X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=blobdiff_plain;f=src%2Fintel%2Fvulkan%2Fgen8_cmd_buffer.c;h=92aa2b6d8c4faeb0bbf7db4c42b3c36a1dd30b74;hp=884152da20766f0132dc8b225d31b28fd6782dbc;hb=e4590c07500941ae1afa82db5e4fea2111a68604;hpb=89e421369c54a8edc4930352e062f03b5c66b886 diff --git a/src/intel/vulkan/gen8_cmd_buffer.c b/src/intel/vulkan/gen8_cmd_buffer.c index 884152da207..92aa2b6d8c4 100644 --- a/src/intel/vulkan/gen8_cmd_buffer.c +++ b/src/intel/vulkan/gen8_cmd_buffer.c @@ -31,316 +31,471 @@ #include "genxml/gen_macros.h" #include "genxml/genX_pack.h" - -static uint32_t -cmd_buffer_flush_push_constants(struct anv_cmd_buffer *cmd_buffer) -{ - static const uint32_t push_constant_opcodes[] = { - [MESA_SHADER_VERTEX] = 21, - [MESA_SHADER_TESS_CTRL] = 25, /* HS */ - [MESA_SHADER_TESS_EVAL] = 26, /* DS */ - [MESA_SHADER_GEOMETRY] = 22, - [MESA_SHADER_FRAGMENT] = 23, - [MESA_SHADER_COMPUTE] = 0, - }; - - VkShaderStageFlags flushed = 0; - - anv_foreach_stage(stage, cmd_buffer->state.push_constants_dirty) { - if (stage == MESA_SHADER_COMPUTE) - continue; - - struct anv_state state = anv_cmd_buffer_push_constants(cmd_buffer, stage); - - if (state.offset == 0) { - anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_VS), - ._3DCommandSubOpcode = push_constant_opcodes[stage]); - } else { - anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CONSTANT_VS), - ._3DCommandSubOpcode = push_constant_opcodes[stage], - .ConstantBody = { - .PointerToConstantBuffer2 = { &cmd_buffer->device->dynamic_state_block_pool.bo, state.offset }, - .ConstantBuffer2ReadLength = DIV_ROUND_UP(state.alloc_size, 32), - }); - } - } - - cmd_buffer->state.push_constants_dirty &= ~VK_SHADER_STAGE_ALL_GRAPHICS; - - return flushed; -} +#include "common/gen_guardband.h" #if GEN_GEN == 8 -static void -emit_viewport_state(struct anv_cmd_buffer *cmd_buffer, - uint32_t count, const VkViewport *viewports) +void +gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer *cmd_buffer) { + struct anv_framebuffer *fb = cmd_buffer->state.framebuffer; + uint32_t count = cmd_buffer->state.gfx.dynamic.viewport.count; + const VkViewport *viewports = + cmd_buffer->state.gfx.dynamic.viewport.viewports; struct anv_state sf_clip_state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 64, 64); - struct anv_state cc_state = - anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32); for (uint32_t i = 0; i < count; i++) { const VkViewport *vp = &viewports[i]; /* The gen7 state struct has just the matrix and guardband fields, the * gen8 struct adds the min/max viewport fields. */ - struct GENX(SF_CLIP_VIEWPORT) sf_clip_viewport = { + struct GENX(SF_CLIP_VIEWPORT) sfv = { .ViewportMatrixElementm00 = vp->width / 2, .ViewportMatrixElementm11 = vp->height / 2, - .ViewportMatrixElementm22 = 1.0, + .ViewportMatrixElementm22 = vp->maxDepth - vp->minDepth, .ViewportMatrixElementm30 = vp->x + vp->width / 2, .ViewportMatrixElementm31 = vp->y + vp->height / 2, - .ViewportMatrixElementm32 = 0.0, + .ViewportMatrixElementm32 = vp->minDepth, .XMinClipGuardband = -1.0f, .XMaxClipGuardband = 1.0f, .YMinClipGuardband = -1.0f, .YMaxClipGuardband = 1.0f, .XMinViewPort = vp->x, .XMaxViewPort = vp->x + vp->width - 1, - .YMinViewPort = vp->y, - .YMaxViewPort = vp->y + vp->height - 1, + .YMinViewPort = MIN2(vp->y, vp->y + vp->height), + .YMaxViewPort = MAX2(vp->y, vp->y + vp->height) - 1, }; - struct GENX(CC_VIEWPORT) cc_viewport = { - .MinimumDepth = vp->minDepth, - .MaximumDepth = vp->maxDepth - }; - - GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64, - &sf_clip_viewport); - GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport); - } + if (fb) { + /* We can only calculate a "real" guardband clip if we know the + * framebuffer at the time we emit the packet. Otherwise, we have + * fall back to a worst-case guardband of [-1, 1]. + */ + gen_calculate_guardband_size(fb->width, fb->height, + sfv.ViewportMatrixElementm00, + sfv.ViewportMatrixElementm11, + sfv.ViewportMatrixElementm30, + sfv.ViewportMatrixElementm31, + &sfv.XMinClipGuardband, + &sfv.XMaxClipGuardband, + &sfv.YMinClipGuardband, + &sfv.YMaxClipGuardband); + } - if (!cmd_buffer->device->info.has_llc) { - anv_state_clflush(sf_clip_state); - anv_state_clflush(cc_state); + GENX(SF_CLIP_VIEWPORT_pack)(NULL, sf_clip_state.map + i * 64, &sfv); } anv_batch_emit(&cmd_buffer->batch, - GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), - .CCViewportPointer = cc_state.offset); - anv_batch_emit(&cmd_buffer->batch, - GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), - .SFClipViewportPointer = sf_clip_state.offset); + GENX(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP), clip) { + clip.SFClipViewportPointer = sf_clip_state.offset; + } } void -gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer *cmd_buffer) +gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer *cmd_buffer, + bool depth_clamp_enable) { - if (cmd_buffer->state.dynamic.viewport.count > 0) { - emit_viewport_state(cmd_buffer, cmd_buffer->state.dynamic.viewport.count, - cmd_buffer->state.dynamic.viewport.viewports); - } else { - /* If viewport count is 0, this is taken to mean "use the default" */ - emit_viewport_state(cmd_buffer, 1, - &(VkViewport) { - .x = 0.0f, - .y = 0.0f, - .width = cmd_buffer->state.framebuffer->width, - .height = cmd_buffer->state.framebuffer->height, - .minDepth = 0.0f, - .maxDepth = 1.0f, - }); + uint32_t count = cmd_buffer->state.gfx.dynamic.viewport.count; + const VkViewport *viewports = + cmd_buffer->state.gfx.dynamic.viewport.viewports; + struct anv_state cc_state = + anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, count * 8, 32); + + for (uint32_t i = 0; i < count; i++) { + const VkViewport *vp = &viewports[i]; + + /* From the Vulkan spec: + * + * "It is valid for minDepth to be greater than or equal to + * maxDepth." + */ + float min_depth = MIN2(vp->minDepth, vp->maxDepth); + float max_depth = MAX2(vp->minDepth, vp->maxDepth); + + struct GENX(CC_VIEWPORT) cc_viewport = { + .MinimumDepth = depth_clamp_enable ? min_depth : 0.0f, + .MaximumDepth = depth_clamp_enable ? max_depth : 1.0f, + }; + + GENX(CC_VIEWPORT_pack)(NULL, cc_state.map + i * 8, &cc_viewport); + } + + anv_batch_emit(&cmd_buffer->batch, + GENX(3DSTATE_VIEWPORT_STATE_POINTERS_CC), cc) { + cc.CCViewportPointer = cc_state.offset; } } #endif -static void -emit_lri(struct anv_batch *batch, uint32_t reg, uint32_t imm) +void +genX(cmd_buffer_enable_pma_fix)(struct anv_cmd_buffer *cmd_buffer, bool enable) { - anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_IMM), - .RegisterOffset = reg, - .DataDWord = imm); -} + if (cmd_buffer->state.pma_fix_enabled == enable) + return; -#define GEN8_L3CNTLREG 0x7034 + cmd_buffer->state.pma_fix_enabled = enable; -static void -config_l3(struct anv_cmd_buffer *cmd_buffer, bool enable_slm) -{ - /* References for GL state: + /* According to the Broadwell PIPE_CONTROL documentation, software should + * emit a PIPE_CONTROL with the CS Stall and Depth Cache Flush bits set + * prior to the LRI. If stencil buffer writes are enabled, then a Render + * Cache Flush is also necessary. * - * - commits e307cfa..228d5a3 - * - src/mesa/drivers/dri/i965/gen7_l3_state.c + * The Skylake docs say to use a depth stall rather than a command + * streamer stall. However, the hardware seems to violently disagree. + * A full command streamer stall seems to be needed in both cases. */ - - uint32_t val = enable_slm ? - /* All = 48 ways; URB = 16 ways; DC and RO = 0, SLM = 1 */ - 0x60000021 : - /* All = 48 ways; URB = 48 ways; DC, RO and SLM = 0 */ - 0x60000060; - bool changed = cmd_buffer->state.current_l3_config != val; - - if (changed) { - /* According to the hardware docs, the L3 partitioning can only be changed - * while the pipeline is completely drained and the caches are flushed, - * which involves a first PIPE_CONTROL flush which stalls the pipeline and - * initiates invalidation of the relevant caches... + anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) { + pc.DepthCacheFlushEnable = true; + pc.CommandStreamerStallEnable = true; + pc.RenderTargetCacheFlushEnable = true; +#if GEN_GEN >= 12 + pc.TileCacheFlushEnable = true; + + /* GEN:BUG:1409600907: "PIPE_CONTROL with Depth Stall Enable bit must + * be set with any PIPE_CONTROL with Depth Flush Enable bit set. */ - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .TextureCacheInvalidationEnable = true, - .ConstantCacheInvalidationEnable = true, - .InstructionCacheInvalidateEnable = true, - .DCFlushEnable = true, - .PostSyncOperation = NoWrite, - .CommandStreamerStallEnable = true); - - /* ...followed by a second stalling flush which guarantees that - * invalidation is complete when the L3 configuration registers are - * modified. - */ - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .DCFlushEnable = true, - .PostSyncOperation = NoWrite, - .CommandStreamerStallEnable = true); + pc.DepthStallEnable = true; +#endif + } + +#if GEN_GEN == 9 - emit_lri(&cmd_buffer->batch, GEN8_L3CNTLREG, val); - cmd_buffer->state.current_l3_config = val; + uint32_t cache_mode; + anv_pack_struct(&cache_mode, GENX(CACHE_MODE_0), + .STCPMAOptimizationEnable = enable, + .STCPMAOptimizationEnableMask = true); + anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) { + lri.RegisterOffset = GENX(CACHE_MODE_0_num); + lri.DataDWord = cache_mode; } -} -static void -__emit_genx_sf_state(struct anv_cmd_buffer *cmd_buffer) -{ - uint32_t sf_dw[GENX(3DSTATE_SF_length)]; - struct GENX(3DSTATE_SF) sf = { - GENX(3DSTATE_SF_header), - .LineWidth = cmd_buffer->state.dynamic.line_width, - }; - GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf); - /* FIXME: gen9.fs */ - anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, - cmd_buffer->state.pipeline->gen8.sf); -} +#elif GEN_GEN == 8 + + uint32_t cache_mode; + anv_pack_struct(&cache_mode, GENX(CACHE_MODE_1), + .NPPMAFixEnable = enable, + .NPEarlyZFailsDisable = enable, + .NPPMAFixEnableMask = true, + .NPEarlyZFailsDisableMask = true); + anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) { + lri.RegisterOffset = GENX(CACHE_MODE_1_num); + lri.DataDWord = cache_mode; + } -#include "genxml/gen9_pack.h" -static void -__emit_gen9_sf_state(struct anv_cmd_buffer *cmd_buffer) -{ - uint32_t sf_dw[GENX(3DSTATE_SF_length)]; - struct GEN9_3DSTATE_SF sf = { - GEN9_3DSTATE_SF_header, - .LineWidth = cmd_buffer->state.dynamic.line_width, - }; - GEN9_3DSTATE_SF_pack(NULL, sf_dw, &sf); - /* FIXME: gen9.fs */ - anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, - cmd_buffer->state.pipeline->gen8.sf); -} +#endif /* GEN_GEN == 8 */ -static void -__emit_sf_state(struct anv_cmd_buffer *cmd_buffer) -{ - if (cmd_buffer->device->info.is_cherryview) - __emit_gen9_sf_state(cmd_buffer); - else - __emit_genx_sf_state(cmd_buffer); + /* After the LRI, a PIPE_CONTROL with both the Depth Stall and Depth Cache + * Flush bits is often necessary. We do it regardless because it's easier. + * The render cache flush is also necessary if stencil writes are enabled. + * + * Again, the Skylake docs give a different set of flushes but the BDW + * flushes seem to work just as well. + */ + anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), pc) { + pc.DepthStallEnable = true; + pc.DepthCacheFlushEnable = true; + pc.RenderTargetCacheFlushEnable = true; +#if GEN_GEN >= 12 + pc.TileCacheFlushEnable = true; +#endif + } } -void -genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer) +UNUSED static bool +want_depth_pma_fix(struct anv_cmd_buffer *cmd_buffer) { - struct anv_pipeline *pipeline = cmd_buffer->state.pipeline; - uint32_t *p; - - uint32_t vb_emit = cmd_buffer->state.vb_dirty & pipeline->vb_used; + assert(GEN_GEN == 8); - assert((pipeline->active_stages & VK_SHADER_STAGE_COMPUTE_BIT) == 0); - - config_l3(cmd_buffer, false); - - genX(flush_pipeline_select_3d)(cmd_buffer); + /* From the Broadwell PRM Vol. 2c CACHE_MODE_1::NP_PMA_FIX_ENABLE: + * + * SW must set this bit in order to enable this fix when following + * expression is TRUE. + * + * 3DSTATE_WM::ForceThreadDispatch != 1 && + * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) && + * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) && + * (3DSTATE_DEPTH_BUFFER::HIZ Enable) && + * !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) && + * (3DSTATE_PS_EXTRA::PixelShaderValid) && + * !