From c30f23c1231e8443e0880efa5326ccecf6eec034 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 6 May 2009 21:44:13 -0700 Subject: i965: Remove bad constant buffer constant-reg-already-loaded optimization. Thanks to branching, the state of c->current_const[i].index at the point of emitting constant loads for this instruction may not match the actual constant currently loaded in the reg at runtime. Fixes a regression in my GLSL program for idr's class since b58b3a786aa38dcc9d72144c2cc691151e46e3d5. --- src/mesa/drivers/dri/i965/brw_wm_glsl.c | 24 +++++++++++------------- 1 file changed, 11 insertions(+), 13 deletions(-) (limited to 'src/mesa') diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 094c1af2fe..c06c820b4d 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -294,23 +294,21 @@ static void fetch_constants(struct brw_wm_compile *c, if (src->File == PROGRAM_STATE_VAR || src->File == PROGRAM_CONSTANT || src->File == PROGRAM_UNIFORM) { - if (c->current_const[i].index != src->Index) { - c->current_const[i].index = src->Index; + c->current_const[i].index = src->Index; #if 0 - printf(" fetch const[%d] for arg %d into reg %d\n", - src->Index, i, c->current_const[i].reg.nr); + printf(" fetch const[%d] for arg %d into reg %d\n", + src->Index, i, c->current_const[i].reg.nr); #endif - /* need to fetch the constant now */ - brw_dp_READ_4(p, - c->current_const[i].reg, /* writeback dest */ - 1, /* msg_reg */ - src->RelAddr, /* relative indexing? */ - 16 * src->Index, /* byte offset */ - SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ - ); - } + /* need to fetch the constant now */ + brw_dp_READ_4(p, + c->current_const[i].reg, /* writeback dest */ + 1, /* msg_reg */ + src->RelAddr, /* relative indexing? */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ + ); } } } -- cgit v1.2.3 From fc6e02ce6210d6615af0058f1b57e7ee37a6527f Mon Sep 17 00:00:00 2001 From: Roland Scheidegger Date: Fri, 26 Jun 2009 20:38:07 +0200 Subject: i965: fix fetching constants from constant buffer in glsl path the driver used to overwrite grf0 then use implicit move by send instruction to move contents of grf0 to mrf1. However, we must not overwrite grf0 since it's still used later for fb write. Instead, do the move directly do mrf1 (we could use implicit move from another grf reg to mrf1 but since we need a mov to encode the data anyway it doesn't seem to make sense). I think the dp_READ/WRITE_16 functions may suffer from the same issue. While here also remove unnecessary msg_reg_nr parameter from the dataport functions since always message register 1 is used. --- src/mesa/drivers/dri/i965/brw_eu.h | 3 --- src/mesa/drivers/dri/i965/brw_eu_emit.c | 27 ++++++++++++++++----------- src/mesa/drivers/dri/i965/brw_wm_emit.c | 2 -- src/mesa/drivers/dri/i965/brw_wm_glsl.c | 1 - 4 files changed, 16 insertions(+), 17 deletions(-) (limited to 'src/mesa') diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h index 62c98bd8bb..003332f78c 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.h +++ b/src/mesa/drivers/dri/i965/brw_eu.h @@ -855,12 +855,10 @@ void brw_math( struct brw_compile *p, void brw_dp_READ_16( struct brw_compile *p, struct brw_reg dest, - GLuint msg_reg_nr, GLuint scratch_offset ); void brw_dp_READ_4( struct brw_compile *p, struct brw_reg dest, - GLuint msg_reg_nr, GLboolean relAddr, GLuint location, GLuint bind_table_index ); @@ -875,7 +873,6 @@ void brw_dp_READ_4_vs( struct brw_compile *p, void brw_dp_WRITE_16( struct brw_compile *p, struct brw_reg src, - GLuint msg_reg_nr, GLuint scratch_offset ); /* If/else/endif. Works by manipulating the execution flags on each diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index 60ea44f7a9..2a147fb8c3 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -865,9 +865,9 @@ void brw_math_16( struct brw_compile *p, */ void brw_dp_WRITE_16( struct brw_compile *p, struct brw_reg src, - GLuint msg_reg_nr, GLuint scratch_offset ) { + GLuint msg_reg_nr = 1; { brw_push_insn_state(p); brw_set_mask_control(p, BRW_MASK_DISABLE); @@ -877,7 +877,7 @@ void brw_dp_WRITE_16( struct brw_compile *p, brw_MOV(p, retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D), brw_imm_d(scratch_offset)); - + brw_pop_insn_state(p); } @@ -912,9 +912,9 @@ void brw_dp_WRITE_16( struct brw_compile *p, */ void brw_dp_READ_16( struct brw_compile *p, struct brw_reg dest, - GLuint msg_reg_nr, GLuint scratch_offset ) { + GLuint msg_reg_nr = 1; { brw_push_insn_state(p); brw_set_compression_control(p, BRW_COMPRESSION_NONE); @@ -924,7 +924,7 @@ void brw_dp_READ_16( struct brw_compile *p, brw_MOV(p, retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_D), brw_imm_d(scratch_offset)); - + brw_pop_insn_state(p); } @@ -958,21 +958,26 @@ void brw_dp_READ_16( struct brw_compile *p, */ void brw_dp_READ_4( struct brw_compile *p, struct brw_reg dest, - GLuint msg_reg_nr, GLboolean relAddr, GLuint location, GLuint bind_table_index ) { + /* XXX: relAddr not implemented */ + GLuint msg_reg_nr = 1; { + struct brw_reg b; brw_push_insn_state(p); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_set_mask_control(p, BRW_MASK_DISABLE); - /* set message header global offset field (reg 0, element 2) */ - /* Note that grf[0] will be copied to mrf[1] implicitly by the SEND instr */ - brw_MOV(p, - retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_UD), - brw_imm_d(location)); + /* Setup MRF[1] with location/offset into const buffer */ + b = brw_message_reg(msg_reg_nr); + b = retype(b, BRW_REGISTER_TYPE_UD); + /* XXX I think we're setting all the dwords of MRF[1] to 'location'. + * when the docs say only dword[2] should be set. Hmmm. But it works. + */ + brw_MOV(p, b, brw_imm_ud(location)); brw_pop_insn_state(p); } @@ -988,7 +993,7 @@ void brw_dp_READ_4( struct brw_compile *p, dest = retype(vec8(dest), BRW_REGISTER_TYPE_UW); brw_set_dest(insn, dest); - brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW)); + brw_set_src0(insn, brw_null_reg()); brw_set_dp_read_message(insn, bind_table_index, diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index 72fc21d2eb..a870e75e52 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -1044,7 +1044,6 @@ static void emit_spill( struct brw_wm_compile *c, */ brw_dp_WRITE_16(p, retype(vec16(brw_vec8_grf(0, 0)), BRW_REGISTER_TYPE_UW), - 1, slot); } @@ -1072,7 +1071,6 @@ static void emit_unspill( struct brw_wm_compile *c, brw_dp_READ_16(p, retype(vec16(reg), BRW_REGISTER_TYPE_UW), - 1, slot); } diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index c06c820b4d..a907e1b788 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -304,7 +304,6 @@ static void fetch_constants(struct brw_wm_compile *c, /* need to fetch the constant now */ brw_dp_READ_4(p, c->current_const[i].reg, /* writeback dest */ - 1, /* msg_reg */ src->RelAddr, /* relative indexing? */ 16 * src->Index, /* byte offset */ SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ -- cgit v1.2.3 From a4d952358d8fd0f9d135c91c37969acdcf394f0a Mon Sep 17 00:00:00 2001 From: Zack Rusin Date: Mon, 6 Jul 2009 21:33:56 -0400 Subject: gallium: more fog extraction fixes fix the cases when fog coord/front face/point coord