skipping output index reg so the colors + * get written into appropriate output vectors. + */ if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) { vp->outputs[VERT_RESULT_COL0] = cur_reg++; + } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) || + vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) { + cur_reg++; } if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) { - vp->outputs[VERT_RESULT_COL1] = - vp->outputs[VERT_RESULT_COL0] + 1; - cur_reg = vp->outputs[VERT_RESULT_COL1] + 1; + vp->outputs[VERT_RESULT_COL1] = cur_reg++; + } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) || + vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) { + cur_reg++; } if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) { - vp->outputs[VERT_RESULT_BFC0] = - vp->outputs[VERT_RESULT_COL0] + 2; - cur_reg = vp->outputs[VERT_RESULT_BFC0] + 2; + vp->outputs[VERT_RESULT_BFC0] = cur_reg++; + } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) { + cur_reg++; } if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) { - vp->outputs[VERT_RESULT_BFC1] = - vp->outputs[VERT_RESULT_COL0] + 3; - cur_reg = vp->outputs[VERT_RESULT_BFC1] + 1; + vp->outputs[VERT_RESULT_BFC1] = cur_reg++; + } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) { + cur_reg++; } for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) { @@ -1007,14 +1012,13 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, struct prog_src_register src[3]; vp->pos_end = 0; /* Not supported yet */ - vp->program.length = 0; - /*vp->num_temporaries=mesa_vp->Base.NumTemporaries; */ + vp->hw_code.length = 0; vp->translated = GL_TRUE; - vp->native = GL_TRUE; + vp->error = GL_FALSE; t_inputs_outputs(vp); - for (inst = vp->program.body.i; vpi->Opcode != OPCODE_END; + for (inst = vp->hw_code.body.d; vpi->Opcode != OPCODE_END; vpi++, inst += 4) { FREE_TEMPS(); @@ -1176,38 +1180,15 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp, &u_temp_i); break; default: - assert(0); + vp->error = GL_TRUE; break; } } - /* Some outputs may be artificially added, to match the inputs - of the fragment program. Blank the outputs here. */ - for (i = 0; i < VERT_RESULT_MAX; i++) { - if (vp->key.OutputsAdded & (1 << i)) { - inst[0] = PVS_OP_DST_OPERAND(VE_ADD, - GL_FALSE, - GL_FALSE, - vp->outputs[i], - VSF_FLAG_ALL, - PVS_DST_REG_OUT); - inst[1] = __CONST(0, SWIZZLE_ZERO); - inst[2] = __CONST(0, SWIZZLE_ZERO); - inst[3] = __CONST(0, SWIZZLE_ZERO); - inst += 4; - } + vp->hw_code.length = (inst - vp->hw_code.body.d); + if (vp->hw_code.length >= VSF_MAX_FRAGMENT_LENGTH) { + vp->error = GL_TRUE; } - - vp->program.length = (inst - vp->program.body.i); - if (vp->program.length >= VSF_MAX_FRAGMENT_LENGTH) { - vp->program.length = 0; - vp->native = GL_FALSE; - } -#if 0 - fprintf(stderr, "hw program:\n"); - for (i = 0; i < vp->program.length; i++) - fprintf(stderr, "%08x\n", vp->program.body.d[i]); -#endif } /* DP4 version seems to trigger some hw peculiarity */ @@ -1386,6 +1367,49 @@ static struct r300_vertex_program *build_program(struct r300_vertex_program_key pos_as_texcoord(vp, &mesa_vp->Base); } + if (RADEON_DEBUG & DEBUG_VERTS) { + fprintf(stderr, "Vertex program after native rewrite:\n"); + _mesa_print_program(&mesa_vp->Base); + fflush(stdout); + } + + /* Some outputs may be artificially added, to match the inputs of the fragment program. + * Issue 16 of vertex program spec says that all vertex attributes that are unwritten by + * vertex program are undefined, so just use MOV [vertex_result], CONST[0] + */ + { + int i, count = 0; + for (i = 0; i < VERT_RESULT_MAX; ++i) { + if (vp->key.OutputsAdded & (1 << i)) { + ++count; + } + } + + if (count > 0) { + struct prog_instruction *inst; + + _mesa_insert_instructions(&mesa_vp->Base, mesa_vp->Base.NumInstructions - 1, count); + inst = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions - 1 - count]; + + for (i = 0; i < VERT_RESULT_MAX; ++i) { + if (vp->key.OutputsAdded & (1 << i)) { + inst->Opcode = OPCODE_MOV; + + inst->DstReg.File = PROGRAM_OUTPUT; + inst->DstReg.Index = i; + inst->DstReg.WriteMask = WRITEMASK_XYZW; + inst->DstReg.CondMask = COND_TR; + + inst->SrcReg[0].File = PROGRAM_CONSTANT; + inst->SrcReg[0].Index = 0; + inst->SrcReg[0].Swizzle = SWIZZLE_XYZW; + + ++inst; + } + } + } + } + assert(mesa_vp->Base.NumInstructions); vp->num_temporaries = mesa_vp->Base.NumTemporaries; r300TranslateVertexShader(vp, mesa_vp->Base.Instructions); @@ -1432,7 +1456,12 @@ void r300SelectVertexShader(r300ContextPtr r300) wpos_idx = i; } - add_outputs(&wanted_key, VERT_RESULT_HPOS); + if (vpc->mesa_program.IsPositionInvariant) { + wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS); + wanted_key.OutputsWritten |= (1 << VERT_RESULT_HPOS); + } else { + add_outputs(&wanted_key, VERT_RESULT_HPOS); + } if (InputsRead & FRAG_BIT_COL0) { add_outputs(&wanted_key, VERT_RESULT_COL0); @@ -1442,27 +1471,103 @@ void r300SelectVertexShader(r300ContextPtr r300) add_outputs(&wanted_key, VERT_RESULT_COL1); } + if (InputsRead & FRAG_BIT_FOGC) { + add_outputs(&wanted_key, VERT_RESULT_FOGC); + } + for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { if (InputsRead & (FRAG_BIT_TEX0 << i)) { add_outputs(&wanted_key, VERT_RESULT_TEX0 + i); } } - if (vpc->mesa_program.IsPositionInvariant) { - /* we wan't position don't we ? */ - wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS); - } - for (vp = vpc->progs; vp; vp = vp->next) if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key)) == 0) { r300->selected_vp = vp; return; } - //_mesa_print_program(&vpc->mesa_program.Base); + + if (RADEON_DEBUG & DEBUG_VERTS) { + fprintf(stderr, "Initial vertex program:\n"); + _mesa_print_program(&vpc->mesa_program.Base); + fflush(stdout); + } vp = build_program(&wanted_key, &vpc->mesa_program, wpos_idx); vp->next = vpc->progs; vpc->progs = vp; r300->selected_vp = vp; } + +#define bump_vpu_count(ptr, new_count) do { \ + drm_r300_cmd_header_t* _p=((drm_r300_cmd_header_t*)(ptr)); \ + int _nc=(new_count)/4; \ + assert(_nc < 256); \ + if(_nc>_p->vpu.count)_p->vpu.count=_nc; \ + } while(0) + +static void r300EmitVertexProgram(r300ContextPtr r300, int dest, struct r300_vertex_shader_hw_code *code) +{ + int i; + + assert((code->length > 0) && (code->length % 4 == 0)); + + switch ((dest >> 8) & 0xf) { + case 0: + R300_STATECHANGE(r300, vpi); + for (i = 0; i < code->length; i++) + r300->hw.vpi.cmd[R300_VPI_INSTR_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]); + bump_vpu_count(r300->hw.vpi.cmd, code->length + 4 * (dest & 0xff)); + break; + case 2: + R300_STATECHANGE(r300, vpp); + for (i = 0; i < code->length; i++) + r300->hw.vpp.cmd[R300_VPP_PARAM_0 + i + 4 * (dest & 0xff)] = (code->body.d[i]); + bump_vpu_count(r300->hw.vpp.cmd, code->length + 4 * (dest & 0xff)); + break; + case 4: + R300_STATECHANGE(r300, vps); + for (i = 0; i < code->length; i++) + r300->hw.vps.cmd[1 + i + 4 * (dest & 0xff)] = (code->body.d[i]); + bump_vpu_count(r300->hw.vps.cmd, code->length + 4 * (dest & 0xff)); + break; + default: + fprintf(stderr, "%s:%s don't know how to handle dest %04x\n", __FILE__, __FUNCTION__, dest); + _mesa_exit(-1); + } +} + +void r300SetupVertexProgram(r300ContextPtr rmesa) +{ + GLcontext *ctx = rmesa->radeon.glCtx; + struct r300_vertex_program *prog = rmesa->selected_vp; + int inst_count = 0; + int param_count = 0; + + /* Reset state, in case we don't use something */ + ((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0; + ((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0; + ((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0; + + R300_STATECHANGE(rmesa, vpp); + param_count = r300VertexProgUpdateParams(ctx, + (struct r300_vertex_program_cont *) + ctx->VertexProgram._Current, + (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]); + bump_vpu_count(rmesa->hw.vpp.cmd, param_count); + param_count /= 4; + + r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->hw_code)); + inst_count = (prog->hw_code.length / 4) - 1; + + r300VapCntl(rmesa, _mesa_bitcount(prog->key.InputsRead), + _mesa_bitcount(prog->key.OutputsWritten), prog->num_temporaries); + + R300_STATECHANGE(rmesa, pvs); + rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = (0 << R300_PVS_FIRST_INST_SHIFT) | (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) | + (inst_count << R300_PVS_LAST_INST_SHIFT); + + rmesa->hw.pvs.cmd[R300_PVS_CNTL_2] = (0 << R300_PVS_CONST_BASE_OFFSET_SHIFT) | (param_count << R300_PVS_MAX_CONST_ADDR_SHIFT); + rmesa->hw.pvs.cmd[R300_PVS_CNTL_3] = (inst_count << R300_PVS_LAST_VTX_SRC_INST_SHIFT); +} diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.h b/src/mesa/drivers/dri/r300/r300_vertprog.h index 2f35f02bc8..b552e3fb1b 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.h +++ b/src/mesa/drivers/dri/r300/r300_vertprog.h @@ -32,4 +32,6 @@ #endif +void r300SetupVertexProgram(r300ContextPtr rmesa); + #endif diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.c b/src/mesa/drivers/dri/r300/r500_fragprog.c index 300559d0b4..4d58cf2162 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog.c +++ b/src/mesa/drivers/dri/r300/r500_fragprog.c @@ -27,10 +27,6 @@ #include "r500_fragprog.h" -#include "radeon_nqssadce.h" -#include "radeon_program_alu.h" - - static void reset_srcreg(struct prog_src_register* reg) { _mesa_bzero(reg, sizeof(*reg)); @@ -58,12 +54,12 @@ static struct prog_src_register shadow_ambient(struct gl_program *program, int t * - introduce a temporary register when write masks are needed * */ -static GLboolean transform_TEX( +GLboolean r500_transform_TEX( struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data) { - struct r500_fragment_program_compiler *compiler = - (struct r500_fragment_program_compiler*)data; + struct r300_fragment_program_compiler *compiler = + (struct r300_fragment_program_compiler*)data; struct prog_instruction inst = *orig_inst; struct prog_instruction* tgt; GLboolean destredirect = GL_FALSE; @@ -188,121 +184,7 @@ static GLboolean transform_TEX( return GL_TRUE; } - -static void update_params(r300ContextPtr r300, struct r500_fragment_program *fp) -{ - struct gl_fragment_program *mp = &fp->mesa_program; - - /* Ask Mesa nicely to fill in ParameterValues for us */ - if (mp->Base.Parameters) - _mesa_load_state_parameters(r300->radeon.glCtx, mp->Base.Parameters); -} - - -/** - * Transform the program to support fragment.position. - * - * Introduce a small fragment at the start of the program that will be - * the only code that directly reads the FRAG_ATTRIB_WPOS input. - * All other code pieces that reference that input will be rewritten - * to read from a newly allocated temporary. - * - * \todo if/when r5xx supports the radeon_program architecture, this is a - * likely candidate for code sharing. - */ -static void insert_WPOS_trailer(struct r500_fragment_program_compiler *compiler) -{ - GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead; - - if (!(InputsRead & FRAG_BIT_WPOS)) - return; - - static gl_state_index tokens[STATE_LENGTH] = { - STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0 - }; - struct prog_instruction *fpi; - GLuint window_index; - int i = 0; - GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY); - - _mesa_insert_instructions(compiler->program, 0, 3); - fpi = compiler->program->Instructions; - - /* perspective divide */ - fpi[i].Opcode = OPCODE_RCP; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_W; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW; - i++; - - fpi[i].Opcode = OPCODE_MUL; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_INPUT; - fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS; - fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW; - - fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[1].Index = tempregi; - fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW; - i++; - - /* viewport transformation */ - window_index = _mesa_add_state_reference(compiler->program->Parameters, tokens); - - fpi[i].Opcode = OPCODE_MAD; - - fpi[i].DstReg.File = PROGRAM_TEMPORARY; - fpi[i].DstReg.Index = tempregi; - fpi[i].DstReg.WriteMask = WRITEMASK_XYZ; - fpi[i].DstReg.CondMask = COND_TR; - - fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[0].Index = tempregi; - fpi[i].SrcReg[0].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[1].Index = window_index; - fpi[i].SrcReg[1].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - - fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR; - fpi[i].SrcReg[2].Index = window_index; - fpi[i].SrcReg[2].Swizzle = - MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO); - i++; - - for (; i < compiler->program->NumInstructions; ++i) { - int reg; - for (reg = 0; reg < 3; reg++) { - if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT && - fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) { - fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY; - fpi[i].SrcReg[reg].Index = tempregi; - } - } - } -} - - -static void nqssadce_init(struct nqssadce_state* s) -{ - s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW; - s->Outputs[FRAG_RESULT_DEPTH].Sourced = WRITEMASK_W; -} - -static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg) +GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg) { GLuint relevant; int i; @@ -314,22 +196,20 @@ static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg) if (reg.Abs) return GL_FALSE; + if (opcode == OPCODE_KIL && (reg.Swizzle != SWIZZLE_NOOP || reg.Negate != NEGATE_NONE)) + return GL_FALSE; + if (reg.Negate) reg.Negate ^= NEGATE_XYZW; - if (opcode == OPCODE_KIL) { - if (reg.Swizzle != SWIZZLE_NOOP) - return GL_FALSE; - } else { - for(i = 0; i < 4; ++i) { - GLuint swz = GET_SWZ(reg.Swizzle, i); - if (swz == SWIZZLE_NIL) { - reg.Negate &= ~(1 << i); - continue; - } - if (swz >= 4) - return GL_FALSE; + for(i = 0; i < 4; ++i) { + GLuint swz = GET_SWZ(reg.Swizzle, i); + if (swz == SWIZZLE_NIL) { + reg.Negate &= ~(1 << i); + continue; } + if (swz >= 4) + return GL_FALSE; } if (reg.Negate) @@ -367,8 +247,7 @@ static GLboolean is_native_swizzle(GLuint opcode, struct prog_src_register reg) * The only thing we *cannot* do in an ALU instruction is per-component * negation. Therefore, we split the MOV into two instructions when necessary. */ -static void nqssadce_build_swizzle(struct nqssadce_state *s, - struct prog_dst_register dst, struct prog_src_register src) +void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src) { struct prog_instruction *inst; GLuint negatebase[2] = { 0, 0 }; @@ -392,130 +271,12 @@ static void nqssadce_build_swizzle(struct nqssadce_state *s, inst->DstReg = dst; inst->DstReg.WriteMask = negatebase[i]; inst->SrcReg[0] = src; + inst->SrcReg[0].Negate = (i == 0) ? NEGATE_NONE : NEGATE_XYZW; inst++; s->IP++; } } -static GLuint build_dtm(GLuint depthmode) -{ - switch(depthmode) { - default: - case GL_LUMINANCE: return 0; - case GL_INTENSITY: return 1; - case GL_ALPHA: return 2; - } -} - -static GLuint build_func(GLuint comparefunc) -{ - return comparefunc - GL_NEVER; -} - - -/** - * Collect all external state that is relevant for compiling the given - * fragment program. - */ -static void build_state( - r300ContextPtr r300, - struct r500_fragment_program *fp, - struct r500_fragment_program_external_state *state) -{ - int unit; - - _mesa_bzero(state, sizeof(*state)); - - for(unit = 0; unit < 16; ++unit) { - if (fp->mesa_program.Base.ShadowSamplers & (1 << unit)) { - struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current; - - state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode); - state->unit[unit].texture_compare_func = build_func(tex->CompareFunc); - } - } -} - -static void dump_program(struct r500_fragment_program_code *code); - -void r500TranslateFragmentShader(r300ContextPtr r300, - struct r500_fragment_program *fp) -{ - struct r500_fragment_program_external_state state; - - build_state(r300, fp, &state); - if (_mesa_memcmp(&fp->state, &state, sizeof(state))) { - /* TODO: cache compiled programs */ - fp->translated = GL_FALSE; - _mesa_memcpy(&fp->state, &state, sizeof(state)); - } - - if (!fp->translated) { - struct r500_fragment_program_compiler compiler; - - compiler.r300 = r300; - compiler.fp = fp; - compiler.code = &fp->code; - compiler.program = _mesa_clone_program(r300->radeon.glCtx, &fp->mesa_program.Base); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: Initial program:\n"); - _mesa_print_program(compiler.program); - } - - insert_WPOS_trailer(&compiler); - - struct radeon_program_transformation transformations[] = { - { &transform_TEX, &compiler }, - { &radeonTransformALU, 0 }, - { &radeonTransformDeriv, 0 }, - { &radeonTransformTrigScale, 0 } - }; - radeonLocalTransform(r300->radeon.glCtx, compiler.program, - 4, transformations); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: after native rewrite:\n"); - _mesa_print_program(compiler.program); - } - - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadce_init, - .IsNativeSwizzle = &is_native_swizzle, - .BuildSwizzle = &nqssadce_build_swizzle, - .RewriteDepthOut = GL_TRUE - }; - radeonNqssaDce(r300->radeon.glCtx, compiler.program, &nqssadce); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - _mesa_printf("Compiler: after NqSSA-DCE:\n"); - _mesa_print_program(compiler.program); - } - - fp->translated = r500FragmentProgramEmit(&compiler); - - /* Subtle: Rescue any parameters that have been added during transformations */ - _mesa_free_parameter_list(fp->mesa_program.Base.Parameters); - fp->mesa_program.Base.Parameters = compiler.program->Parameters; - compiler.program->Parameters = 0; - - _mesa_reference_program(r300->radeon.glCtx, &compiler.program, 0); - - r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM | - _NEW_PROGRAM_CONSTANTS); - - if (RADEON_DEBUG & DEBUG_PIXEL) { - if (fp->translated) { - _mesa_printf("Machine-readable code:\n"); - dump_program(&fp->code); - } - } - - } - - update_params(r300, fp); - -} static char *toswiz(int swiz_val) { switch(swiz_val) { @@ -614,9 +375,9 @@ static char *to_texop(int val) return NULL; } -static void dump_program(struct r500_fragment_program_code *code) +void r500FragmentProgramDump(union rX00_fragment_program_code *c) { - + struct r500_fragment_program_code *code = &c->r500; fprintf(stderr, "R500 Fragment Program:\n--------\n"); int n; diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.h b/src/mesa/drivers/dri/r300/r500_fragprog.h index 1e45538f80..1179bf6607 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog.h +++ b/src/mesa/drivers/dri/r300/r500_fragprog.h @@ -33,30 +33,20 @@ #ifndef __R500_FRAGPROG_H_ #define __R500_FRAGPROG_H_ -#include "main/glheader.h" -#include "main/macros.h" -#include "main/enums.h" #include "shader/prog_parameter.h" -#include "shader/prog_print.h" -#include "shader/program.h" #include "shader/prog_instruction.h" #include "r300_context.h" -#include "r300_state.h" -#include "radeon_program.h" +#include "radeon_nqssadce.h" -struct r500_fragment_program; +extern GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler); -extern void r500TranslateFragmentShader(r300ContextPtr r300, - struct r500_fragment_program *fp); +extern void r500FragmentProgramDump(union rX00_fragment_program_code *c); -struct r500_fragment_program_compiler { - r300ContextPtr r300; - struct r500_fragment_program *fp; - struct r500_fragment_program_code *code; - struct gl_program *program; -}; +extern GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg); -extern GLboolean r500FragmentProgramEmit(struct r500_fragment_program_compiler *compiler); +extern void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src); + +extern GLboolean r500_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data); #endif diff --git a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/r500_fragprog_emit.c index 4631235f0d..30f4514897 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/r500_fragprog_emit.c @@ -49,8 +49,8 @@ #define PROG_CODE \ - struct r500_fragment_program_compiler *c = (struct r500_fragment_program_compiler*)data; \ - struct r500_fragment_program_code *code = c->code + struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \ + struct r500_fragment_program_code *code = &c->code->r500 #define error(fmt, args...) do { \ fprintf(stderr, "%s::%s(): " fmt "\n", \ @@ -72,7 +72,7 @@ static GLboolean emit_const(void *data, GLuint file, GLuint idx, GLuint *hwindex } if (*hwindex >= code->const_nr) { - if (*hwindex >= PFS_NUM_CONST_REGS) { + if (*hwindex >= R500_PFS_NUM_CONST_REGS) { error("Out of hw constants!\n"); return GL_FALSE; } @@ -299,9 +299,9 @@ static const struct radeon_pair_handler pair_handler = { .MaxHwTemps = 128 }; -GLboolean r500FragmentProgramEmit(struct r500_fragment_program_compiler *compiler) +GLboolean r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler) { - struct r500_fragment_program_code *code = compiler->code; + struct r500_fragment_program_code *code = &compiler->code->r500; _mesa_bzero(code, sizeof(*code)); code->max_temp_idx = 1; diff --git a/src/mesa/drivers/dri/r300/radeon_context.c b/src/mesa/drivers/dri/r300/radeon_context.c deleted file mode 100644 index 5267fe9a77..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_context.c +++ /dev/null @@ -1,330 +0,0 @@ -/* -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/** - * \file radeon_context.c - * Common context initialization. - * - * \author Keith Whitwell <keith@tungstengraphics.com> - */ - -#include <dlfcn.h> - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/context.h" -#include "main/state.h" -#include "main/matrix.h" -#include "main/framebuffer.h" - -#include "drivers/common/driverfuncs.h" -#include "swrast/swrast.h" - -#include "radeon_screen.h" -#include "radeon_ioctl.h" -#include "radeon_macros.h" -#include "radeon_reg.h" - -#include "radeon_state.h" -#include "r300_state.h" - -#include "utils.h" -#include "vblank.h" -#include "xmlpool.h" /* for symbolic values of enum-type options */ - -#define DRIVER_DATE "20060815" - - -/* Return various strings for glGetString(). - */ -static const GLubyte *radeonGetString(GLcontext * ctx, GLenum name) -{ - radeonContextPtr radeon = RADEON_CONTEXT(ctx); - static char buffer[128]; - - switch (name) { - case GL_VENDOR: - if (IS_R300_CLASS(radeon->radeonScreen)) - return (GLubyte *) "DRI R300 Project"; - else - return (GLubyte *) "Tungsten Graphics, Inc."; - - case GL_RENDERER: - { - unsigned offset; - GLuint agp_mode = (radeon->radeonScreen->card_type==RADEON_CARD_PCI) ? 0 : - radeon->radeonScreen->AGPMode; - const char* chipname; - - if (IS_R300_CLASS(radeon->radeonScreen)) - chipname = "R300"; - else - chipname = "R200"; - - offset = driGetRendererString(buffer, chipname, DRIVER_DATE, - agp_mode); - - if (IS_R300_CLASS(radeon->radeonScreen)) { - sprintf(&buffer[offset], " %sTCL", - (radeon->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) - ? "" : "NO-"); - } else { - sprintf(&buffer[offset], " %sTCL", - !(radeon->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE) - ? "" : "NO-"); - } - - return (GLubyte *) buffer; - } - - default: - return NULL; - } -} - -/* Initialize the driver's misc functions. - */ -static void radeonInitDriverFuncs(struct dd_function_table *functions) -{ - functions->GetString = radeonGetString; -} - - -/** - * Create and initialize all common fields of the context, - * including the Mesa context itself. - */ -GLboolean radeonInitContext(radeonContextPtr radeon, - struct dd_function_table* functions, - const __GLcontextModes * glVisual, - __DRIcontextPrivate * driContextPriv, - void *sharedContextPrivate) -{ - __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv; - radeonScreenPtr screen = (radeonScreenPtr) (sPriv->private); - GLcontext* ctx; - GLcontext* shareCtx; - int fthrottle_mode; - - /* Fill in additional standard functions. */ - radeonInitDriverFuncs(functions); - - radeon->radeonScreen = screen; - /* Allocate and initialize the Mesa context */ - if (sharedContextPrivate) - shareCtx = ((radeonContextPtr)sharedContextPrivate)->glCtx; - else - shareCtx = NULL; - radeon->glCtx = _mesa_create_context(glVisual, shareCtx, - functions, (void *)radeon); - if (!radeon->glCtx) - return GL_FALSE; - - ctx = radeon->glCtx; - driContextPriv->driverPrivate = radeon; - - /* DRI fields */ - radeon->dri.context = driContextPriv; - radeon->dri.screen = sPriv; - radeon->dri.drawable = NULL; - radeon->dri.readable = NULL; - radeon->dri.hwContext = driContextPriv->hHWContext; - radeon->dri.hwLock = &sPriv->pSAREA->lock; - radeon->dri.fd = sPriv->fd; - radeon->dri.drmMinor = sPriv->drm_version.minor; - - radeon->sarea = (drm_radeon_sarea_t *) ((GLubyte *) sPriv->pSAREA + - screen->sarea_priv_offset); - - /* Setup IRQs */ - fthrottle_mode = driQueryOptioni(&radeon->optionCache, "fthrottle_mode"); - radeon->iw.irq_seq = -1; - radeon->irqsEmitted = 0; - radeon->do_irqs = (fthrottle_mode == DRI_CONF_FTHROTTLE_IRQS && - radeon->radeonScreen->irq); - - radeon->do_usleeps = (fthrottle_mode == DRI_CONF_FTHROTTLE_USLEEPS); - - if (!radeon->do_irqs) - fprintf(stderr, - "IRQ's not enabled, falling back to %s: %d %d\n", - radeon->do_usleeps ? "usleeps" : "busy waits", - fthrottle_mode, radeon->radeonScreen->irq); - - (*sPriv->systemTime->getUST) (&radeon->swap_ust); - - return GL_TRUE; -} - - -/** - * Cleanup common context fields. - * Called by r200DestroyContext/r300DestroyContext - */ -void radeonCleanupContext(radeonContextPtr radeon) -{ - /* _mesa_destroy_context() might result in calls to functions that - * depend on the DriverCtx, so don't set it to NULL before. - * - * radeon->glCtx->DriverCtx = NULL; - */ - - /* free the Mesa context */ - _mesa_destroy_context(radeon->glCtx); - - if (radeon->state.scissor.pClipRects) { - FREE(radeon->state.scissor.pClipRects); - radeon->state.scissor.pClipRects = 0; - } -} - - -/** - * Swap front and back buffer. - */ -void radeonSwapBuffers(__DRIdrawablePrivate * dPriv) -{ - if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { - radeonContextPtr radeon; - GLcontext *ctx; - - radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - ctx = radeon->glCtx; - - if (ctx->Visual.doubleBufferMode) { - _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ - if (radeon->doPageFlip) { - radeonPageFlip(dPriv); - } else { - radeonCopyBuffer(dPriv, NULL); - } - } - } else { - /* XXX this shouldn't be an error but we can't handle it for now */ - _mesa_problem(NULL, "%s: drawable has no context!", - __FUNCTION__); - } -} - -void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, - int x, int y, int w, int h ) -{ - if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) { - radeonContextPtr radeon; - GLcontext *ctx; - - radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - ctx = radeon->glCtx; - - if (ctx->Visual.doubleBufferMode) { - drm_clip_rect_t rect; - rect.x1 = x + dPriv->x; - rect.y1 = (dPriv->h - y - h) + dPriv->y; - rect.x2 = rect.x1 + w; - rect.y2 = rect.y1 + h; - _mesa_notifySwapBuffers(ctx); /* flush pending rendering comands */ - radeonCopyBuffer(dPriv, &rect); - } - } else { - /* XXX this shouldn't be an error but we can't handle it for now */ - _mesa_problem(NULL, "%s: drawable has no context!", - __FUNCTION__); - } -} - -/* Force the context `c' to be the current context and associate with it - * buffer `b'. - */ -GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv, - __DRIdrawablePrivate * driDrawPriv, - __DRIdrawablePrivate * driReadPriv) -{ - if (driContextPriv) { - radeonContextPtr radeon = - (radeonContextPtr) driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx %p\n", __FUNCTION__, - radeon->glCtx); - - if (radeon->dri.drawable != driDrawPriv) { - if (driDrawPriv->swap_interval == (unsigned)-1) { - driDrawPriv->vblFlags = - (radeon->radeonScreen->irq != 0) - ? driGetDefaultVBlankFlags(&radeon-> - optionCache) - : VBLANK_FLAG_NO_IRQ; - - driDrawableInitVBlank(driDrawPriv); - } - } - - radeon->dri.readable = driReadPriv; - - if (radeon->dri.drawable != driDrawPriv || - radeon->lastStamp != driDrawPriv->lastStamp) { - radeon->dri.drawable = driDrawPriv; - - radeonSetCliprects(radeon); - r300UpdateViewportOffset(radeon->glCtx); - } - - _mesa_make_current(radeon->glCtx, - (GLframebuffer *) driDrawPriv-> - driverPrivate, - (GLframebuffer *) driReadPriv-> - driverPrivate); - - _mesa_update_state(radeon->glCtx); - - radeonUpdatePageFlipping(radeon); - } else { - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx is null\n", __FUNCTION__); - _mesa_make_current(0, 0, 0); - } - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "End %s\n", __FUNCTION__); - return GL_TRUE; -} - -/* Force the context `c' to be unbound from its buffer. - */ -GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv) -{ - radeonContextPtr radeon = (radeonContextPtr) driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_DRI) - fprintf(stderr, "%s ctx %p\n", __FUNCTION__, - radeon->glCtx); - - return GL_TRUE; -} - diff --git a/src/mesa/drivers/dri/r300/radeon_context.h b/src/mesa/drivers/dri/r300/radeon_context.h index 47cbc22a72..250570f6b8 100644 --- a/src/mesa/drivers/dri/r300/radeon_context.h +++ b/src/mesa/drivers/dri/r300/radeon_context.h @@ -49,20 +49,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include "drm.h" #include "dri_util.h" -struct radeon_context; -typedef struct radeon_context radeonContextRec; -typedef struct radeon_context *radeonContextPtr; - -/* Rasterizing fallbacks */ -/* See correponding strings in r200_swtcl.c */ -#define RADEON_FALLBACK_TEXTURE 0x0001 -#define RADEON_FALLBACK_DRAW_BUFFER 0x0002 -#define RADEON_FALLBACK_STENCIL 0x0004 -#define RADEON_FALLBACK_RENDER_MODE 0x0008 -#define RADEON_FALLBACK_BLEND_EQ 0x0010 -#define RADEON_FALLBACK_BLEND_FUNC 0x0020 -#define RADEON_FALLBACK_DISABLE 0x0040 -#define RADEON_FALLBACK_BORDER_MODE 0x0080 +#include "radeon_screen.h" #if R200_MERGED extern void radeonFallback(GLcontext * ctx, GLuint bit, GLboolean mode); @@ -79,155 +66,11 @@ extern void radeonFallback(GLcontext * ctx, GLuint bit, GLboolean mode); /* TCL fallbacks */ extern void radeonTclFallback(GLcontext * ctx, GLuint bit, GLboolean mode); -#define RADEON_TCL_FALLBACK_RASTER 0x0001 /* rasterization */ -#define RADEON_TCL_FALLBACK_UNFILLED 0x0002 /* unfilled tris */ -#define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE 0x0004 /* twoside tris */ -#define RADEON_TCL_FALLBACK_MATERIAL 0x0008 /* material in vb */ -#define RADEON_TCL_FALLBACK_TEXGEN_0 0x0010 /* texgen, unit 0 */ -#define RADEON_TCL_FALLBACK_TEXGEN_1 0x0020 /* texgen, unit 1 */ -#define RADEON_TCL_FALLBACK_TEXGEN_2 0x0040 /* texgen, unit 2 */ -#define RADEON_TCL_FALLBACK_TEXGEN_3 0x0080 /* texgen, unit 3 */ -#define RADEON_TCL_FALLBACK_TEXGEN_4 0x0100 /* texgen, unit 4 */ -#define RADEON_TCL_FALLBACK_TEXGEN_5 0x0200 /* texgen, unit 5 */ -#define RADEON_TCL_FALLBACK_TCL_DISABLE 0x0400 /* user disable */ -#define RADEON_TCL_FALLBACK_BITMAP 0x0800 /* draw bitmap with points */ -#define RADEON_TCL_FALLBACK_VERTEX_PROGRAM 0x1000 /* vertex program active */ - #if R200_MERGED #define TCL_FALLBACK( ctx, bit, mode ) radeonTclFallback( ctx, bit, mode ) #else #define TCL_FALLBACK( ctx, bit, mode ) ; #endif -struct radeon_dri_mirror { - __DRIcontextPrivate *context; /* DRI context */ - __DRIscreenPrivate *screen; /* DRI screen */ - /** - * DRI drawable bound to this context for drawing. - */ - __DRIdrawablePrivate *drawable; - - /** - * DRI drawable bound to this context for reading. - */ - __DRIdrawablePrivate *readable; - - drm_context_t hwContext; - drm_hw_lock_t *hwLock; - int fd; - int drmMinor; -}; - -/** - * Derived state for internal purposes. - */ -struct radeon_scissor_state { - drm_clip_rect_t rect; - GLboolean enabled; - - GLuint numClipRects; /* Cliprects active */ - GLuint numAllocedClipRects; /* Cliprects available */ - drm_clip_rect_t *pClipRects; -}; - -struct radeon_colorbuffer_state { - GLuint clear; - GLint drawOffset, drawPitch; -}; - -struct radeon_state { - struct radeon_colorbuffer_state color; - struct radeon_scissor_state scissor; -}; - -/** - * Common per-context variables shared by R200 and R300. - * R200- and R300-specific code "derive" their own context from this - * structure. - */ -struct radeon_context { - GLcontext *glCtx; /* Mesa context */ - radeonScreenPtr radeonScreen; /* Screen private DRI data */ - - /* Fallback state */ - GLuint Fallback; - GLuint TclFallback; - - /* Page flipping */ - GLuint doPageFlip; - - /* Drawable, cliprect and scissor information */ - GLuint numClipRects; /* Cliprects for the draw buffer */ - drm_clip_rect_t *pClipRects; - unsigned int lastStamp; - GLboolean lost_context; - drm_radeon_sarea_t *sarea; /* Private SAREA data */ - - /* Mirrors of some DRI state */ - struct radeon_dri_mirror dri; - - /* Busy waiting */ - GLuint do_usleeps; - GLuint do_irqs; - GLuint irqsEmitted; - drm_radeon_irq_wait_t iw; - - /* buffer swap */ - int64_t swap_ust; - int64_t swap_missed_ust; - - GLuint swap_count; - GLuint swap_missed_count; - - /* Derived state */ - struct radeon_state state; - - /* Configuration cache - */ - driOptionCache optionCache; -}; - -#define RADEON_CONTEXT(glctx) ((radeonContextPtr)(ctx->DriverCtx)) - -extern void radeonSwapBuffers(__DRIdrawablePrivate * dPriv); -extern void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv, - int x, int y, int w, int h); -extern GLboolean radeonInitContext(radeonContextPtr radeon, - struct dd_function_table *functions, - const __GLcontextModes * glVisual, - __DRIcontextPrivate * driContextPriv, - void *sharedContextPrivate); -extern void radeonCleanupContext(radeonContextPtr radeon); -extern GLboolean radeonMakeCurrent(__DRIcontextPrivate * driContextPriv, - __DRIdrawablePrivate * driDrawPriv, - __DRIdrawablePrivate * driReadPriv); -extern GLboolean radeonUnbindContext(__DRIcontextPrivate * driContextPriv); - -/* ================================================================ - * Debugging: - */ -#define DO_DEBUG 1 - -#if DO_DEBUG -extern int RADEON_DEBUG; -#else -#define RADEON_DEBUG 0 -#endif - -#define DEBUG_TEXTURE 0x0001 -#define DEBUG_STATE 0x0002 -#define DEBUG_IOCTL 0x0004 -#define DEBUG_PRIMS 0x0008 -#define DEBUG_VERTS 0x0010 -#define DEBUG_FALLBACKS 0x0020 -#define DEBUG_VFMT 0x0040 -#define DEBUG_CODEGEN 0x0080 -#define DEBUG_VERBOSE 0x0100 -#define DEBUG_DRI 0x0200 -#define DEBUG_DMA 0x0400 -#define DEBUG_SANITY 0x0800 -#define DEBUG_SYNC 0x1000 -#define DEBUG_PIXEL 0x2000 -#define DEBUG_MEMORY 0x4000 #endif /* __RADEON_CONTEXT_H__ */ diff --git a/src/mesa/drivers/dri/r300/radeon_ioctl.c b/src/mesa/drivers/dri/r300/radeon_ioctl.c deleted file mode 100644 index f042a7b943..