diff options
Diffstat (limited to 'src/mesa/drivers/dri/radeon/radeon_ioctl.c')
-rw-r--r-- | src/mesa/drivers/dri/radeon/radeon_ioctl.c | 1474 |
1 files changed, 515 insertions, 959 deletions
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.c b/src/mesa/drivers/dri/radeon/radeon_ioctl.c index 09acf6b4f8..22584f4817 100644 --- a/src/mesa/drivers/dri/radeon/radeon_ioctl.c +++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.c @@ -37,12 +37,27 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #include <sched.h> #include <errno.h> +#include "main/attrib.h" +#include "main/enable.h" +#include "main/blend.h" +#include "main/bufferobj.h" +#include "main/buffers.h" +#include "main/depth.h" +#include "main/shaders.h" +#include "main/texstate.h" +#include "main/varray.h" +#include "glapi/dispatch.h" +#include "swrast/swrast.h" +#include "main/stencil.h" +#include "main/matrix.h" + #include "main/glheader.h" #include "main/imports.h" #include "main/simple_list.h" #include "swrast/swrast.h" #include "radeon_context.h" +#include "radeon_common.h" #include "radeon_state.h" #include "radeon_ioctl.h" #include "radeon_tcl.h" @@ -58,75 +73,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. #define RADEON_IDLE_RETRY 16 -static void radeonWaitForIdle( radeonContextPtr rmesa ); -static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, - const char * caller ); - -static void print_state_atom( struct radeon_state_atom *state ) -{ - int i; - - fprintf(stderr, "emit %s/%d\n", state->name, state->cmd_size); - - if (RADEON_DEBUG & DEBUG_VERBOSE) - for (i = 0 ; i < state->cmd_size ; i++) - fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]); - -} - -static void radeonSaveHwState( radeonContextPtr rmesa ) -{ - struct radeon_state_atom *atom; - char * dest = rmesa->backup_store.cmd_buf; - - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "%s\n", __FUNCTION__); - - rmesa->backup_store.cmd_used = 0; - - foreach( atom, &rmesa->hw.atomlist ) { - if ( atom->check( rmesa->glCtx ) ) { - int size = atom->cmd_size * 4; - memcpy( dest, atom->cmd, size); - dest += size; - rmesa->backup_store.cmd_used += size; - if (RADEON_DEBUG & DEBUG_STATE) - print_state_atom( atom ); - } - } - - assert( rmesa->backup_store.cmd_used <= RADEON_CMD_BUF_SZ ); - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Returning to radeonEmitState\n"); -} - -/* At this point we were in FlushCmdBufLocked but we had lost our context, so - * we need to unwire our current cmdbuf, hook the one with the saved state in - * it, flush it, and then put the current one back. This is so commands at the - * start of a cmdbuf can rely on the state being kept from the previous one. - */ -static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa ) -{ - GLuint nr_released_bufs; - struct radeon_store saved_store; - - if (rmesa->backup_store.cmd_used == 0) - return; - - if (RADEON_DEBUG & DEBUG_STATE) - fprintf(stderr, "Emitting backup state on lost context\n"); - - rmesa->lost_context = GL_FALSE; - - nr_released_bufs = rmesa->dma.nr_released_bufs; - saved_store = rmesa->store; - rmesa->dma.nr_released_bufs = 0; - rmesa->store = rmesa->backup_store; - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); - rmesa->dma.nr_released_bufs = nr_released_bufs; - rmesa->store = saved_store; -} - /* ============================================================= * Kernel command buffer handling */ @@ -134,965 +80,572 @@ static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa ) /* The state atoms will be emitted in the order they appear in the atom list, * so this step is important. */ -void radeonSetUpAtomList( radeonContextPtr rmesa ) +void radeonSetUpAtomList( r100ContextPtr rmesa ) { - int i, mtu = rmesa->glCtx->Const.MaxTextureUnits; - - make_empty_list(&rmesa->hw.atomlist); - rmesa->hw.atomlist.name = "atom-list"; - - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ctx); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.set); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lin); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msk); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.vpt); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tcl); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.msc); + int i, mtu = rmesa->radeon.glCtx->Const.MaxTextureUnits; + + make_empty_list(&rmesa->radeon.hw.atomlist); + rmesa->radeon.hw.atomlist.name = "atom-list"; + + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ctx); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.set); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lin); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msk); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.vpt); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tcl); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.msc); for (i = 0; i < mtu; ++i) { - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.tex[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.txr[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.cube[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.tex[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.txr[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.cube[i]); } - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.zbs); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mtl); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.zbs); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mtl); for (i = 0; i < 3 + mtu; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.mat[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.mat[i]); for (i = 0; i < 8; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.lit[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.lit[i]); for (i = 0; i < 6; ++i) - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.ucp[i]); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.eye); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.grd); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.fog); - insert_at_tail(&rmesa->hw.atomlist, &rmesa->hw.glt); -} - -void radeonEmitState( radeonContextPtr rmesa ) -{ - struct radeon_state_atom *atom; - char *dest; - - if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS)) - fprintf(stderr, "%s\n", __FUNCTION__); - - if (rmesa->save_on_next_emit) { - radeonSaveHwState(rmesa); - rmesa->save_on_next_emit = GL_FALSE; - } - - /* this code used to return here but now it emits zbs */ - - /* 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 - * radeonAllocCmdBuf code here without all the checks. - */ - radeonEnsureCmdBufSpace(rmesa, rmesa->hw.max_state_size); - dest = rmesa->store.cmd_buf + rmesa->store.cmd_used; - - /* We always always emit zbs, this is due to a bug found by keithw in - the hardware and rediscovered after Erics changes by me. - if you ever touch this code make sure you emit zbs otherwise - you get tcl lockups on at least M7/7500 class of chips - airlied */ - rmesa->hw.zbs.dirty=1; - - if (RADEON_DEBUG & DEBUG_STATE) { - foreach(atom, &rmesa->hw.atomlist) { - if (atom->dirty || rmesa->hw.all_dirty) { - if (atom->check(rmesa->glCtx)) - print_state_atom(atom); - else - fprintf(stderr, "skip state %s\n", atom->name); - } - } - } - - foreach(atom, &rmesa->hw.atomlist) { - if (rmesa->hw.all_dirty) - atom->dirty = GL_TRUE; - if (!(rmesa->radeonScreen->chip_flags & RADEON_CHIPSET_TCL) && - atom->is_tcl) - atom->dirty = GL_FALSE; - if (atom->dirty) { - if (atom->check(rmesa->glCtx)) { - int size = atom->cmd_size * 4; - memcpy(dest, atom->cmd, size); - dest += size; - rmesa->store.cmd_used += size; - atom->dirty = GL_FALSE; - } - } - } - - assert(rmesa->store.cmd_used <= RADEON_CMD_BUF_SZ); - - rmesa->hw.is_dirty = GL_FALSE; - rmesa->hw.all_dirty = GL_FALSE; + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.ucp[i]); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.eye); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.grd); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.fog); + insert_at_tail(&rmesa->radeon.hw.atomlist, &rmesa->hw.glt); } /* Fire a section of the retained (indexed_verts) buffer as a regular * primtive. */ -extern void radeonEmitVbufPrim( radeonContextPtr rmesa, +extern void radeonEmitVbufPrim( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint vertex_nr ) { - drm_radeon_cmd_header_t *cmd; - + BATCH_LOCALS(&rmesa->radeon); assert(!(primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND)); - radeonEmitState( rmesa ); + radeonEmitState(&rmesa->radeon); - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s cmd_used/4: %d\n", __FUNCTION__, - rmesa->store.cmd_used/4); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VBUF_BUFSZ, - __FUNCTION__ ); #if RADEON_OLD_PACKETS - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM | (3 << 16); - cmd[2].i = rmesa->ioctl.vertex_offset; - cmd[3].i = vertex_nr; - cmd[4].i = vertex_format; - cmd[5].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_LIST | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | - (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x offt 0x%x vfmt 0x%x vfcntl %x \n", - __FUNCTION__, - cmd[1].i, cmd[2].i, cmd[4].i, cmd[5].i); -#else - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_VBUF | (1 << 16); - cmd[2].i = vertex_format; - cmd[3].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_LIST | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_MAOS_ENABLE | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | - (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); - - - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x vfmt 0x%x vfcntl %x \n", - __FUNCTION__, - cmd[1].i, cmd[2].i, cmd[3].i); + BEGIN_BATCH(8); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 3); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + } else { + OUT_BATCH(rmesa->ioctl.vertex_offset); + } + + OUT_BATCH(vertex_nr); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | RADEON_CP_VC_CNTL_PRIM_WALK_LIST | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | + (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); + + if (rmesa->radeon.radeonScreen->kernel_mm) { + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->ioctl.bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + + END_BATCH(); + +#else + BEGIN_BATCH(4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_DRAW_VBUF, 1); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_LIST | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_MAOS_ENABLE | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE | + (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT)); + END_BATCH(); #endif } - -void radeonFlushElts( radeonContextPtr rmesa ) +void radeonFlushElts( GLcontext *ctx ) { - int *cmd = (int *)(rmesa->store.cmd_buf + rmesa->store.elts_start); - int dwords; -#if RADEON_OLD_PACKETS - int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 24)) / 2; -#else - int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 16)) / 2; -#endif - + r100ContextPtr rmesa = R100_CONTEXT(ctx); + BATCH_LOCALS(&rmesa->radeon); + int nr; + uint32_t *cmd = (uint32_t *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_start); + int dwords = (rmesa->radeon.cmdbuf.cs->section_ndw - rmesa->radeon.cmdbuf.cs->section_cdw); + if (RADEON_DEBUG & DEBUG_IOCTL) fprintf(stderr, "%s\n", __FUNCTION__); - assert( rmesa->dma.flush == radeonFlushElts ); - rmesa->dma.flush = NULL; + assert( rmesa->radeon.dma.flush == radeonFlushElts ); + rmesa->radeon.dma.flush = NULL; - /* Cope with odd number of elts: - */ - rmesa->store.cmd_used = (rmesa->store.cmd_used + 2) & ~2; - dwords = (rmesa->store.cmd_used - rmesa->store.elts_start) / 4; + nr = rmesa->tcl.elt_used; + +#if RADEON_OLD_PACKETS + if (rmesa->radeon.radeonScreen->kernel_mm) { + dwords -= 2; + } +#endif #if RADEON_OLD_PACKETS - cmd[1] |= (dwords - 3) << 16; + cmd[1] |= (dwords + 3) << 16; cmd[5] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT; #else - cmd[1] |= (dwords - 3) << 16; + cmd[1] |= (dwords + 2) << 16; cmd[3] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT; #endif + rmesa->radeon.cmdbuf.cs->cdw += dwords; + rmesa->radeon.cmdbuf.cs->section_cdw += dwords; + +#if RADEON_OLD_PACKETS + if (rmesa->radeon.radeonScreen->kernel_mm) { + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->ioctl.bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } +#endif + + END_BATCH(); + if (RADEON_DEBUG & DEBUG_SYNC) { fprintf(stderr, "%s: Syncing\n", __FUNCTION__); - radeonFinish( rmesa->glCtx ); + radeonFinish( rmesa->radeon.glCtx ); } -} +} -GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa, +GLushort *radeonAllocEltsOpenEnded( r100ContextPtr rmesa, GLuint vertex_format, GLuint primitive, GLuint min_nr ) { - drm_radeon_cmd_header_t *cmd; GLushort *retval; + int align_min_nr; + BATCH_LOCALS(&rmesa->radeon); if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, min_nr); + fprintf(stderr, "%s %d prim %x\n", __FUNCTION__, min_nr, primitive); assert((primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND)); - radeonEmitState( rmesa ); + radeonEmitState(&rmesa->radeon); - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, - ELTS_BUFSZ(min_nr), - __FUNCTION__ ); + rmesa->tcl.elt_cmd_start = rmesa->radeon.cmdbuf.cs->cdw; + + /* round up min_nr to align the state */ + align_min_nr = (min_nr + 1) & ~1; + #if RADEON_OLD_PACKETS - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM; - cmd[2].i = rmesa->ioctl.vertex_offset; - cmd[3].i = 0xffff; - cmd[4].i = vertex_format; - cmd[5].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_IND | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); - - retval = (GLushort *)(cmd+6); -#else - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP; - cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_INDX; - cmd[2].i = vertex_format; - cmd[3].i = (primitive | - RADEON_CP_VC_CNTL_PRIM_WALK_IND | - RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | - RADEON_CP_VC_CNTL_MAOS_ENABLE | - RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); - - retval = (GLushort *)(cmd+4); + BEGIN_BATCH_NO_AUTOSTATE(2+ELTS_BUFSZ(align_min_nr)/4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM, 0); + if (!rmesa->radeon.radeonScreen->kernel_mm) { + OUT_BATCH_RELOC(rmesa->ioctl.vertex_offset, rmesa->ioctl.bo, rmesa->ioctl.vertex_offset, RADEON_GEM_DOMAIN_GTT, 0, 0); + } else { + OUT_BATCH(rmesa->ioctl.vertex_offset); + } + OUT_BATCH(0xffff); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_IND | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); + +#else + BEGIN_BATCH_NO_AUTOSTATE(ELTS_BUFSZ(align_min_nr)/4); + OUT_BATCH_PACKET3_CLIP(RADEON_CP_PACKET3_DRAW_INDX, 0); + OUT_BATCH(vertex_format); + OUT_BATCH(primitive | + RADEON_CP_VC_CNTL_PRIM_WALK_IND | + RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA | + RADEON_CP_VC_CNTL_MAOS_ENABLE | + RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE); #endif - if (RADEON_DEBUG & DEBUG_PRIMS) - fprintf(stderr, "%s: header 0x%x vfmt 0x%x prim %x \n", - __FUNCTION__, - cmd[1].i, vertex_format, primitive); - assert(!rmesa->dma.flush); - rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; - rmesa->dma.flush = radeonFlushElts; + rmesa->tcl.elt_cmd_offset = rmesa->radeon.cmdbuf.cs->cdw; + rmesa->tcl.elt_used = min_nr; + + retval = (GLushort *)(rmesa->radeon.cmdbuf.cs->packets + rmesa->tcl.elt_cmd_offset); + + if (RADEON_DEBUG & DEBUG_PRIMS) + fprintf(stderr, "%s: header prim %x \n", + __FUNCTION__, primitive); - rmesa->store.elts_start = ((char *)cmd) - rmesa->store.cmd_buf; + assert(!rmesa->radeon.dma.flush); + rmesa->radeon.glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES; + rmesa->radeon.dma.flush = radeonFlushElts; return retval; } - - -void radeonEmitVertexAOS( radeonContextPtr rmesa, +void radeonEmitVertexAOS( r100ContextPtr rmesa, GLuint vertex_size, + struct radeon_bo *bo, GLuint offset ) { #if RADEON_OLD_PACKETS - rmesa->ioctl.vertex_size = vertex_size; rmesa->ioctl.vertex_offset = offset; + rmesa->ioctl.bo = bo; #else - drm_radeon_cmd_header_t *cmd; + BATCH_LOCALS(&rmesa->radeon); if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_IOCTL)) fprintf(stderr, "%s: vertex_size 0x%x offset 0x%x \n", __FUNCTION__, vertex_size, offset); - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, VERT_AOS_BUFSZ, - __FUNCTION__ ); + BEGIN_BATCH(7); + OUT_BATCH_PACKET3(RADEON_CP_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(); - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (2 << 16); - cmd[2].i = 1; - cmd[3].i = vertex_size | (vertex_size << 8); - cmd[4].i = offset; #endif } -void radeonEmitAOS( radeonContextPtr rmesa, - struct radeon_dma_region **component, +void radeonEmitAOS( r100ContextPtr rmesa, GLuint nr, GLuint offset ) { #if RADEON_OLD_PACKETS assert( nr == 1 ); - assert( component[0]->aos_size == component[0]->aos_stride ); - rmesa->ioctl.vertex_size = component[0]->aos_size; + rmesa->ioctl.bo = rmesa->tcl.aos[0].bo; rmesa->ioctl.vertex_offset = - (component[0]->aos_start + offset * component[0]->aos_stride * 4); + (rmesa->tcl.aos[0].offset + offset * rmesa->tcl.aos[0].stride * 4); #else - drm_radeon_cmd_header_t *cmd; - int sz = AOS_BUFSZ(nr); + BATCH_LOCALS(&rmesa->radeon); + uint32_t voffset; + // int sz = AOS_BUFSZ(nr); + int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2; int i; - int *tmp; if (RADEON_DEBUG & DEBUG_IOCTL) fprintf(stderr, "%s\n", __FUNCTION__); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, sz, - __FUNCTION__ ); - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (((sz / sizeof(int))-3) << 16); - cmd[2].i = nr; - tmp = &cmd[0].i; - cmd += 3; - - for (i = 0 ; i < nr ; i++) { - if (i & 1) { - cmd[0].i |= ((component[i]->aos_stride << 24) | - (component[i]->aos_size << 16)); - cmd[2].i = (component[i]->aos_start + - offset * component[i]->aos_stride * 4); - cmd += 3; + BEGIN_BATCH(sz+2+(nr * 2)); + OUT_BATCH_PACKET3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, sz - 1); + OUT_BATCH(nr); + + if (!rmesa->radeon.radeonScreen->kernel_mm) { + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->tcl.aos[i].components << 0) | + (rmesa->tcl.aos[i].stride << 8) | + (rmesa->tcl.aos[i + 1].components << 16) | + (rmesa->tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->tcl.aos[i + 0].offset + + offset * 4 * rmesa->tcl.aos[i + 0].stride; + OUT_BATCH_RELOC(voffset, + rmesa->tcl.aos[i].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->tcl.aos[i + 1].offset + + offset * 4 * rmesa->tcl.aos[i + 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->tcl.aos[i+1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } - else { - cmd[0].i = ((component[i]->aos_stride << 8) | - (component[i]->aos_size << 0)); - cmd[1].i = (component[i]->aos_start + - offset * component[i]->aos_stride * 4); - } - } - - if (RADEON_DEBUG & DEBUG_VERTS) { - fprintf(stderr, "%s:\n", __FUNCTION__); - for (i = 0 ; i < sz ; i++) - fprintf(stderr, " %d: %x\n", i, tmp[i]); - } -#endif -} - -/* using already shifted color_fmt! */ -void radeonEmitBlit( radeonContextPtr rmesa, /* FIXME: which drmMinor is required? */ - 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_radeon_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_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 8 * sizeof(int), - __FUNCTION__ ); - - - cmd[0].i = 0; - cmd[0].header.cmd_type = RADEON_CMD_PACKET3; - cmd[1].i = RADEON_CP_PACKET3_CNTL_BITBLT_MULTI | (5 << 16); - cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL | - RADEON_GMC_DST_PITCH_OFFSET_CNTL | - RADEON_GMC_BRUSH_NONE | - color_fmt | - 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].i = ((src_pitch/64)<<22) | (src_offset >> 10); - cmd[4].i = ((dst_pitch/64)<<22) | (dst_offset >> 10); - cmd[5].i = (srcx << 16) | srcy; - cmd[6].i = (dstx << 16) | dsty; /* dst */ - cmd[7].i = (w << 16) | h; -} - - -void radeonEmitWait( radeonContextPtr rmesa, GLuint flags ) -{ - drm_radeon_cmd_header_t *cmd; - - assert( !(flags & ~(RADEON_WAIT_2D|RADEON_WAIT_3D)) ); - - cmd = (drm_radeon_cmd_header_t *)radeonAllocCmdBuf( rmesa, 1 * sizeof(int), - __FUNCTION__ ); - cmd[0].i = 0; - cmd[0].wait.cmd_type = RADEON_CMD_WAIT; - cmd[0].wait.flags = flags; -} - - -static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, - const char * caller ) -{ - int ret, i; - drm_radeon_cmd_buffer_t cmd; - - if (rmesa->lost_context) - radeonBackUpAndEmitLostStateLocked(rmesa); - - if (RADEON_DEBUG & DEBUG_IOCTL) { - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (RADEON_DEBUG & DEBUG_VERBOSE) - for (i = 0 ; i < rmesa->store.cmd_used ; i += 4 ) - fprintf(stderr, "%d: %x\n", i/4, - *(int *)(&rmesa->store.cmd_buf[i])); - } - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "%s: Releasing %d buffers\n", __FUNCTION__, - rmesa->dma.nr_released_bufs); - - - if (RADEON_DEBUG & DEBUG_SANITY) { - if (rmesa->state.scissor.enabled) - ret = radeonSanityCmdBuffer( rmesa, - rmesa->state.scissor.numClipRects, - rmesa->state.scissor.pClipRects); - else - ret = radeonSanityCmdBuffer( rmesa, - rmesa->numClipRects, - rmesa->pClipRects); - if (ret) { - fprintf(stderr, "drmSanityCommandWrite: %d\n", ret); - goto out; + + if (nr & 1) { + OUT_BATCH((rmesa->tcl.aos[nr - 1].components << 0) | + (rmesa->tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->tcl.aos[nr - 1].offset + + offset * 4 * rmesa->tcl.aos[nr - 1].stride; + OUT_BATCH_RELOC(voffset, + rmesa->tcl.aos[nr - 1].bo, + voffset, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } - } - - - cmd.bufsz = rmesa->store.cmd_used; - cmd.buf = rmesa->store.cmd_buf; - - if (rmesa->state.scissor.enabled) { - cmd.nbox = rmesa->state.scissor.numClipRects; - cmd.boxes = rmesa->state.scissor.pClipRects; } else { - cmd.nbox = rmesa->numClipRects; - cmd.boxes = rmesa->pClipRects; - } - - ret = drmCommandWrite( rmesa->dri.fd, - DRM_RADEON_CMDBUF, - &cmd, sizeof(cmd) ); - - if (ret) - fprintf(stderr, "drmCommandWrite: %d\n", ret); - - if (RADEON_DEBUG & DEBUG_SYNC) { - fprintf(stderr, "\nSyncing in %s\n\n", __FUNCTION__); - radeonWaitForIdleLocked( rmesa ); - } - - out: - rmesa->store.primnr = 0; - rmesa->store.statenr = 0; - rmesa->store.cmd_used = 0; - rmesa->dma.nr_released_bufs = 0; - rmesa->save_on_next_emit = 1; - - return ret; -} - - -/* Note: does not emit any commands to avoid recursion on - * radeonAllocCmdBuf. - */ -void radeonFlushCmdBuf( radeonContextPtr rmesa, const char *caller ) -{ - int ret; - - - LOCK_HARDWARE( rmesa ); - - ret = radeonFlushCmdBufLocked( rmesa, caller ); - - UNLOCK_HARDWARE( rmesa ); - - if (ret) { - fprintf(stderr, "drm_radeon_cmd_buffer_t: %d (exiting)\n", ret); - exit(ret); - } -} - -/* ============================================================= - * Hardware vertex