(3DSTATE_WM_HZ_OP::DepthBufferClear || + * 3DSTATE_WM_HZ_OP::DepthBufferResolve || + * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable || + * 3DSTATE_WM_HZ_OP::StencilBufferClear) && + * (3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable) && + * (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels || + * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget || + * 3DSTATE_PS_BLEND::AlphaToCoverageEnable || + * 3DSTATE_PS_BLEND::AlphaTestEnable || + * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) && + * 3DSTATE_WM::ForceKillPix != ForceOff && + * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable && + * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) || + * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable && + * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE && + * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) || + * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF)) + */ - if (vb_emit) { - const uint32_t num_buffers = __builtin_popcount(vb_emit); - const uint32_t num_dwords = 1 + num_buffers * 4; + /* These are always true: + * 3DSTATE_WM::ForceThreadDispatch != 1 && + * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) + */ - p = anv_batch_emitn(&cmd_buffer->batch, num_dwords, - GENX(3DSTATE_VERTEX_BUFFERS)); - uint32_t vb, i = 0; - for_each_bit(vb, vb_emit) { - struct anv_buffer *buffer = cmd_buffer->state.vertex_bindings[vb].buffer; - uint32_t offset = cmd_buffer->state.vertex_bindings[vb].offset; + /* We only enable the PMA fix if we know for certain that HiZ is enabled. + * If we don't know whether HiZ is enabled or not, we disable the PMA fix + * and there is no harm. + * + * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) && + * 3DSTATE_DEPTH_BUFFER::HIZ Enable + */ + if (!cmd_buffer->state.hiz_enabled) + return false; + + /* 3DSTATE_PS_EXTRA::PixelShaderValid */ + struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline; + if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT)) + return false; + + /* !(3DSTATE_WM::EDSC_Mode == EDSC_PREPS) */ + const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline); + if (wm_prog_data->early_fragment_tests) + return false; + + /* We never use anv_pipeline for HiZ ops so this is trivially true: + * !(3DSTATE_WM_HZ_OP::DepthBufferClear || + * 3DSTATE_WM_HZ_OP::DepthBufferResolve || + * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable || + * 3DSTATE_WM_HZ_OP::StencilBufferClear) + */ - struct GENX(VERTEX_BUFFER_STATE) state = { - .VertexBufferIndex = vb, - .MemoryObjectControlState = GENX(MOCS), - .AddressModifyEnable = true, - .BufferPitch = pipeline->binding_stride[vb], - .BufferStartingAddress = { buffer->bo, buffer->offset + offset }, - .BufferSize = buffer->size - offset - }; + /* 3DSTATE_WM_DEPTH_STENCIL::DepthTestEnable */ + if (!pipeline->depth_test_enable) + return false; + + /* (((3DSTATE_PS_EXTRA::PixelShaderKillsPixels || + * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget || + * 3DSTATE_PS_BLEND::AlphaToCoverageEnable || + * 3DSTATE_PS_BLEND::AlphaTestEnable || + * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) && + * 3DSTATE_WM::ForceKillPix != ForceOff && + * ((3DSTATE_WM_DEPTH_STENCIL::DepthWriteEnable && + * 3DSTATE_DEPTH_BUFFER::DEPTH_WRITE_ENABLE) || + * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable && + * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE && + * 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE))) || + * (3DSTATE_PS_EXTRA:: Pixel Shader Computed Depth mode != PSCDEPTH_OFF)) + */ + return (pipeline->kill_pixel && (pipeline->writes_depth || + pipeline->writes_stencil)) || + wm_prog_data->computed_depth_mode != PSCDEPTH_OFF; +} - GENX(VERTEX_BUFFER_STATE_pack)(&cmd_buffer->batch, &p[1 + i * 4], &state); - i++; - } - } +UNUSED static bool +want_stencil_pma_fix(struct anv_cmd_buffer *cmd_buffer) +{ + if (GEN_GEN > 9) + return false; + assert(GEN_GEN == 9); - if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_PIPELINE) { - /* If somebody compiled a pipeline after starting a command buffer the - * scratch bo may have grown since we started this cmd buffer (and - * emitted STATE_BASE_ADDRESS). If we're binding that pipeline now, - * reemit STATE_BASE_ADDRESS so that we use the bigger scratch bo. */ - if (cmd_buffer->state.scratch_size < pipeline->total_scratch) - anv_cmd_buffer_emit_state_base_address(cmd_buffer); + /* From the Skylake PRM Vol. 2c CACHE_MODE_1::STC PMA Optimization Enable: + * + * Clearing this bit will force the STC cache to wait for pending + * retirement of pixels at the HZ-read stage and do the STC-test for + * Non-promoted, R-computed and Computed depth modes instead of + * postponing the STC-test to