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_ioctl.c +++ /dev/null @@ -1,396 +0,0 @@ -/* -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Keith Whitwell <keith@tungstengraphics.com> - */ - -#include <sched.h> -#include <errno.h> - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/macros.h" -#include "main/context.h" -#include "swrast/swrast.h" -#include "r300_context.h" -#include "radeon_ioctl.h" -#include "r300_ioctl.h" -#include "r300_state.h" -#include "radeon_reg.h" - -#include "drirenderbuffer.h" -#include "vblank.h" - -static void radeonWaitForIdle(radeonContextPtr radeon); - -/* ================================================================ - * SwapBuffers with client-side throttling - */ - -static uint32_t radeonGetLastFrame(radeonContextPtr radeon) -{ - drm_radeon_getparam_t gp; - int ret; - uint32_t frame = 0; - - gp.param = RADEON_PARAM_LAST_FRAME; - gp.value = (int *)&frame; - ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM, - &gp, sizeof(gp)); - if (ret) { - fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, - ret); - exit(1); - } - - return frame; -} - -uint32_t radeonGetAge(radeonContextPtr radeon) -{ - drm_radeon_getparam_t gp; - int ret; - uint32_t age = 0; - - gp.param = RADEON_PARAM_LAST_CLEAR; - gp.value = (int *)&age; - ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM, - &gp, sizeof(gp)); - if (ret) { - fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, - ret); - exit(1); - } - - return age; -} - -static void radeonEmitIrqLocked(radeonContextPtr radeon) -{ - drm_radeon_irq_emit_t ie; - int ret; - - ie.irq_seq = &radeon->iw.irq_seq; - ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_IRQ_EMIT, - &ie, sizeof(ie)); - if (ret) { - fprintf(stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__, - ret); - exit(1); - } -} - -static void radeonWaitIrq(radeonContextPtr radeon) -{ - int ret; - - do { - ret = drmCommandWrite(radeon->dri.fd, DRM_RADEON_IRQ_WAIT, - &radeon->iw, sizeof(radeon->iw)); - } while (ret && (errno == EINTR || errno == EBUSY)); - - if (ret) { - fprintf(stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, - ret); - exit(1); - } -} - -static void radeonWaitForFrameCompletion(radeonContextPtr radeon) -{ - drm_radeon_sarea_t *sarea = radeon->sarea; - - if (radeon->do_irqs) { - if (radeonGetLastFrame(radeon) < sarea->last_frame) { - if (!radeon->irqsEmitted) { - while (radeonGetLastFrame(radeon) < - sarea->last_frame) ; - } else { - UNLOCK_HARDWARE(radeon); - radeonWaitIrq(radeon); - LOCK_HARDWARE(radeon); - } - radeon->irqsEmitted = 10; - } - - if (radeon->irqsEmitted) { - radeonEmitIrqLocked(radeon); - radeon->irqsEmitted--; - } - } else { - while (radeonGetLastFrame(radeon) < sarea->last_frame) { - UNLOCK_HARDWARE(radeon); - if (radeon->do_usleeps) - DO_USLEEP(1); - LOCK_HARDWARE(radeon); - } - } -} - -/* Copy the back color buffer to the front color buffer. - */ -void radeonCopyBuffer(__DRIdrawablePrivate * dPriv, - const drm_clip_rect_t * rect) -{ - radeonContextPtr radeon; - GLint nbox, i, ret; - GLboolean missed_target; - int64_t ust; - __DRIscreenPrivate *psp = dPriv->driScreenPriv; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_IOCTL) { - fprintf(stderr, "\n%s( %p )\n\n", __FUNCTION__, - (void *)radeon->glCtx); - } - - r300Flush(radeon->glCtx); - - LOCK_HARDWARE(radeon); - - /* Throttle the frame rate -- only allow one pending swap buffers - * request at a time. - */ - radeonWaitForFrameCompletion(radeon); - if (!rect) - { - UNLOCK_HARDWARE(radeon); - driWaitForVBlank(dPriv, &missed_target); - LOCK_HARDWARE(radeon); - } - - nbox = dPriv->numClipRects; /* must be in locked region */ - - for (i = 0; i < nbox;) { - GLint nr = MIN2(i + RADEON_NR_SAREA_CLIPRECTS, nbox); - drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = radeon->sarea->boxes; - GLint n = 0; - - for ( ; i < nr ; i++ ) { - - *b = box[i]; - - if (rect) - { - if (rect->x1 > b->x1) - b->x1 = rect->x1; - if (rect->y1 > b->y1) - b->y1 = rect->y1; - if (rect->x2 < b->x2) - b->x2 = rect->x2; - if (rect->y2 < b->y2) - b->y2 = rect->y2; - - if (b->x1 >= b->x2 || b->y1 >= b->y2) - continue; - } - - b++; - n++; - } - radeon->sarea->nbox = n; - - if (!n) - continue; - - ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_SWAP); - - if (ret) { - fprintf(stderr, "DRM_RADEON_SWAP: return = %d\n", - ret); - UNLOCK_HARDWARE(radeon); - exit(1); - } - } - - UNLOCK_HARDWARE(radeon); - if (!rect) - { - ((r300ContextPtr)radeon)->hw.all_dirty = GL_TRUE; - - radeon->swap_count++; - (*psp->systemTime->getUST) (&ust); - if (missed_target) { - radeon->swap_missed_count++; - radeon->swap_missed_ust = ust - radeon->swap_ust; - } - - radeon->swap_ust = ust; - - sched_yield(); - } -} - -void radeonPageFlip(__DRIdrawablePrivate * dPriv) -{ - radeonContextPtr radeon; - GLint ret; - GLboolean missed_target; - __DRIscreenPrivate *psp = dPriv->driScreenPriv; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - - if (RADEON_DEBUG & DEBUG_IOCTL) { - fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__, - radeon->sarea->pfCurrentPage); - } - - r300Flush(radeon->glCtx); - LOCK_HARDWARE(radeon); - - if (!dPriv->numClipRects) { - UNLOCK_HARDWARE(radeon); - usleep(10000); /* throttle invisible client 10ms */ - return; - } - - /* Need to do this for the perf box placement: - */ - { - drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = radeon->sarea->boxes; - b[0] = box[0]; - radeon->sarea->nbox = 1; - } - - /* Throttle the frame rate -- only allow a few pending swap buffers - * request at a time. - */ - radeonWaitForFrameCompletion(radeon); - UNLOCK_HARDWARE(radeon); - driWaitForVBlank(dPriv, &missed_target); - if (missed_target) { - radeon->swap_missed_count++; - (void)(*psp->systemTime->getUST) (&radeon->swap_missed_ust); - } - LOCK_HARDWARE(radeon); - - ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_FLIP); - - UNLOCK_HARDWARE(radeon); - - if (ret) { - fprintf(stderr, "DRM_RADEON_FLIP: return = %d\n", ret); - exit(1); - } - - radeon->swap_count++; - (void)(*psp->systemTime->getUST) (&radeon->swap_ust); - - driFlipRenderbuffers(radeon->glCtx->WinSysDrawBuffer, - radeon->sarea->pfCurrentPage); - - if (radeon->sarea->pfCurrentPage == 1) { - radeon->state.color.drawOffset = radeon->radeonScreen->frontOffset; - radeon->state.color.drawPitch = radeon->radeonScreen->frontPitch; - } else { - radeon->state.color.drawOffset = radeon->radeonScreen->backOffset; - radeon->state.color.drawPitch = radeon->radeonScreen->backPitch; - } - - if (IS_R300_CLASS(radeon->radeonScreen)) { - r300ContextPtr r300 = (r300ContextPtr)radeon; - R300_STATECHANGE(r300, cb); - r300->hw.cb.cmd[R300_CB_OFFSET] = r300->radeon.state.color.drawOffset + - r300->radeon.radeonScreen->fbLocation; - r300->hw.cb.cmd[R300_CB_PITCH] = r300->radeon.state.color.drawPitch; - - if (r300->radeon.radeonScreen->cpp == 4) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_ARGB8888; - else - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_FORMAT_RGB565; - - if (r300->radeon.sarea->tiling_enabled) - r300->hw.cb.cmd[R300_CB_PITCH] |= R300_COLOR_TILE_ENABLE; - } -} - -void radeonWaitForIdleLocked(radeonContextPtr radeon) -{ - int ret; - int i = 0; - - do { - ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_CP_IDLE); - if (ret) - DO_USLEEP(1); - } while (ret && ++i < 100); - - if (ret < 0) { - UNLOCK_HARDWARE(radeon); - fprintf(stderr, "Error: R300 timed out... exiting\n"); - exit(-1); - } -} - -static void radeonWaitForIdle(radeonContextPtr radeon) -{ - LOCK_HARDWARE(radeon); - radeonWaitForIdleLocked(radeon); - UNLOCK_HARDWARE(radeon); -} - -void radeonFlush(GLcontext * ctx) -{ - radeonContextPtr radeon = RADEON_CONTEXT(ctx); - - if (IS_R300_CLASS(radeon->radeonScreen)) - r300Flush(ctx); -} - - -/* Make sure all commands have been sent to the hardware and have - * completed processing. - */ -void radeonFinish(GLcontext * ctx) -{ - radeonContextPtr radeon = RADEON_CONTEXT(ctx); - - radeonFlush(ctx); - - if (radeon->do_irqs) { - LOCK_HARDWARE(radeon); - radeonEmitIrqLocked(radeon); - UNLOCK_HARDWARE(radeon); - radeonWaitIrq(radeon); - } else - radeonWaitForIdle(radeon); -} diff --git a/src/mesa/drivers/dri/r300/radeon_ioctl.h b/src/mesa/drivers/dri/r300/radeon_ioctl.h deleted file mode 100644 index 3add775b82..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_ioctl.h +++ /dev/null @@ -1,57 +0,0 @@ -/* -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Keith Whitwell <keith@tungstengraphics.com> - */ - -#ifndef __RADEON_IOCTL_H__ -#define __RADEON_IOCTL_H__ - -#include "main/simple_list.h" -#include "radeon_dri.h" -#include "radeon_lock.h" - -#include "xf86drm.h" -#include "drm.h" -#if 0 -#include "r200context.h" -#endif -#include "radeon_drm.h" - -extern void radeonCopyBuffer(__DRIdrawablePrivate * drawable, - const drm_clip_rect_t * rect); -extern void radeonPageFlip(__DRIdrawablePrivate * drawable); -extern void radeonFlush(GLcontext * ctx); -extern void radeonFinish(GLcontext * ctx); -extern void radeonWaitForIdleLocked(radeonContextPtr radeon); -extern uint32_t radeonGetAge(radeonContextPtr radeon); - -#endif /* __RADEON_IOCTL_H__ */ diff --git a/src/mesa/drivers/dri/r300/radeon_lock.c b/src/mesa/drivers/dri/r300/radeon_lock.c deleted file mode 100644 index 4f47afd5dc..