are used in the same shader. --- src/mesa/state_tracker/st_atom_shader.c | 21 +++++++------------- src/mesa/state_tracker/st_mesa_to_tgsi.c | 9 ++++++++- src/mesa/state_tracker/st_program.c | 34 ++++++++++++++++++++++---------- 3 files changed, 39 insertions(+), 25 deletions(-) (limited to 'src/mesa') diff --git a/src/mesa/state_tracker/st_atom_shader.c b/src/mesa/state_tracker/st_atom_shader.c index c02ccc3528..299aa762c2 100644 --- a/src/mesa/state_tracker/st_atom_shader.c +++ b/src/mesa/state_tracker/st_atom_shader.c @@ -137,22 +137,15 @@ find_translated_vp(struct st_context *st, for (inAttr = 0; inAttr < FRAG_ATTRIB_MAX; inAttr++) { if (fragInputsRead & (1 << inAttr)) { + stfp->input_to_slot[inAttr] = numIn; + numIn++; if ((fragInputsRead & FRAG_BIT_FOGC)) { - if (stfp->Base.UsesPointCoord) { - stfp->input_to_slot[inAttr] = numIn; - numIn++; + /* leave placeholders for the + * extra registers we extract from fog */ + if (stfp->Base.UsesFrontFacing || + stfp->Base.UsesPointCoord) { + numIn += 2; } - if (stfp->Base.UsesFrontFacing) { - stfp->input_to_slot[inAttr] = numIn; - numIn++; - } - if (stfp->Base.UsesFogFragCoord) { - stfp->input_to_slot[inAttr] = numIn; - numIn++; - } - } else { - stfp->input_to_slot[inAttr] = numIn; - numIn++; } } else { diff --git a/src/mesa/state_tracker/st_mesa_to_tgsi.c b/src/mesa/state_tracker/st_mesa_to_tgsi.c index 3140ebe04a..8aef3fc6dc 100644 --- a/src/mesa/state_tracker/st_mesa_to_tgsi.c +++ b/src/mesa/state_tracker/st_mesa_to_tgsi.c @@ -122,8 +122,15 @@ map_register_file_index( SWIZZLE_X, SWIZZLE_X, SWIZZLE_X); + /* register after fog */ + return inputMapping[index] + 1; } else { - /* fixme: point coord */ + *swizzle = MAKE_SWIZZLE4(SWIZZLE_Z, + SWIZZLE_W, + SWIZZLE_Z, + SWIZZLE_W); + /* register after frontface */ + return inputMapping[index] + 2; } } /* inputs are mapped according to the user-defined map */ diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c index 9a346fbde0..18d1046831 100644 --- a/src/mesa/state_tracker/st_program.c +++ b/src/mesa/state_tracker/st_program.c @@ -433,20 +433,34 @@ st_translate_fragment_program(struct st_context *st, stfp->input_semantic_index[slot] = 1; interpMode[slot] = TGSI_INTERPOLATE_LINEAR; break; - case FRAG_ATTRIB_FOGC: - if (stfp->Base.UsesPointCoord) { - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC; - stfp->input_semantic_index[slot] = num_generic++; - interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; - } else if (stfp->Base.UsesFrontFacing) { - stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FACE; - stfp->input_semantic_index[slot] = 0; - interpMode[slot] = TGSI_INTERPOLATE_CONSTANT; - } else { + case FRAG_ATTRIB_FOGC: { + int extra_decls = 0; + if (stfp->Base.UsesFogFragCoord) { stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG; stfp->input_semantic_index[slot] = 0; interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE; + input_flags[slot] = stfp->Base.Base.InputFlags[attr]; + ++extra_decls; } + if (stfp->Base.UsesFrontFacing) { + GLint idx = slot + extra_decls; + stfp->input_semantic_name[idx] = TGSI_SEMANTIC_FACE; + stfp->input_semantic_index[idx] = 0; + interpMode[idx] = TGSI_INTERPOLATE_CONSTANT; + input_flags[idx] = stfp->Base.Base.InputFlags[attr]; + ++extra_decls; + } + if (stfp->Base.UsesPointCoord) { + GLint idx = slot + extra_decls; + stfp->input_semantic_name[idx] = TGSI_SEMANTIC_GENERIC; + stfp->input_semantic_index[idx] = num_generic++; + interpMode[idx] = TGSI_INTERPOLATE_PERSPECTIVE; + input_flags[idx] = stfp->Base.Base.InputFlags[attr]; + ++extra_decls; + } + fs_num_inputs += extra_decls - 1; + continue; + } break; case FRAG_ATTRIB_TEX0: case FRAG_ATTRIB_TEX1: -- cgit v1.2.3 From 96601ec8e0e35d540c1819a48df7f08ec5874ab7 Mon Sep 17 00:00:00 2001 From: Michel Dänzer Date: Tue, 7 Jul 2009 12:17:04 +0200 Subject: gallium: Only set FRONT_STATUS_COPY_OF_BACK if there is a back buffer. Fixes potential crash when SwapBuffers is called but there's no back buffer. --- src/mesa/state_tracker/st_framebuffer.