buffer handling - */ - - -void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa ) -{ - struct radeon_dma_buffer *dmabuf; - int fd = rmesa->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 ); - } - - if (rmesa->dma.current.buf) - radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ ); - - if (rmesa->dma.nr_released_bufs > 4) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); - - dma.context = rmesa->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); /* no need to validate */ - - ret = drmDMA( fd, &dma ); - - if (ret != 0) { - /* Free some up this way? - */ - if (rmesa->dma.nr_released_bufs) { - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); + for (i = 0; i + 1 < nr; i += 2) { + OUT_BATCH((rmesa->tcl.aos[i].components << 0) | + (rmesa->tcl.aos[i].stride << 8) | + (rmesa->tcl.aos[i + 1].components << 16) | + (rmesa->tcl.aos[i + 1].stride << 24)); + + voffset = rmesa->tcl.aos[i + 0].offset + + offset * 4 * rmesa->tcl.aos[i + 0].stride; + OUT_BATCH(voffset); + voffset = rmesa->tcl.aos[i + 1].offset + + offset * 4 * rmesa->tcl.aos[i + 1].stride; + OUT_BATCH(voffset); } - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Waiting for buffers\n"); - - radeonWaitForIdleLocked( rmesa ); - ret = drmDMA( fd, &dma ); - - if ( ret != 0 ) { - UNLOCK_HARDWARE( rmesa ); - fprintf( stderr, "Error: Could not get dma buffer... exiting\n" ); - exit( -1 ); + if (nr & 1) { + OUT_BATCH((rmesa->tcl.aos[nr - 1].components << 0) | + (rmesa->tcl.aos[nr - 1].stride << 8)); + voffset = rmesa->tcl.aos[nr - 1].offset + + offset * 4 * rmesa->tcl.aos[nr - 1].stride; + OUT_BATCH(voffset); + } + for (i = 0; i + 1 < nr; i += 2) { + voffset = rmesa->tcl.aos[i + 0].offset + + offset * 4 * rmesa->tcl.aos[i + 0].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->tcl.aos[i+0].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + voffset = rmesa->tcl.aos[i + 1].offset + + offset * 4 * rmesa->tcl.aos[i + 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->tcl.aos[i+1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); + } + if (nr & 1) { + voffset = rmesa->tcl.aos[nr - 1].offset + + offset * 4 * rmesa->tcl.aos[nr - 1].stride; + radeon_cs_write_reloc(rmesa->radeon.cmdbuf.cs, + rmesa->tcl.aos[nr-1].bo, + RADEON_GEM_DOMAIN_GTT, + 0, 0); } } + END_BATCH(); - UNLOCK_HARDWARE(rmesa); - - if (RADEON_DEBUG & DEBUG_DMA) - fprintf(stderr, "Allocated buffer %d\n", index); - - dmabuf = CALLOC_STRUCT( radeon_dma_buffer ); - dmabuf->buf = &rmesa->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; - - rmesa->c_vertexBuffers++; +#endif } -void radeonReleaseDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - const char *caller ) -{ - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); - - if (!region->buf) - return; +/* ================================================================ + * Buffer clear + */ +#define RADEON_MAX_CLEARS 256 - 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 *)radeonAllocCmdBuf( rmesa, sizeof(*cmd), - __FUNCTION__ ); - cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD; - cmd->dma.buf_idx = region->buf->buf->idx; - FREE(region->buf); - rmesa->dma.nr_released_bufs++; - } - - region->buf = NULL; - 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 radeonAllocDmaRegion( radeonContextPtr rmesa, - struct radeon_dma_region *region, - int bytes, - int alignment ) +static void +r100_meta_set_passthrough_transform(r100ContextPtr r100) { - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s %d\n", __FUNCTION__, bytes); - - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); + GLcontext *ctx = r100->radeon.glCtx; - if (region->buf) - radeonReleaseDmaRegion( rmesa, region, __FUNCTION__ ); + r100->meta.saved_vp_x = ctx->Viewport.X; + r100->meta.saved_vp_y = ctx->Viewport.Y; + r100->meta.saved_vp_width = ctx->Viewport.Width; + r100->meta.saved_vp_height = ctx->Viewport.Height; + r100->meta.saved_matrix_mode = ctx->Transform.MatrixMode; - alignment--; - rmesa->dma.current.start = rmesa->dma.current.ptr = - (rmesa->dma.current.ptr + alignment) & ~alignment; + _mesa_Viewport(0, 0, ctx->DrawBuffer->Width, ctx->DrawBuffer->Height); - if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) - radeonRefillCurrentDmaRegion( rmesa ); + _mesa_MatrixMode(GL_PROJECTION); + _mesa_PushMatrix(); + _mesa_LoadIdentity(); + _mesa_Ortho(0, ctx->DrawBuffer->Width, 0, ctx->DrawBuffer->Height, 1, -1); - 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; + _mesa_MatrixMode(GL_MODELVIEW); + _mesa_PushMatrix(); + _mesa_LoadIdentity(); } -/* ================================================================ - * SwapBuffers with client-side throttling - */ - -static uint32_t radeonGetLastFrame (radeonContextPtr rmesa) +static void +r100_meta_restore_transform(r100ContextPtr r100) { - drm_radeon_getparam_t gp; - int ret; - uint32_t frame; - - gp.param = RADEON_PARAM_LAST_FRAME; - gp.value = (int *)&frame; - ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM, - &gp, sizeof(gp) ); - - if ( ret ) { - fprintf( stderr, "%s: drm_radeon_getparam_t: %d\n", __FUNCTION__, ret ); - exit(1); - } + _mesa_MatrixMode(GL_PROJECTION); + _mesa_PopMatrix(); + _mesa_MatrixMode(GL_MODELVIEW); + _mesa_PopMatrix(); - return frame; -} + _mesa_MatrixMode(r100->meta.saved_matrix_mode); -static void radeonEmitIrqLocked( radeonContextPtr rmesa ) -{ - drm_radeon_irq_emit_t ie; - int ret; - - ie.irq_seq = &rmesa->iw.irq_seq; - ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_IRQ_EMIT, - &ie, sizeof(ie) ); - if ( ret ) { - fprintf( stderr, "%s: drm_radeon_irq_emit_t: %d\n", __FUNCTION__, ret ); - exit(1); - } + _mesa_Viewport(r100->meta.saved_vp_x, r100->meta.saved_vp_y, + r100->meta.saved_vp_width, r100->meta.saved_vp_height); } - -static