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_lock.c +++ /dev/null @@ -1,137 +0,0 @@ -/************************************************************************** - -Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and - VA Linux Systems Inc., Fremont, California. -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Gareth Hughes <gareth@valinux.com> - * Keith Whitwell <keith@tungstengraphics.com> - * Kevin E. Martin <martin@valinux.com> - */ - -#include "radeon_lock.h" -#include "radeon_ioctl.h" -#include "radeon_state.h" -#include "r300_context.h" -#include "r300_state.h" - -#include "main/framebuffer.h" - -#include "drirenderbuffer.h" - -#if DEBUG_LOCKING -char *prevLockFile = NULL; -int prevLockLine = 0; -#endif - -/* Turn on/off page flipping according to the flags in the sarea: - */ -void radeonUpdatePageFlipping(radeonContextPtr rmesa) -{ - int use_back; - - rmesa->doPageFlip = rmesa->sarea->pfState; - if (rmesa->glCtx->WinSysDrawBuffer) { - driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer, - rmesa->sarea->pfCurrentPage); - r300UpdateDrawBuffer(rmesa->glCtx); - } - - use_back = rmesa->glCtx->DrawBuffer ? - (rmesa->glCtx->DrawBuffer->_ColorDrawBufferIndexes[0] == - BUFFER_BACK_LEFT) : 1; - use_back ^= (rmesa->sarea->pfCurrentPage == 1); - - if (use_back) { - rmesa->state.color.drawOffset = - rmesa->radeonScreen->backOffset; - rmesa->state.color.drawPitch = rmesa->radeonScreen->backPitch; - } else { - rmesa->state.color.drawOffset = - rmesa->radeonScreen->frontOffset; - rmesa->state.color.drawPitch = - rmesa->radeonScreen->frontPitch; - } -} - -/* Update the hardware state. This is called if another context has - * grabbed the hardware lock, which includes the X server. This - * function also updates the driver's window state after the X server - * moves, resizes or restacks a window -- the change will be reflected - * in the drawable position and clip rects. Since the X server grabs - * the hardware lock when it changes the window state, this routine will - * automatically be called after such a change. - */ -void radeonGetLock(radeonContextPtr rmesa, GLuint flags) -{ - __DRIdrawablePrivate *const drawable = rmesa->dri.drawable; - __DRIdrawablePrivate *const readable = rmesa->dri.readable; - __DRIscreenPrivate *sPriv = rmesa->dri.screen; - drm_radeon_sarea_t *sarea = rmesa->sarea; - r300ContextPtr r300 = (r300ContextPtr) rmesa; - - assert(drawable != NULL); - - drmGetLock(rmesa->dri.fd, rmesa->dri.hwContext, flags); - - /* The window might have moved, so we might need to get new clip - * rects. - * - * NOTE: This releases and regrabs the hw lock to allow the X server - * to respond to the DRI protocol request for new drawable info. - * Since the hardware state depends on having the latest drawable - * clip rects, all state checking must be done _after_ this call. - */ - DRI_VALIDATE_DRAWABLE_INFO(sPriv, drawable); - if (drawable != readable) { - DRI_VALIDATE_DRAWABLE_INFO(sPriv, readable); - } - - if (rmesa->lastStamp != drawable->lastStamp) { - radeonUpdatePageFlipping(rmesa); - radeonSetCliprects(rmesa); - r300UpdateViewportOffset(rmesa->glCtx); - driUpdateFramebufferSize(rmesa->glCtx, drawable); - } - - if (sarea->ctx_owner != rmesa->dri.hwContext) { - int i; - - sarea->ctx_owner = rmesa->dri.hwContext; - for (i = 0; i < r300->nr_heaps; i++) { - DRI_AGE_TEXTURES(r300->texture_heaps[i]); - } - } - - rmesa->lost_context = GL_TRUE; -} diff --git a/src/mesa/drivers/dri/r300/radeon_lock.h b/src/mesa/drivers/dri/r300/radeon_lock.h deleted file mode 100644 index a344837f47..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_lock.h +++ /dev/null @@ -1,115 +0,0 @@ -/************************************************************************** - -Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and - VA Linux Systems Inc., Fremont, California. -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Gareth Hughes <gareth@valinux.com> - * Keith Whitwell <keith@tungstengraphics.com> - * Kevin E. Martin <martin@valinux.com> - */ - -#ifndef __RADEON_LOCK_H__ -#define __RADEON_LOCK_H__ - -#include "radeon_context.h" - -extern void radeonGetLock(radeonContextPtr rmesa, GLuint flags); -extern void radeonUpdatePageFlipping(radeonContextPtr rmesa); - -/* Turn DEBUG_LOCKING on to find locking conflicts. - */ -#define DEBUG_LOCKING 0 - -#if DEBUG_LOCKING -extern char *prevLockFile; -extern int prevLockLine; - -#define DEBUG_LOCK() \ - do { \ - prevLockFile = (__FILE__); \ - prevLockLine = (__LINE__); \ - } while (0) - -#define DEBUG_RESET() \ - do { \ - prevLockFile = 0; \ - prevLockLine = 0; \ - } while (0) - -#define DEBUG_CHECK_LOCK() \ - do { \ - if (prevLockFile) { \ - fprintf(stderr, \ - "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n", \ - prevLockFile, prevLockLine, __FILE__, __LINE__); \ - exit(1); \ - } \ - } while (0) - -#else - -#define DEBUG_LOCK() -#define DEBUG_RESET() -#define DEBUG_CHECK_LOCK() - -#endif - -/* - * !!! We may want to separate locks from locks with validation. This - * could be used to improve performance for those things commands that - * do not do any drawing !!! - */ - -/* Lock the hardware and validate our state. - */ -#define LOCK_HARDWARE( rmesa ) \ - do { \ - char __ret = 0; \ - DEBUG_CHECK_LOCK(); \ - DRM_CAS((rmesa)->dri.hwLock, (rmesa)->dri.hwContext, \ - (DRM_LOCK_HELD | (rmesa)->dri.hwContext), __ret); \ - if (__ret) \ - radeonGetLock((rmesa), 0); \ - DEBUG_LOCK(); \ - } while (0) - -#define UNLOCK_HARDWARE( rmesa ) \ - do { \ - DRM_UNLOCK((rmesa)->dri.fd, \ - (rmesa)->dri.hwLock, \ - (rmesa)->dri.hwContext); \ - DEBUG_RESET(); \ - } while (0) - -#endif /* __RADEON_LOCK_H__ */ diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/radeon_program_pair.c index 2e21f7bf66..906d36e522 100644 --- a/src/mesa/drivers/dri/r300/radeon_program_pair.c +++ b/src/mesa/drivers/dri/r300/radeon_program_pair.c @@ -35,7 +35,7 @@ #include "radeon_program_pair.h" -#include "radeon_context.h" +#include "radeon_common.h" #include "shader/prog_print.h" @@ -47,7 +47,6 @@ struct pair_state_instruction { GLuint IsTex:1; /**< Is a texture instruction */ - GLuint IsOutput:1; /**< Is output instruction */ GLuint NeedRGB:1; /**< Needs the RGB ALU */ GLuint NeedAlpha:1; /**< Needs the Alpha ALU */ GLuint IsTranscendent:1; /**< Is a special transcendent instruction */ @@ -124,7 +123,6 @@ struct pair_state { GLboolean Debug; GLboolean Verbose; void *UserData; - GLubyte NumKillInsts; /** * Translate Mesa registers to hardware registers @@ -151,11 +149,6 @@ struct pair_state { struct pair_state_instruction *ReadyTEX; /** - * Linked list of deferred instructions - */ - struct pair_state_instruction *DeferredInsts; - - /** * Pool of @ref reg_value structures for fast allocation. */ struct reg_value *ValuePool; @@ -238,9 +231,7 @@ static void instruction_ready(struct pair_state *s, int ip) if (s->Verbose) _mesa_printf("instruction_ready(%i)\n", ip); - if (s->NumKillInsts > 0 && pairinst->IsOutput) - add_pairinst_to_list(&s->DeferredInsts, pairinst); - else if (pairinst->IsTex) + if (pairinst->IsTex) add_pairinst_to_list(&s->ReadyTEX, pairinst); else if (!pairinst->NeedAlpha) add_pairinst_to_list(&s->ReadyRGB, pairinst); @@ -348,8 +339,6 @@ static void classify_instruction(struct pair_state *s, error("Unknown opcode %d\n", inst->Opcode); break; } - - pairinst->IsOutput = (inst->DstReg.File == PROGRAM_OUTPUT); } @@ -613,16 +602,14 @@ static void emit_all_tex(struct pair_state *s) struct prog_instruction *inst = s->Program->Instructions + ip; commit_instruction(s, ip); - if (inst->Opcode == OPCODE_KIL) - --s->NumKillInsts; - else + if (inst->Opcode != OPCODE_KIL) inst->DstReg.Index = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index); - inst->SrcReg[0].Index = get_hw_reg(s, inst->SrcReg[0].File, inst->SrcReg[0].Index); if (s->Debug) { _mesa_printf(" "); _mesa_print_instruction(inst); + fflush(stdout); } s->Error = s->Error || !s->Handler->EmitTex(s->UserData, inst); } @@ -875,17 +862,6 @@ static void emit_alu(struct pair_state *s) s->Error = s->Error || !s->Handler->EmitPaired(s->UserData, &pair); } -static GLubyte countKillInsts(struct gl_program *prog) -{ - GLubyte i, count = 0; - - for (i = 0; i < prog->NumInstructions; ++i) { - if (prog->Instructions[i].Opcode == OPCODE_KIL) - ++count; - } - - return count; -} GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, const struct radeon_pair_handler* handler, void *userdata) @@ -899,7 +875,6 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, s.UserData = userdata; s.