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'src/mesa') diff --git a/src/mesa/state_tracker/st_framebuffer.c b/src/mesa/state_tracker/st_framebuffer.c index 33a90ea7db..fe99fc0892 100644 --- a/src/mesa/state_tracker/st_framebuffer.c +++ b/src/mesa/state_tracker/st_framebuffer.c @@ -280,7 +280,8 @@ st_notify_swapbuffers(struct st_framebuffer *stfb) PIPE_FLUSH_SWAPBUFFERS | PIPE_FLUSH_FRAME, NULL ); - ctx->st->frontbuffer_status = FRONT_STATUS_COPY_OF_BACK; + if (st_renderbuffer(stfb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer)) + ctx->st->frontbuffer_status = FRONT_STATUS_COPY_OF_BACK; } } -- cgit v1.2.3 From 71633abafc935c25da9731bab48c228ceb9b4097 Mon Sep 17 00:00:00 2001 From: Michel Dänzer Date: Tue, 7 Jul 2009 14:49:52 +0200 Subject: gallium: Fixes for clobbering stencil values in combined depth/stencil textures. Also fix one case where a 32 bit depth value was incorrectly converted to a combined depth/stencil value. --- src/gallium/auxiliary/util/u_tile.c | 28 ++++++++++++++++++++++++- src/mesa/state_tracker/st_cb_drawpixels.c | 12 ++++++++--- src/mesa/state_tracker/st_cb_texture.c | 35 ++++++++++++++++++++++++------- 3 files changed, 64 insertions(+), 11 deletions(-) (limited to 'src/mesa') diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c index 9747a55cbf..a0c8ed88f7 100644 --- a/src/gallium/auxiliary/util/u_tile.c +++ b/src/gallium/auxiliary/util/u_tile.c @@ -1202,6 +1202,19 @@ pipe_put_tile_z(struct pipe_transfer *pt, } break; case PIPE_FORMAT_S8Z24_UNORM: + { + uint *pDest = (uint *) (map + y * pt->stride + x*4); + assert(pt->usage == PIPE_TRANSFER_READ_WRITE); + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + /* convert 32-bit Z to 24-bit Z, preserve stencil */ + pDest[j] = (pDest[j] & 0xff000000) | ptrc[j] >> 8; + } + pDest += pt->stride/4; + ptrc += srcStride; + } + } + break; case PIPE_FORMAT_X8Z24_UNORM: { uint *pDest = (uint *) (map + y * pt->stride + x*4); @@ -1216,13 +1229,26 @@ pipe_put_tile_z(struct pipe_transfer *pt, } break; case PIPE_FORMAT_Z24S8_UNORM: + { + uint *pDest = (uint *) (map + y * pt->stride + x*4); + assert(pt->usage == PIPE_TRANSFER_READ_WRITE); + for (i = 0; i < h; i++) { + for (j = 0; j < w; j++) { + /* convert 32-bit Z to 24-bit Z, preserve stencil */ + pDest[j] = (pDest[j] & 0xff) | (ptrc[j] & 0xffffff00); + } + pDest += pt->stride/4; + ptrc += srcStride; + } + } + break; case PIPE_FORMAT_Z24X8_UNORM: { uint *pDest = (uint *) (map + y * pt->stride + x*4); for (i = 0; i < h; i++) { for (j = 0; j < w; j++) { /* convert 32-bit Z to 24-bit Z (0 stencil) */ - pDest[j] = ptrc[j] << 8; + pDest[j] = ptrc[j] & 0xffffff00; } pDest += pt->stride/4; ptrc += srcStride; diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c index 2027b713ce..8b5094a04f 100644 --- a/src/mesa/state_tracker/st_cb_drawpixels.c +++ b/src/mesa/state_tracker/st_cb_drawpixels.c @@ -1037,10 +1037,16 @@ st_CopyPixels(GLcontext *ctx, GLint srcx, GLint srcy, st_cond_flush_get_tex_transfer(st, rbRead->texture, 0, 0, 0, PIPE_TRANSFER_READ, srcx, srcy, width, height); + struct pipe_transfer *ptTex; + enum pipe_transfer_usage