RCPFE. + * + * STC_TEST_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE && + * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable + * + * STC_WRITE_EN = 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE && + * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable && + * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE) + * + * COMP_STC_EN = STC_TEST_EN && + * 3DSTATE_PS_EXTRA::PixelShaderComputesStencil + * + * SW parses the pipeline states to generate the following logical + * signal indicating if PMA FIX can be enabled. + * + * STC_PMA_OPT = + * 3DSTATE_WM::ForceThreadDispatch != 1 && + * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) && + * 3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL && + * 3DSTATE_DEPTH_BUFFER::HIZ Enable && + * !(3DSTATE_WM::EDSC_Mode == 2) && + * 3DSTATE_PS_EXTRA::PixelShaderValid && + * !(3DSTATE_WM_HZ_OP::DepthBufferClear || + * 3DSTATE_WM_HZ_OP::DepthBufferResolve || + * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable || + * 3DSTATE_WM_HZ_OP::StencilBufferClear) && + * (COMP_STC_EN || STC_WRITE_EN) && + * ((3DSTATE_PS_EXTRA::PixelShaderKillsPixels || + * 3DSTATE_WM::ForceKillPix == ON || + * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget || + * 3DSTATE_PS_BLEND::AlphaToCoverageEnable || + * 3DSTATE_PS_BLEND::AlphaTestEnable || + * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) || + * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF)) + */ - anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch); + /* These are always true: + * 3DSTATE_WM::ForceThreadDispatch != 1 && + * !(3DSTATE_RASTER::ForceSampleCount != NUMRASTSAMPLES_0) + */ - /* From the BDW PRM for 3DSTATE_PUSH_CONSTANT_ALLOC_VS: - * - * "The 3DSTATE_CONSTANT_VS must be reprogrammed prior to - * the next 3DPRIMITIVE command after programming the - * 3DSTATE_PUSH_CONSTANT_ALLOC_VS" - * - * Since 3DSTATE_PUSH_CONSTANT_ALLOC_VS is programmed as part of - * pipeline setup, we need to dirty push constants. - */ - cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS; - } + /* We only enable the PMA fix if we know for certain that HiZ is enabled. + * If we don't know whether HiZ is enabled or not, we disable the PMA fix + * and there is no harm. + * + * (3DSTATE_DEPTH_BUFFER::SURFACE_TYPE != NULL) && + * 3DSTATE_DEPTH_BUFFER::HIZ Enable + */ + if (!cmd_buffer->state.hiz_enabled) + return false; + + /* We can't possibly know if HiZ is enabled without the framebuffer */ + assert(cmd_buffer->state.framebuffer); + + /* HiZ is enabled so we had better have a depth buffer with HiZ */ + const struct anv_image_view *ds_iview = + anv_cmd_buffer_get_depth_stencil_view(cmd_buffer); + assert(ds_iview && ds_iview->image->planes[0].aux_usage == ISL_AUX_USAGE_HIZ); + + /* 3DSTATE_PS_EXTRA::PixelShaderValid */ + struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline; + if (!anv_pipeline_has_stage(pipeline, MESA_SHADER_FRAGMENT)) + return false; + + /* !(3DSTATE_WM::EDSC_Mode == 2) */ + const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline); + if (wm_prog_data->early_fragment_tests) + return false; + + /* We never use anv_pipeline for HiZ ops so this is trivially true: + * !(3DSTATE_WM_HZ_OP::DepthBufferClear || + * 3DSTATE_WM_HZ_OP::DepthBufferResolve || + * 3DSTATE_WM_HZ_OP::Hierarchical Depth Buffer Resolve Enable || + * 3DSTATE_WM_HZ_OP::StencilBufferClear) + */ - /* We emit the binding tables and sampler tables first, then emit push - * constants and then finally emit binding table and sampler table - * pointers. It has to happen in this order, since emitting the binding - * tables may change the push constants (in case of storage images). After - * emitting push constants, on SKL+ we have to emit the corresponding - * 3DSTATE_BINDING_TABLE_POINTER_* for the push constants to take effect. + /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE && + * 3DSTATE_WM_DEPTH_STENCIL::StencilTestEnable */ - uint32_t dirty = 0; - if (cmd_buffer->state.descriptors_dirty) - dirty = gen7_cmd_buffer_flush_descriptor_sets(cmd_buffer); + const bool stc_test_en = + (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) && + pipeline->stencil_test_enable; - if (cmd_buffer->state.push_constants_dirty) - dirty |= cmd_buffer_flush_push_constants(cmd_buffer); + /* 3DSTATE_STENCIL_BUFFER::STENCIL_BUFFER_ENABLE && + * (3DSTATE_WM_DEPTH_STENCIL::Stencil Buffer Write Enable && + * 3DSTATE_DEPTH_BUFFER::STENCIL_WRITE_ENABLE) + */ + const bool stc_write_en = + (ds_iview->image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) && + (cmd_buffer->state.gfx.dynamic.stencil_write_mask.front || + cmd_buffer->state.gfx.dynamic.stencil_write_mask.back) && + pipeline->writes_stencil; + + /* STC_TEST_EN && 3DSTATE_PS_EXTRA::PixelShaderComputesStencil */ + const bool comp_stc_en = stc_test_en && wm_prog_data->computed_stencil; + + /* COMP_STC_EN || STC_WRITE_EN */ + if (!