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE; s.Verbose = GL_FALSE && s.Debug; - s.NumKillInsts = countKillInsts(program); s.Instructions = (struct pair_state_instruction*)_mesa_calloc( sizeof(struct pair_state_instruction)*s.Program->NumInstructions); @@ -918,21 +893,6 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program, if (s.ReadyTEX) emit_all_tex(&s); - if (!s.NumKillInsts) { - struct pair_state_instruction *pairinst = s.DeferredInsts; - while (pairinst) { - if (!pairinst->NeedAlpha) - add_pairinst_to_list(&s.ReadyRGB, pairinst); - else if (!pairinst->NeedRGB) - add_pairinst_to_list(&s.ReadyAlpha, pairinst); - else - add_pairinst_to_list(&s.ReadyFullALU, pairinst); - - pairinst = pairinst->NextReady; - } - s.DeferredInsts = NULL; - } - while(s.ReadyFullALU || s.ReadyRGB || s.ReadyAlpha) emit_alu(&s); } diff --git a/src/mesa/drivers/dri/r300/radeon_span.c b/src/mesa/drivers/dri/r300/radeon_span.c deleted file mode 100644 index 16f9fb99e6..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_span.c +++ /dev/null @@ -1,349 +0,0 @@ -/************************************************************************** - -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. -Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and - VA Linux Systems Inc., Fremont, California. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Kevin E. Martin <martin@valinux.com> - * Gareth Hughes <gareth@valinux.com> - * Keith Whitwell <keith@tungstengraphics.com> - * - */ - -#include "main/glheader.h" -#include "swrast/swrast.h" - -#include "r300_state.h" -#include "radeon_ioctl.h" -#include "r300_ioctl.h" -#include "radeon_span.h" - -#include "drirenderbuffer.h" - -#define DBG 0 - -/* - * Note that all information needed to access pixels in a renderbuffer - * should be obtained through the gl_renderbuffer parameter, not per-context - * information. - */ -#define LOCAL_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ - const GLuint bottom = dPriv->h - 1; \ - GLubyte *buf = (GLubyte *) drb->flippedData \ - + (dPriv->y * drb->flippedPitch + dPriv->x) * drb->cpp; \ - GLuint p; \ - (void) p; - -#define LOCAL_DEPTH_VARS \ - driRenderbuffer *drb = (driRenderbuffer *) rb; \ - const __DRIdrawablePrivate *dPriv = drb->dPriv; \ - const GLuint bottom = dPriv->h - 1; \ - GLuint xo = dPriv->x; \ - GLuint yo = dPriv->y; \ - GLubyte *buf = (GLubyte *) drb->Base.Data; - -#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS - -#define Y_FLIP(Y) (bottom - (Y)) - -#define HW_LOCK() - -#define HW_UNLOCK() - -/* ================================================================ - * Color buffer - */ - -/* 16 bit, RGB565 color spanline and pixel functions - */ -#define SPANTMP_PIXEL_FMT GL_RGB -#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5 - -#define TAG(x) radeon##x##_RGB565 -#define TAG2(x,y) radeon##x##_RGB565##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 2) -#include "spantmp2.h" - -/* 32 bit, ARGB8888 color spanline and pixel functions - */ -#define SPANTMP_PIXEL_FMT GL_BGRA -#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV - -#define TAG(x) radeon##x##_ARGB8888 -#define TAG2(x,y) radeon##x##_ARGB8888##y -#define GET_PTR(X,Y) (buf + ((Y) * drb->flippedPitch + (X)) * 4) -#include "spantmp2.h" - -/* ================================================================ - * Depth buffer - */ - -/* The Radeon family has depth tiling on all the time, so we have to convert - * the x,y coordinates into the memory bus address (mba) in the same - * manner as the engine. In each case, the linear block address (ba) - * is calculated, and then wired with x and y to produce the final - * memory address. - * The chip will do address translation on its own if the surface registers - * are set up correctly. It is not quite enough to get it working with hyperz - * too... - */ - -static GLuint radeon_mba_z32(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 4 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0..1] = 0 */ - -#ifdef COMPILE_R300 - ba = (y / 8) * (pitch / 8) + (x / 8); -#else - ba = (y / 16) * (pitch / 16) + (x / 16); -#endif - - address |= (x & 0x7) << 2; /* a[2..4] = x[0..2] */ - address |= (y & 0x3) << 5; /* a[5..6] = y[0..1] */ - address |= (((x & 0x10) >> 2) ^ (y & 0x4)) << 5; /* a[7] = x[4] ^ y[2] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= (((x & 0x8) << 1) ^ (y & 0x10)) << 7; /* a[11] = x[3] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - -static INLINE GLuint -radeon_mba_z16(const driRenderbuffer * drb, GLint x, GLint y) -{ - GLuint pitch = drb->pitch; - if (drb->depthHasSurface) { - return 2 * (x + y * pitch); - } else { - GLuint ba, address = 0; /* a[0] = 0 */ - - ba = (y / 16) * (pitch / 32) + (x / 32); - - address |= (x & 0x7) << 1; /* a[1..3] = x[0..2] */ - address |= (y & 0x7) << 4; /* a[4..6] = y[0..2] */ - address |= (x & 0x8) << 4; /* a[7] = x[3] */ - address |= (ba & 0x3) << 8; /* a[8..9] = ba[0..1] */ - address |= (y & 0x8) << 7; /* a[10] = y[3] */ - address |= ((x & 0x10) ^ (y & 0x10)) << 7; /* a[11] = x[4] ^ y[4] */ - address |= (ba & ~0x3) << 10; /* a[12..] = ba[2..] */ - - return address; - } -} - -/* 16-bit depth buffer functions - */ -#define VALUE_TYPE GLushort - -#define WRITE_DEPTH( _x, _y, d ) \ - *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )) = d; - -#define READ_DEPTH( d, _x, _y ) \ - d = *(GLushort *)(buf + radeon_mba_z16( drb, _x + xo, _y + yo )); - -#define TAG(x) radeon##x##_z16 -#include "depthtmp.h" - -/* 24 bit depth, 8 bit stencil depthbuffer functions - * - * Careful: It looks like the R300 uses ZZZS byte order while the R200 - * uses SZZZ for 24 bit depth, 8 bit stencil mode. - */ -#define VALUE_TYPE GLuint - -#ifdef COMPILE_R300 -#define WRITE_DEPTH( _x, _y, d ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - tmp &= 0x000000ff; \ - tmp |= ((d << 8) & 0xffffff00); \ - *(GLuint *)(buf + offset) = tmp; \ -} while (0) -#else -#define WRITE_DEPTH( _x, _y, d ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - tmp &= 0xff000000; \ - tmp |= ((d) & 0x00ffffff); \ - *(GLuint *)(buf + offset) = tmp; \ -} while (0) -#endif - -#ifdef COMPILE_R300 -#define READ_DEPTH( d, _x, _y ) \ - do { \ - d = (*(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0xffffff00) >> 8; \ - }while(0) -#else -#define READ_DEPTH( d, _x, _y ) \ - d = *(GLuint *)(buf + radeon_mba_z32( drb, _x + xo, \ - _y + yo )) & 0x00ffffff; -#endif - -#define TAG(x) radeon##x##_z24_s8 -#include "depthtmp.h" - -/* ================================================================ - * Stencil buffer - */ - -/* 24 bit depth, 8 bit stencil depthbuffer functions - */ -#ifdef COMPILE_R300 -#define WRITE_STENCIL( _x, _y, d ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - tmp &= 0xffffff00; \ - tmp |= (d) & 0xff; \ - *(GLuint *)(buf + offset) = tmp; \ -} while (0) -#else -#define WRITE_STENCIL( _x, _y, d ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - tmp &= 0x00ffffff; \ - tmp |= (((d) & 0xff) << 24); \ - *(GLuint *)(buf + offset) = tmp; \ -} while (0) -#endif - -#ifdef COMPILE_R300 -#define READ_STENCIL( d, _x, _y ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - d = tmp & 0x000000ff; \ -} while (0) -#else -#define READ_STENCIL( d, _x, _y ) \ -do { \ - GLuint offset = radeon_mba_z32( drb, _x + xo, _y + yo ); \ - GLuint tmp = *(GLuint *)(buf + offset); \ - d = (tmp & 0xff000000) >> 24; \ -} while (0) -#endif - -#define TAG(x) radeon##x##_z24_s8 -#include "stenciltmp.h" - -/* Move locking out to get reasonable span performance (10x better - * than doing this in HW_LOCK above). WaitForIdle() is the main - * culprit. - */ - -static void radeonSpanRenderStart(GLcontext * ctx) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); -#ifdef COMPILE_R300 - r300ContextPtr r300 = (r300ContextPtr) rmesa; - R300_FIREVERTICES(r300); -#else - RADEON_FIREVERTICES(rmesa); -#endif - LOCK_HARDWARE(rmesa); - radeonWaitForIdleLocked(rmesa); - - /* Read the first pixel in the frame buffer. This should - * be a noop, right? In fact without this conform fails as reading - * from the framebuffer sometimes produces old results -- the - * on-card read cache gets mixed up and doesn't notice that the - * framebuffer has been updated. - * - * Note that we should probably be reading some otherwise unused - * region of VRAM, otherwise we might get incorrect results when - * reading pixels from the top left of the screen. - * - * I found this problem on an R420 with glean's texCube test. - * Note that the R200 span code also *writes* the first pixel in the - * framebuffer, but I've found this to be unnecessary. - * -- Nicolai Hähnle, June 2008 - */ - { - int p; - driRenderbuffer *drb = - (driRenderbuffer *) ctx->WinSysDrawBuffer->_ColorDrawBuffers[0]; - volatile int *buf = - (volatile int *)(rmesa->dri.screen->pFB + drb->offset); - p = *buf; - } -} - -static void radeonSpanRenderFinish(GLcontext * ctx) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - _swrast_flush(ctx); - UNLOCK_HARDWARE(rmesa); -} - -void radeonInitSpanFuncs(GLcontext * ctx) -{ - struct swrast_device_driver *swdd = - _swrast_GetDeviceDriverReference(ctx); - swdd->SpanRenderStart = radeonSpanRenderStart; - swdd->SpanRenderFinish = radeonSpanRenderFinish; -} - -/** - * Plug in the Get/Put routines for the given driRenderbuffer. - */ -void radeonSetSpanFunctions(driRenderbuffer * drb, const GLvisual * vis) -{ - if (drb->Base.InternalFormat == GL_RGBA) { - if (vis->redBits == 5 && vis->greenBits == 6 - && vis->blueBits == 5) { - radeonInitPointers_RGB565(&drb->Base); - } else { - radeonInitPointers_ARGB8888(&drb->Base); - } - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) { - radeonInitDepthPointers_z16(&drb->Base); - } else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) { - radeonInitDepthPointers_z24_s8(&drb->Base); - } else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) { - radeonInitStencilPointers_z24_s8(&drb->Base); - } -} diff --git a/src/mesa/drivers/dri/r300/radeon_state.c b/src/mesa/drivers/dri/r300/radeon_state.c deleted file mode 100644 index c401da6c54..