void radeonWaitIrq( radeonContextPtr rmesa ) +/** + * Perform glClear where mask contains only color, depth, and/or stencil. + * + * The implementation is based on calling into Mesa to set GL state and + * performing normal triangle rendering. The intent of this path is to + * have as generic a path as possible, so that any driver could make use of + * it. + */ +static void radeon_clear_tris(GLcontext *ctx, GLbitfield mask) { - int ret; - - do { - ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_IRQ_WAIT, - &rmesa->iw, sizeof(rmesa->iw) ); - } while (ret && (errno == EINTR || errno == EBUSY)); - - if ( ret ) { - fprintf( stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, ret ); - exit(1); + r100ContextPtr rmesa = R100_CONTEXT(ctx); + GLfloat vertices[4][3]; + GLfloat color[4][4]; + GLfloat dst_z; + struct gl_framebuffer *fb = ctx->DrawBuffer; + int i; + GLboolean saved_fp_enable = GL_FALSE, saved_vp_enable = GL_FALSE; + GLboolean saved_shader_program = 0; + unsigned int saved_active_texture; + + assert((mask & ~(BUFFER_BIT_BACK_LEFT | BUFFER_BIT_FRONT_LEFT | + BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL)) == 0); + + _mesa_PushAttrib(GL_COLOR_BUFFER_BIT | + GL_CURRENT_BIT | + GL_DEPTH_BUFFER_BIT | + GL_ENABLE_BIT | + GL_STENCIL_BUFFER_BIT | + GL_TRANSFORM_BIT | + GL_CURRENT_BIT); + _mesa_PushClientAttrib(GL_CLIENT_VERTEX_ARRAY_BIT); + saved_active_texture = ctx->Texture.CurrentUnit; + + /* Disable existing GL state we don't want to apply to a clear. */ + _mesa_Disable(GL_ALPHA_TEST); + _mesa_Disable(GL_BLEND); + _mesa_Disable(GL_CULL_FACE); + _mesa_Disable(GL_FOG); + _mesa_Disable(GL_POLYGON_SMOOTH); + _mesa_Disable(GL_POLYGON_STIPPLE); + _mesa_Disable(GL_POLYGON_OFFSET_FILL); + _mesa_Disable(GL_LIGHTING); + _mesa_Disable(GL_CLIP_PLANE0); + _mesa_Disable(GL_CLIP_PLANE1); + _mesa_Disable(GL_CLIP_PLANE2); + _mesa_Disable(GL_CLIP_PLANE3); + _mesa_Disable(GL_CLIP_PLANE4); + _mesa_Disable(GL_CLIP_PLANE5); + if (ctx->Extensions.ARB_fragment_program && ctx->FragmentProgram.Enabled) { + saved_fp_enable = GL_TRUE; + _mesa_Disable(GL_FRAGMENT_PROGRAM_ARB); + } + if (ctx->Extensions.ARB_vertex_program && ctx->VertexProgram.Enabled) { + saved_vp_enable = GL_TRUE; + _mesa_Disable(GL_VERTEX_PROGRAM_ARB); + } + if (ctx->Extensions.ARB_shader_objects && ctx->Shader.CurrentProgram) { + saved_shader_program = ctx->Shader.CurrentProgram->Name; + _mesa_UseProgramObjectARB(0); } -} - - -static void radeonWaitForFrameCompletion( radeonContextPtr rmesa ) -{ - drm_radeon_sarea_t *sarea = rmesa->sarea; - - if (rmesa->do_irqs) { - if (radeonGetLastFrame(rmesa) < sarea->last_frame) { - if (!rmesa->irqsEmitted) { - while (radeonGetLastFrame (rmesa) < sarea->last_frame) - ; - } - else { - UNLOCK_HARDWARE( rmesa ); - radeonWaitIrq( rmesa ); - LOCK_HARDWARE( rmesa ); + + if (ctx->Texture._EnabledUnits != 0) { + int i; + + for (i = 0; i < ctx->Const.MaxTextureUnits; i++) { + _mesa_ActiveTextureARB(GL_TEXTURE0 + i); + _mesa_Disable(GL_TEXTURE_1D); + _mesa_Disable(GL_TEXTURE_2D); + _mesa_Disable(GL_TEXTURE_3D); + if (ctx->Extensions.ARB_texture_cube_map) + _mesa_Disable(GL_TEXTURE_CUBE_MAP_ARB); + if (ctx->Extensions.NV_texture_rectangle) + _mesa_Disable(GL_TEXTURE_RECTANGLE_NV); + if (ctx->Extensions.MESA_texture_array) { + _mesa_Disable(GL_TEXTURE_1D_ARRAY_EXT); + _mesa_Disable(GL_TEXTURE_2D_ARRAY_EXT); } - rmesa->irqsEmitted = 10; - } - - if (rmesa->irqsEmitted) { - radeonEmitIrqLocked( rmesa ); - rmesa->irqsEmitted--; - } - } - else { - while (radeonGetLastFrame (rmesa) < sarea->last_frame) { - UNLOCK_HARDWARE( rmesa ); - if (rmesa->do_usleeps) - DO_USLEEP( 1 ); - LOCK_HARDWARE( rmesa ); } } -} - -/* Copy the back color buffer to the front color buffer. - */ -void radeonCopyBuffer( __DRIdrawablePrivate *dPriv, - const drm_clip_rect_t *rect) -{ - radeonContextPtr rmesa; - GLint nbox, i, ret; - GLboolean missed_target; - int64_t ust; - __DRIscreenPrivate *psp; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx ); + + r100_meta_set_passthrough_transform(rmesa); + + for (i = 0; i < 4; i++) { + color[i][0] = ctx->Color.ClearColor[0]; + color[i][1] = ctx->Color.ClearColor[1]; + color[i][2] = ctx->Color.ClearColor[2]; + color[i][3] = ctx->Color.ClearColor[3]; } - RADEON_FIREVERTICES( rmesa ); - LOCK_HARDWARE( rmesa ); - - /* Throttle the frame rate -- only allow one pending swap buffers - * request at a time. + /* convert clear Z from [0,1] to NDC coord in [-1,1] */ + dst_z = -1.0 + 2.0 * ctx->Depth.Clear; + + /* Prepare the vertices, which are the same regardless of which buffer we're + * drawing to. */ - radeonWaitForFrameCompletion( rmesa ); - if (!rect) - { - UNLOCK_HARDWARE( rmesa ); - driWaitForVBlank( dPriv, & missed_target ); - LOCK_HARDWARE( rmesa ); - } - - 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 = rmesa->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; - } + vertices[0][0] = fb->_Xmin; + vertices[0][1] = fb->_Ymin; + vertices[0][2] = dst_z; + vertices[1][0] = fb->_Xmax; + vertices[1][1] = fb->_Ymin; + vertices[1][2] = dst_z; + vertices[2][0] = fb->_Xmax; + vertices[2][1] = fb->_Ymax; + vertices[2][2] = dst_z; + vertices[3][0] = fb->_Xmin; + vertices[3][1] = fb->_Ymax; + vertices[3][2] = dst_z; + + _mesa_ColorPointer(4, GL_FLOAT, 4 * sizeof(GLfloat), &color); + _mesa_VertexPointer(3, GL_FLOAT, 3 * sizeof(GLfloat), &vertices); + _mesa_Enable(GL_COLOR_ARRAY); + _mesa_Enable(GL_VERTEX_ARRAY); + + while (mask != 0) { + GLuint this_mask = 0; + + if (mask & BUFFER_BIT_BACK_LEFT) + this_mask = BUFFER_BIT_BACK_LEFT; + else if (mask & BUFFER_BIT_FRONT_LEFT) + this_mask = BUFFER_BIT_FRONT_LEFT; + + /* Clear depth/stencil in the same pass as color. */ + this_mask |= (mask & (BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL)); + + /* Select the current color buffer and use the color write mask if + * we have one, otherwise don't write any color channels. + */ + if (this_mask & BUFFER_BIT_FRONT_LEFT) + _mesa_DrawBuffer(GL_FRONT_LEFT); + else if (this_mask & BUFFER_BIT_BACK_LEFT) + _mesa_DrawBuffer(GL_BACK_LEFT); + else + _mesa_ColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE); - b++; - n++; + /* Control writing of the depth clear value to depth. */ + if (this_mask & BUFFER_BIT_DEPTH) { + _mesa_DepthFunc(GL_ALWAYS); + _mesa_Enable(GL_DEPTH_TEST); + } else { + _mesa_Disable(GL_DEPTH_TEST); + _mesa_DepthMask(GL_FALSE); } - rmesa->sarea->nbox = n; - - if (!n) - continue; - - ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP ); - if ( ret ) { - fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret ); - UNLOCK_HARDWARE( rmesa ); - exit( 1 ); + /* Control writing of the stencil clear value to stencil. */ + if (this_mask & BUFFER_BIT_STENCIL) { + _mesa_Enable(GL_STENCIL_TEST); + _mesa_StencilOp(GL_REPLACE, GL_REPLACE, GL_REPLACE); + _mesa_StencilFuncSeparate(GL_FRONT, GL_ALWAYS, ctx->Stencil.Clear, + ctx->Stencil.WriteMask[0]); + } else { + _mesa_Disable(GL_STENCIL_TEST); } - } - - UNLOCK_HARDWARE( rmesa ); - if (!rect) - { - psp = dPriv->driScreenPriv; - rmesa->swap_count++; - (*psp->systemTime->getUST)( & ust ); - if ( missed_target ) { - rmesa->swap_missed_count++; - rmesa->swap_missed_ust = ust - rmesa->swap_ust; - } - - rmesa->swap_ust = ust; - rmesa->hw.all_dirty = GL_TRUE; - } -} - -void radeonPageFlip( __DRIdrawablePrivate *dPriv ) -{ - radeonContextPtr rmesa; - GLint ret; - GLboolean missed_target; - __DRIscreenPrivate *psp; - - assert(dPriv); - assert(dPriv->driContextPriv); - assert(dPriv->driContextPriv->driverPrivate); - - rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate; - psp = dPriv->driScreenPriv; - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__, - rmesa->sarea->pfCurrentPage); - } - - RADEON_FIREVERTICES( rmesa ); - LOCK_HARDWARE( rmesa ); - - /* Need to do this for the perf box placement: - */ - if (dPriv->numClipRects) - { - drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = rmesa->sarea->boxes; - b[0] = box[0]; - rmesa->sarea->nbox = 1; - } + CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4)); - /* Throttle the frame rate -- only allow a few pending swap buffers - * request at a time. - */ - radeonWaitForFrameCompletion( rmesa ); - UNLOCK_HARDWARE( rmesa ); - driWaitForVBlank( dPriv, & missed_target ); - if ( missed_target ) { - rmesa->swap_missed_count++; - (void) (*psp->systemTime->getUST)( & rmesa->swap_missed_ust ); + mask &= ~this_mask; } - LOCK_HARDWARE( rmesa ); - ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_FLIP ); + r100_meta_restore_transform(rmesa); - UNLOCK_HARDWARE( rmesa ); + _mesa_ActiveTextureARB(GL_TEXTURE0 + saved_active_texture); + if (saved_fp_enable) + _mesa_Enable(GL_FRAGMENT_PROGRAM_ARB); + if (saved_vp_enable) + _mesa_Enable(GL_VERTEX_PROGRAM_ARB); - if ( ret ) { - fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret ); - exit( 1 ); - } - - rmesa->swap_count++; - (void) (*psp->systemTime->getUST)( & rmesa->swap_ust ); + if (saved_shader_program) + _mesa_UseProgramObjectARB(saved_shader_program); - /* Get ready for drawing next frame. Update the renderbuffers' - * flippedOffset/Pitch fields so we draw into the right place. - */ - driFlipRenderbuffers(rmesa->glCtx->WinSysDrawBuffer, - rmesa->sarea->pfCurrentPage); - - radeonUpdateDrawBuffer(rmesa->glCtx); + _mesa_PopClientAttrib(); + _mesa_PopAttrib(); } +static void radeonUserClear(GLcontext *ctx, GLuint mask) +{ + radeon_clear_tris(ctx, mask); +} -/* ================================================================ - * Buffer clear - */ -#define RADEON_MAX_CLEARS 256 - -static void radeonClear( GLcontext *ctx, GLbitfield mask ) +static void radeonKernelClear(GLcontext *ctx, GLuint flags) { - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - __DRIdrawablePrivate *dPriv = rmesa->dri.drawable; - drm_radeon_sarea_t *sarea = rmesa->sarea; + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = rmesa->radeon.dri.drawable; + drm_radeon_sarea_t *sarea = rmesa->radeon.sarea; uint32_t clear; - GLuint flags = 0; - GLuint color_mask = 0; GLint ret, i; GLint cx, cy, cw, ch; - if ( RADEON_DEBUG & DEBUG_IOCTL ) { - fprintf( stderr, "radeonClear\n"); - } - - { - LOCK_HARDWARE( rmesa ); - UNLOCK_HARDWARE( rmesa ); - if ( dPriv->numClipRects == 0 ) - return; - } - - radeonFlush( ctx ); - - if ( mask & BUFFER_BIT_FRONT_LEFT ) { - flags |= RADEON_FRONT; - color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - mask &= ~BUFFER_BIT_FRONT_LEFT; - } - - if ( mask & BUFFER_BIT_BACK_LEFT ) { - flags |= RADEON_BACK; - color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - mask &= ~BUFFER_BIT_BACK_LEFT; - } - - if ( mask & BUFFER_BIT_DEPTH ) { - flags |= RADEON_DEPTH; - mask &= ~BUFFER_BIT_DEPTH; - } - - if ( (mask & BUFFER_BIT_STENCIL) && rmesa->state.stencil.hwBuffer ) { - flags |= RADEON_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 ) - return; - - if (rmesa->using_hyperz) { - flags |= RADEON_USE_COMP_ZBUF; -/* if (rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL) - flags |= RADEON_USE_HIERZ; */ - if (!