(comp_stc_en || stc_write_en)) + return false; + + /* (3DSTATE_PS_EXTRA::PixelShaderKillsPixels || + * 3DSTATE_WM::ForceKillPix == ON || + * 3DSTATE_PS_EXTRA::oMask Present to RenderTarget || + * 3DSTATE_PS_BLEND::AlphaToCoverageEnable || + * 3DSTATE_PS_BLEND::AlphaTestEnable || + * 3DSTATE_WM_CHROMAKEY::ChromaKeyKillEnable) || + * (3DSTATE_PS_EXTRA::Pixel Shader Computed Depth mode != PSCDEPTH_OFF) + */ + return pipeline->kill_pixel || + wm_prog_data->computed_depth_mode != PSCDEPTH_OFF; +} - if (dirty) - gen7_cmd_buffer_emit_descriptor_pointers(cmd_buffer, dirty); +void +genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer) +{ + struct anv_graphics_pipeline *pipeline = cmd_buffer->state.gfx.pipeline; + struct anv_dynamic_state *d = &cmd_buffer->state.gfx.dynamic; - if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_VIEWPORT) - gen8_cmd_buffer_emit_viewport(cmd_buffer); + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) { + uint32_t sf_dw[GENX(3DSTATE_SF_length)]; + struct GENX(3DSTATE_SF) sf = { + GENX(3DSTATE_SF_header), + }; +#if GEN_GEN == 8 + if (cmd_buffer->device->info.is_cherryview) { + sf.CHVLineWidth = d->line_width; + } else { + sf.LineWidth = d->line_width; + } +#else + sf.LineWidth = d->line_width, +#endif + GENX(3DSTATE_SF_pack)(NULL, sf_dw, &sf); + anv_batch_emit_merge(&cmd_buffer->batch, sf_dw, pipeline->gen8.sf); + } - if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_SCISSOR) - gen7_cmd_buffer_emit_scissor(cmd_buffer); + static const uint32_t vk_to_gen_cullmode[] = { + [VK_CULL_MODE_NONE] = CULLMODE_NONE, + [VK_CULL_MODE_FRONT_BIT] = CULLMODE_FRONT, + [VK_CULL_MODE_BACK_BIT] = CULLMODE_BACK, + [VK_CULL_MODE_FRONT_AND_BACK] = CULLMODE_BOTH + }; + static const uint32_t vk_to_gen_front_face[] = { + [VK_FRONT_FACE_COUNTER_CLOCKWISE] = 1, + [VK_FRONT_FACE_CLOCKWISE] = 0 + }; + static const uint32_t vk_to_gen_primitive_type[] = { + [VK_PRIMITIVE_TOPOLOGY_POINT_LIST] = _3DPRIM_POINTLIST, + [VK_PRIMITIVE_TOPOLOGY_LINE_LIST] = _3DPRIM_LINELIST, + [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP] = _3DPRIM_LINESTRIP, + [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST] = _3DPRIM_TRILIST, + [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP] = _3DPRIM_TRISTRIP, + [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN] = _3DPRIM_TRIFAN, + [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY] = _3DPRIM_LINELIST_ADJ, + [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY] = _3DPRIM_LINESTRIP_ADJ, + [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY] = _3DPRIM_TRILIST_ADJ, + [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY] = _3DPRIM_TRISTRIP_ADJ, + }; - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | - ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)) { - __emit_sf_state(cmd_buffer); - } + static const uint32_t vk_to_gen_compare_op[] = { + [VK_COMPARE_OP_NEVER] = PREFILTEROPNEVER, + [VK_COMPARE_OP_LESS] = PREFILTEROPLESS, + [VK_COMPARE_OP_EQUAL] = PREFILTEROPEQUAL, + [VK_COMPARE_OP_LESS_OR_EQUAL] = PREFILTEROPLEQUAL, + [VK_COMPARE_OP_GREATER] = PREFILTEROPGREATER, + [VK_COMPARE_OP_NOT_EQUAL] = PREFILTEROPNOTEQUAL, + [VK_COMPARE_OP_GREATER_OR_EQUAL] = PREFILTEROPGEQUAL, + [VK_COMPARE_OP_ALWAYS] = PREFILTEROPALWAYS, + }; - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | - ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)){ - bool enable_bias = cmd_buffer->state.dynamic.depth_bias.bias != 0.0f || - cmd_buffer->state.dynamic.depth_bias.slope != 0.0f; + static const uint32_t