transfer_usage; - struct pipe_transfer *ptTex = - st_cond_flush_get_tex_transfer(st, pt, 0, 0, 0, PIPE_TRANSFER_WRITE, - 0, 0, width, height); + if (type == GL_DEPTH && pf_is_depth_and_stencil(pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + + ptTex = st_cond_flush_get_tex_transfer(st, pt, 0, 0, 0, transfer_usage, + 0, 0, width, height); if (type == GL_COLOR) { /* alternate path using get/put_tile() */ diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c index 0a72784ce0..909189f9d3 100644 --- a/src/mesa/state_tracker/st_cb_texture.c +++ b/src/mesa/state_tracker/st_cb_texture.c @@ -529,6 +529,7 @@ st_TexImage(GLcontext * ctx, GLint texelBytes, sizeInBytes; GLuint dstRowStride; struct gl_pixelstore_attrib unpackNB; + enum pipe_transfer_usage transfer_usage; DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__, _mesa_lookup_enum_by_nr(target), level, width, height, depth, border); @@ -680,8 +681,14 @@ st_TexImage(GLcontext * ctx, } if (stImage->pt) { + if (format == GL_DEPTH_COMPONENT && + pf_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + texImage->Data = st_texture_image_map(ctx->st, stImage, 0, - PIPE_TRANSFER_WRITE, 0, 0, + transfer_usage, 0, 0, stImage->base.Width, stImage->base.Height); if(stImage->transfer) @@ -742,7 +749,7 @@ st_TexImage(GLcontext * ctx, st_texture_image_unmap(ctx->st, stImage); /* map next slice of 3D texture */ texImage->Data = st_texture_image_map(ctx->st, stImage, i + 1, - PIPE_TRANSFER_WRITE, 0, 0, + transfer_usage, 0, 0, stImage->base.Width, stImage->base.Height); src += srcImageStride; @@ -1041,6 +1048,7 @@ st_TexSubimage(GLcontext *ctx, GLint dims, GLenum target, GLint level, _mesa_image_image_stride(packing, width, height, format, type); GLint i; const GLubyte *src; + enum pipe_transfer_usage transfer_usage; DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__, _mesa_lookup_enum_by_nr(target), @@ -1072,10 +1080,16 @@ st_TexSubimage(GLcontext *ctx, GLint dims, GLenum target, GLint level, * from uploading the buffer under us. */ if (stImage->pt) { + if (format == GL_DEPTH_COMPONENT && + pf_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + st_teximage_flush_before_map(ctx->st, stImage->pt, 0, level, - PIPE_TRANSFER_WRITE); + transfer_usage); texImage->Data = st_texture_image_map(ctx->st, stImage, zoffset, - PIPE_TRANSFER_WRITE, + transfer_usage, xoffset, yoffset, width, height); } @@ -1106,7 +1120,7 @@ st_TexSubimage(GLcontext *ctx, GLint dims, GLenum target, GLint level, /* map next slice of 3D texture */ texImage->Data = st_texture_image_map(ctx->st, stImage, zoffset + i + 1, - PIPE_TRANSFER_WRITE, + transfer_usage, xoffset, yoffset, width, height); src += srcImageStride; @@ -1274,6 +1288,7 @@ fallback_copy_texsubimage(GLcontext *ctx, GLenum target, GLint level, struct pipe_screen *screen = pipe->screen; struct pipe_transfer *src_trans; GLvoid *texDest; + enum pipe_transfer_usage transfer_usage; assert(width <= MAX_WIDTH); @@ -1288,10 +1303,16 @@ fallback_copy_texsubimage(GLcontext *ctx, GLenum target, GLint level, srcX, srcY, width, height); + if (baseFormat == GL_DEPTH_COMPONENT && + pf_is_depth_and_stencil(stImage->pt->format)) + transfer_usage = PIPE_TRANSFER_READ_WRITE; + else + transfer_usage = PIPE_TRANSFER_WRITE; + st_teximage_flush_before_map(ctx->st, stImage->pt, 0, 0, - PIPE_TRANSFER_WRITE); + transfer_usage); - texDest = st_texture_image_map(ctx->st, stImage, 0, PIPE_TRANSFER_WRITE, + texDest = st_texture_image_map(ctx->st, stImage, 0, transfer_usage, destX, destY, width, height); if (baseFormat == GL_DEPTH_COMPONENT || -- cgit v1.2.3