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_state.c +++ /dev/null @@ -1,244 +0,0 @@ -/************************************************************************** - -Copyright (C) The Weather Channel, Inc. 2002. All Rights Reserved. - -The Weather Channel (TM) funded Tungsten Graphics to develop the -initial release of the Radeon 8500 driver under the XFree86 license. -This notice must be preserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Keith Whitwell <keith@tungstengraphics.com> - */ - -#include "main/glheader.h" -#include "main/imports.h" -#include "main/api_arrayelt.h" -#include "main/enums.h" -#include "main/framebuffer.h" -#include "main/colormac.h" -#include "main/light.h" - -#include "swrast/swrast.h" -#include "vbo/vbo.h" -#include "tnl/tnl.h" -#include "tnl/t_pipeline.h" -#include "swrast_setup/swrast_setup.h" - -#include "radeon_ioctl.h" -#include "radeon_state.h" -#include "r300_ioctl.h" - - -/* ============================================================= - * Scissoring - */ - -static GLboolean intersect_rect(drm_clip_rect_t * out, - drm_clip_rect_t * a, drm_clip_rect_t * b) -{ - *out = *a; - if (b->x1 > out->x1) - out->x1 = b->x1; - if (b->y1 > out->y1) - out->y1 = b->y1; - if (b->x2 < out->x2) - out->x2 = b->x2; - if (b->y2 < out->y2) - out->y2 = b->y2; - if (out->x1 >= out->x2) - return GL_FALSE; - if (out->y1 >= out->y2) - return GL_FALSE; - return GL_TRUE; -} - -void radeonRecalcScissorRects(radeonContextPtr radeon) -{ - drm_clip_rect_t *out; - int i; - - /* Grow cliprect store? - */ - if (radeon->state.scissor.numAllocedClipRects < radeon->numClipRects) { - while (radeon->state.scissor.numAllocedClipRects < - radeon->numClipRects) { - radeon->state.scissor.numAllocedClipRects += 1; /* zero case */ - radeon->state.scissor.numAllocedClipRects *= 2; - } - - if (radeon->state.scissor.pClipRects) - FREE(radeon->state.scissor.pClipRects); - - radeon->state.scissor.pClipRects = - MALLOC(radeon->state.scissor.numAllocedClipRects * - sizeof(drm_clip_rect_t)); - - if (radeon->state.scissor.pClipRects == NULL) { - radeon->state.scissor.numAllocedClipRects = 0; - return; - } - } - - out = radeon->state.scissor.pClipRects; - radeon->state.scissor.numClipRects = 0; - - for (i = 0; i < radeon->numClipRects; i++) { - if (intersect_rect(out, - &radeon->pClipRects[i], - &radeon->state.scissor.rect)) { - radeon->state.scissor.numClipRects++; - out++; - } - } -} - -void radeonUpdateScissor(GLcontext* ctx) -{ - radeonContextPtr radeon = RADEON_CONTEXT(ctx); - - if (radeon->dri.drawable) { - __DRIdrawablePrivate *dPriv = radeon->dri.drawable; - int x1 = dPriv->x + ctx->Scissor.X; - int y1 = dPriv->y + dPriv->h - (ctx->Scissor.Y + ctx->Scissor.Height); - - radeon->state.scissor.rect.x1 = x1; - radeon->state.scissor.rect.y1 = y1; - radeon->state.scissor.rect.x2 = x1 + ctx->Scissor.Width; - radeon->state.scissor.rect.y2 = y1 + ctx->Scissor.Height; - - radeonRecalcScissorRects(radeon); - } -} - -static void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h) -{ - if (ctx->Scissor.Enabled) { - /* We don't pipeline cliprect changes */ - r300Flush(ctx); - radeonUpdateScissor(ctx); - } -} - - -/** - * Update cliprects and scissors. - */ -void radeonSetCliprects(radeonContextPtr radeon) -{ - __DRIdrawablePrivate *const drawable = radeon->dri.drawable; - __DRIdrawablePrivate *const readable = radeon->dri.readable; - GLframebuffer *const draw_fb = (GLframebuffer*)drawable->driverPrivate; - GLframebuffer *const read_fb = (GLframebuffer*)readable->driverPrivate; - - if (draw_fb->_ColorDrawBufferIndexes[0] == BUFFER_BACK_LEFT) { - /* Can't ignore 2d windows if we are page flipping. */ - if (drawable->numBackClipRects == 0 || radeon->doPageFlip || - radeon->sarea->pfCurrentPage == 1) { - radeon->numClipRects = drawable->numClipRects; - radeon->pClipRects = drawable->pClipRects; - } else { - radeon->numClipRects = drawable->numBackClipRects; - radeon->pClipRects = drawable->pBackClipRects; - } - } else { - /* front buffer (or none, or multiple buffers */ - radeon->numClipRects = drawable->numClipRects; - radeon->pClipRects = drawable->pClipRects; - } - - if ((draw_fb->Width != drawable->w) || - (draw_fb->Height != drawable->h)) { - _mesa_resize_framebuffer(radeon->glCtx, draw_fb, - drawable->w, drawable->h); - draw_fb->Initialized = GL_TRUE; - } - - if (drawable != readable) { - if ((read_fb->Width != readable->w) || - (read_fb->Height != readable->h)) { - _mesa_resize_framebuffer(radeon->glCtx, read_fb, - readable->w, readable->h); - read_fb->Initialized = GL_TRUE; - } - } - - if (radeon->state.scissor.enabled) - radeonRecalcScissorRects(radeon); - - radeon->lastStamp = drawable->lastStamp; -} - - -/** - * Handle common enable bits. - * Called as a fallback by r200Enable/r300Enable. - */ -void radeonEnable(GLcontext* ctx, GLenum cap, GLboolean state) -{ - radeonContextPtr radeon = RADEON_CONTEXT(ctx); - - switch(cap) { - case GL_SCISSOR_TEST: - /* We don't pipeline cliprect & scissor changes */ - r300Flush(ctx); - - radeon->state.scissor.enabled = state; - radeonUpdateScissor(ctx); - break; - - default: - return; - } -} - - -/** - * Initialize default state. - * This function is called once at context init time from - * r200InitState/r300InitState - */ -void radeonInitState(radeonContextPtr radeon) -{ - radeon->Fallback = 0; - - if (radeon->glCtx->Visual.doubleBufferMode && radeon->sarea->pfCurrentPage == 0) { - radeon->state.color.drawOffset = radeon->radeonScreen->backOffset; - radeon->state.color.drawPitch = radeon->radeonScreen->backPitch; - } else { - radeon->state.color.drawOffset = radeon->radeonScreen->frontOffset; - radeon->state.color.drawPitch = radeon->radeonScreen->frontPitch; - } -} - - -/** - * Initialize common state functions. - * Called by r200InitStateFuncs/r300InitStateFuncs - */ -void radeonInitStateFuncs(struct dd_function_table *functions) -{ - functions->Scissor = radeonScissor; -} diff --git a/src/mesa/drivers/dri/r300/radeon_state.h b/src/mesa/drivers/dri/r300/radeon_state.h deleted file mode 100644 index 821cb40c7e..0000000000 --- a/src/mesa/drivers/dri/r300/radeon_state.h +++ /dev/null @@ -1,43 +0,0 @@ -/* -Copyright (C) 2004 Nicolai Haehnle. All Rights Reserved. - -Permission is hereby granted, free of charge, to any person obtaining -a copy of this software and associated documentation files (the -"Software"), to deal in the Software without restriction, including -without limitation the rights to use, copy, modify, merge, publish, -distribute, sublicense, and/or sell copies of the Software, and to -permit persons to whom the Software is furnished to do so, subject to -the following conditions: - -The above copyright notice and this permission notice (including the -next paragraph) shall be included in all copies or substantial -portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. -IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE -LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - -**************************************************************************/ - -/* - * Authors: - * Nicolai Haehnle <prefect_@gmx.net> - */ - -#ifndef __RADEON_STATE_H__ -#define __RADEON_STATE_H__ - -extern void radeonRecalcScissorRects(radeonContextPtr radeon); -extern void radeonSetCliprects(radeonContextPtr radeon); -extern void radeonUpdateScissor(GLcontext* ctx); - -extern void radeonEnable(GLcontext* ctx, GLenum cap, GLboolean state); - -extern void radeonInitState(radeonContextPtr radeon); -extern void radeonInitStateFuncs(struct dd_function_table* functions); - -#endif |