vk_to_gen_stencil_op[] = { + [VK_STENCIL_OP_KEEP] = STENCILOP_KEEP, + [VK_STENCIL_OP_ZERO] = STENCILOP_ZERO, + [VK_STENCIL_OP_REPLACE] = STENCILOP_REPLACE, + [VK_STENCIL_OP_INCREMENT_AND_CLAMP] = STENCILOP_INCRSAT, + [VK_STENCIL_OP_DECREMENT_AND_CLAMP] = STENCILOP_DECRSAT, + [VK_STENCIL_OP_INVERT] = STENCILOP_INVERT, + [VK_STENCIL_OP_INCREMENT_AND_WRAP] = STENCILOP_INCR, + [VK_STENCIL_OP_DECREMENT_AND_WRAP] = STENCILOP_DECR, + }; + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS | + ANV_CMD_DIRTY_DYNAMIC_CULL_MODE | + ANV_CMD_DIRTY_DYNAMIC_FRONT_FACE)) { uint32_t raster_dw[GENX(3DSTATE_RASTER_length)]; struct GENX(3DSTATE_RASTER) raster = { GENX(3DSTATE_RASTER_header), - .GlobalDepthOffsetEnableSolid = enable_bias, - .GlobalDepthOffsetEnableWireframe = enable_bias, - .GlobalDepthOffsetEnablePoint = enable_bias, - .GlobalDepthOffsetConstant = cmd_buffer->state.dynamic.depth_bias.bias, - .GlobalDepthOffsetScale = cmd_buffer->state.dynamic.depth_bias.slope, - .GlobalDepthOffsetClamp = cmd_buffer->state.dynamic.depth_bias.clamp + .GlobalDepthOffsetConstant = d->depth_bias.bias, + .GlobalDepthOffsetScale = d->depth_bias.slope, + .GlobalDepthOffsetClamp = d->depth_bias.clamp, + .CullMode = vk_to_gen_cullmode[d->cull_mode], + .FrontWinding = vk_to_gen_front_face[d->front_face], }; GENX(3DSTATE_RASTER_pack)(NULL, raster_dw, &raster); anv_batch_emit_merge(&cmd_buffer->batch, raster_dw, @@ -353,90 +508,107 @@ genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer) * using a big old #if switch here. */ #if GEN_GEN == 8 - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) { - struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) { struct anv_state cc_state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, GENX(COLOR_CALC_STATE_length) * 4, 64); struct GENX(COLOR_CALC_STATE) cc = { - .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0], - .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1], - .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2], - .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3], - .StencilReferenceValue = d->stencil_reference.front, - .BackFaceStencilReferenceValue = d->stencil_reference.back, + .BlendConstantColorRed = d->blend_constants[0], + .BlendConstantColorGreen = d->blend_constants[1], + .BlendConstantColorBlue = d->blend_constants[2], + .BlendConstantColorAlpha = d->blend_constants[3], + .StencilReferenceValue = d->stencil_reference.front & 0xff, + .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff, }; GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc); - if (!cmd_buffer->device->info.has_llc) - anv_state_clflush(cc_state); - - anv_batch_emit(&cmd_buffer->batch, - GENX(3DSTATE_CC_STATE_POINTERS), - .ColorCalcStatePointer = cc_state.offset, - .ColorCalcStatePointerValid = true); + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) { + ccp.ColorCalcStatePointer = cc_state.offset; + ccp.ColorCalcStatePointerValid = true; + } } - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK)) { + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_RENDER_TARGETS | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_OP)) { uint32_t wm_depth_stencil_dw[GENX(3DSTATE_WM_DEPTH_STENCIL_length)]; - struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; struct GENX(3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil) = { GENX(3DSTATE_WM_DEPTH_STENCIL_header), - .StencilBufferWriteEnable = d->stencil_write_mask.front != 0 || - d->stencil_write_mask.back != 0, - .StencilTestMask = d->stencil_compare_mask.front & 0xff, .StencilWriteMask = d->stencil_write_mask.front & 0xff, .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff, .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff, + + .StencilBufferWriteEnable = + (d->stencil_write_mask.front || d->stencil_write_mask.back) && + d->stencil_test_enable, + + .DepthTestEnable = d->depth_test_enable, + .DepthBufferWriteEnable = d->depth_test_enable && d->depth_write_enable, + .DepthTestFunction = vk_to_gen_compare_op[d->depth_compare_op], + .StencilTestEnable = d->stencil_test_enable, + .StencilFailOp = vk_to_gen_stencil_op[d->stencil_op.front.fail_op], + .StencilPassDepthPassOp = vk_to_gen_stencil_op[d->stencil_op.front.pass_op], + .StencilPassDepthFailOp = vk_to_gen_stencil_op[d->stencil_op.front.depth_fail_op], + .StencilTestFunction = vk_to_gen_compare_op[d->stencil_op.front.compare_op], + .BackfaceStencilFailOp = vk_to_gen_stencil_op[d->stencil_op.back.fail_op], + .