(rmesa->state.stencil.hwBuffer) || - ((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) && - ((rmesa->state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) { - flags |= RADEON_CLEAR_FASTZ; - } - } - - LOCK_HARDWARE( rmesa ); + LOCK_HARDWARE( &rmesa->radeon ); /* compute region after locking: */ cx = ctx->DrawBuffer->_Xmin; @@ -1112,7 +665,7 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) gp.param = RADEON_PARAM_LAST_CLEAR; gp.value = (int *)&clear; - ret = drmCommandWriteRead( rmesa->dri.fd, + ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_GETPARAM, &gp, sizeof(gp) ); if ( ret ) { @@ -1124,20 +677,20 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) break; } - if ( rmesa->do_usleeps ) { - UNLOCK_HARDWARE( rmesa ); + if ( rmesa->radeon.do_usleeps ) { + UNLOCK_HARDWARE( &rmesa->radeon ); DO_USLEEP( 1 ); - LOCK_HARDWARE( rmesa ); + LOCK_HARDWARE( &rmesa->radeon ); } } /* Send current state to the hardware */ - radeonFlushCmdBufLocked( rmesa, __FUNCTION__ ); + rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ ); for ( i = 0 ; i < dPriv->numClipRects ; ) { GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, dPriv->numClipRects ); drm_clip_rect_t *box = dPriv->pClipRects; - drm_clip_rect_t *b = rmesa->sarea->boxes; + drm_clip_rect_t *b = rmesa->radeon.sarea->boxes; drm_radeon_clear_t clear; drm_radeon_clear_rect_t depth_boxes[RADEON_NR_SAREA_CLIPRECTS]; GLint n = 0; @@ -1172,105 +725,108 @@ static void radeonClear( GLcontext *ctx, GLbitfield mask ) } } - rmesa->sarea->nbox = n; + rmesa->radeon.sarea->nbox = n; clear.flags = flags; - clear.clear_color = rmesa->state.color.clear; - clear.clear_depth = rmesa->state.depth.clear; + clear.clear_color = rmesa->radeon.state.color.clear; + clear.clear_depth = rmesa->radeon.state.depth.clear; clear.color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; - clear.depth_mask = rmesa->state.stencil.clear; + clear.depth_mask = rmesa->radeon.state.stencil.clear; clear.depth_boxes = depth_boxes; n--; - b = rmesa->sarea->boxes; + b = rmesa->radeon.sarea->boxes; for ( ; n >= 0 ; n-- ) { depth_boxes[n].f[CLEAR_X1] = (float)b[n].x1; depth_boxes[n].f[CLEAR_Y1] = (float)b[n].y1; depth_boxes[n].f[CLEAR_X2] = (float)b[n].x2; depth_boxes[n].f[CLEAR_Y2] = (float)b[n].y2; depth_boxes[n].f[CLEAR_DEPTH] = - (float)rmesa->state.depth.clear; + (float)rmesa->radeon.state.depth.clear; } - ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_CLEAR, + ret = drmCommandWrite( rmesa->radeon.dri.fd, DRM_RADEON_CLEAR, &clear, sizeof(drm_radeon_clear_t)); if ( ret ) { - UNLOCK_HARDWARE( rmesa ); + UNLOCK_HARDWARE( &rmesa->radeon ); fprintf( stderr, "DRM_RADEON_CLEAR: return = %d\n", ret ); exit( 1 ); } } - - UNLOCK_HARDWARE( rmesa ); - rmesa->hw.all_dirty = GL_TRUE; + UNLOCK_HARDWARE( &rmesa->radeon ); } - -void radeonWaitForIdleLocked( radeonContextPtr rmesa ) +static void radeonClear( GLcontext *ctx, GLbitfield mask ) { - int fd = rmesa->dri.fd; - int to = 0; - int ret, i = 0; - - rmesa->c_drawWaits++; - - do { - do { - ret = drmCommandNone( fd, DRM_RADEON_CP_IDLE); - } while ( ret && errno == EBUSY && i++ < RADEON_IDLE_RETRY ); - } while ( ( ret == -EBUSY ) && ( to++ < RADEON_TIMEOUT ) ); - - if ( ret < 0 ) { - UNLOCK_HARDWARE( rmesa ); - fprintf( stderr, "Error: Radeon timed out... exiting\n" ); - exit( -1 ); - } -} + r100ContextPtr rmesa = R100_CONTEXT(ctx); + __DRIdrawablePrivate *dPriv = rmesa->radeon.dri.drawable; + GLuint flags = 0; + GLuint color_mask = 0; + GLuint orig_mask = mask; + if ( RADEON_DEBUG & DEBUG_IOCTL ) { + fprintf( stderr, "radeonClear\n"); + } -static void radeonWaitForIdle( radeonContextPtr rmesa ) -{ - LOCK_HARDWARE(rmesa); - radeonWaitForIdleLocked( rmesa ); - UNLOCK_HARDWARE(rmesa); -} + { + LOCK_HARDWARE( &rmesa->radeon ); + UNLOCK_HARDWARE( &rmesa->radeon ); + if ( dPriv->numClipRects == 0 ) + return; + } + + radeon_firevertices(&rmesa->radeon); + if ( mask & BUFFER_BIT_FRONT_LEFT ) { + flags |= RADEON_FRONT; + color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; + mask &= ~BUFFER_BIT_FRONT_LEFT; + } -void radeonFlush( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT( ctx ); + if ( mask & BUFFER_BIT_BACK_LEFT ) { + flags |= RADEON_BACK; + color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK]; + mask &= ~BUFFER_BIT_BACK_LEFT; + } - if (RADEON_DEBUG & DEBUG_IOCTL) - fprintf(stderr, "%s\n", __FUNCTION__); + if ( mask & BUFFER_BIT_DEPTH ) { + flags |= RADEON_DEPTH; + mask &= ~BUFFER_BIT_DEPTH; + } - if (rmesa->dma.flush) - rmesa->dma.flush( rmesa ); + if ( (mask & BUFFER_BIT_STENCIL) && rmesa->radeon.state.stencil.hwBuffer ) { + flags |= RADEON_STENCIL; + mask &= ~BUFFER_BIT_STENCIL; + } - radeonEmitState( rmesa ); - - if (rmesa->store.cmd_used) - radeonFlushCmdBuf( rmesa, __FUNCTION__ ); -} + if ( mask ) { + if (RADEON_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask); + _swrast_Clear( ctx, mask ); + } -/* Make sure all commands have been sent to the hardware and have - * completed processing. - */ -void radeonFinish( GLcontext *ctx ) -{ - radeonContextPtr rmesa = RADEON_CONTEXT(ctx); - radeonFlush( ctx ); - - if (rmesa->do_irqs) { - LOCK_HARDWARE( rmesa ); - radeonEmitIrqLocked( rmesa ); - UNLOCK_HARDWARE( rmesa ); - radeonWaitIrq( rmesa ); + if ( !flags ) + return; + + if (rmesa->using_hyperz) { + flags |= RADEON_USE_COMP_ZBUF; +/* if (rmesa->radeon.radeonScreen->chipset & RADEON_CHIPSET_TCL) + flags |= RADEON_USE_HIERZ; */ + if (!(rmesa->radeon.state.stencil.hwBuffer) || + ((flags & RADEON_DEPTH) && (flags & RADEON_STENCIL) && + ((rmesa->radeon.state.stencil.clear & RADEON_STENCIL_WRITE_MASK) == RADEON_STENCIL_WRITE_MASK))) { + flags |= RADEON_CLEAR_FASTZ; + } } - else - radeonWaitForIdle( rmesa ); -} + if (rmesa->radeon.radeonScreen->kernel_mm) + radeonUserClear(ctx, orig_mask); + else { + radeonKernelClear(ctx, flags); + rmesa->radeon.hw.all_dirty = GL_TRUE; + } +} void radeonInitIoctlFuncs( GLcontext *ctx ) { |