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[d->stencil_op.back.pass_op], + .BackfaceStencilPassDepthFailOp = vk_to_gen_stencil_op[d->stencil_op.back.depth_fail_op], + .BackfaceStencilTestFunction = vk_to_gen_compare_op[d->stencil_op.back.compare_op], }; GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, wm_depth_stencil_dw, &wm_depth_stencil); anv_batch_emit_merge(&cmd_buffer->batch, wm_depth_stencil_dw, pipeline->gen8.wm_depth_stencil); + + genX(cmd_buffer_enable_pma_fix)(cmd_buffer, + want_depth_pma_fix(cmd_buffer)); } #else - if (cmd_buffer->state.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) { + if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) { struct anv_state cc_state = anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, - GEN9_COLOR_CALC_STATE_length * 4, + GENX(COLOR_CALC_STATE_length) * 4, 64); - struct GEN9_COLOR_CALC_STATE cc = { - .BlendConstantColorRed = cmd_buffer->state.dynamic.blend_constants[0], - .BlendConstantColorGreen = cmd_buffer->state.dynamic.blend_constants[1], - .BlendConstantColorBlue = cmd_buffer->state.dynamic.blend_constants[2], - .BlendConstantColorAlpha = cmd_buffer->state.dynamic.blend_constants[3], + struct GENX(COLOR_CALC_STATE) cc = { + .BlendConstantColorRed = d->blend_constants[0], + .BlendConstantColorGreen = d->blend_constants[1], + .BlendConstantColorBlue = d->blend_constants[2], + .BlendConstantColorAlpha = d->blend_constants[3], }; - GEN9_COLOR_CALC_STATE_pack(NULL, cc_state.map, &cc); - - if (!cmd_buffer->device->info.has_llc) - anv_state_clflush(cc_state); + GENX(COLOR_CALC_STATE_pack)(NULL, cc_state.map, &cc); - anv_batch_emit(&cmd_buffer->batch, - GEN9_3DSTATE_CC_STATE_POINTERS, - .ColorCalcStatePointer = cc_state.offset, - .ColorCalcStatePointerValid = true); + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_CC_STATE_POINTERS), ccp) { + ccp.ColorCalcStatePointer = cc_state.offset; + ccp.ColorCalcStatePointerValid = true; + } } - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | - ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE)) { - uint32_t dwords[GEN9_3DSTATE_WM_DEPTH_STENCIL_length]; - struct anv_dynamic_state *d = &cmd_buffer->state.dynamic; - struct GEN9_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = { - GEN9_3DSTATE_WM_DEPTH_STENCIL_header, - - .StencilBufferWriteEnable = d->stencil_write_mask.front != 0 || - d->stencil_write_mask.back != 0, + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_RENDER_TARGETS | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE | + ANV_CMD_DIRTY_DYNAMIC_STENCIL_OP)) { + uint32_t dwords[GENX(3DSTATE_WM_DEPTH_STENCIL_length)]; + struct GENX(3DSTATE_WM_DEPTH_STENCIL) wm_depth_stencil = { + GENX(3DSTATE_WM_DEPTH_STENCIL_header), .StencilTestMask = d->stencil_compare_mask.front & 0xff, .StencilWriteMask = d->stencil_write_mask.front & 0xff, @@ -444,479 +616,132 @@ genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer) .BackfaceStencilTestMask = d->stencil_compare_mask.back & 0xff, .BackfaceStencilWriteMask = d->stencil_write_mask.back & 0xff, - .StencilReferenceValue = d->stencil_reference.front, - .BackfaceStencilReferenceValue = d->stencil_reference.back + .StencilReferenceValue = d->stencil_reference.front & 0xff, + .BackfaceStencilReferenceValue = d->stencil_reference.back & 0xff, + + .StencilBufferWriteEnable = + (d->stencil_write_mask.front || d->stencil_write_mask.back) && + d->stencil_test_enable, + + .DepthTestEnable = d->depth_test_enable, + .DepthBufferWriteEnable = d->depth_test_enable && d->depth_write_enable, + .DepthTestFunction = vk_to_gen_compare_op[d->depth_compare_op], + .StencilTestEnable = d->stencil_test_enable, + .StencilFailOp = vk_to_gen_stencil_op[d->stencil_op.front.fail_op], + .StencilPassDepthPassOp = vk_to_gen_stencil_op[d->stencil_op.front.pass_op], + .StencilPassDepthFailOp = vk_to_gen_stencil_op[d->stencil_op.front.depth_fail_op], + .StencilTestFunction = vk_to_gen_compare_op[d->stencil_op.front.compare_op], + .BackfaceStencilFailOp = vk_to_gen_stencil_op[d->stencil_op.back.fail_op], + .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[d->stencil_op.back.pass_op], + .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[d->stencil_op.back.depth_fail_op], + .BackfaceStencilTestFunction = vk_to_gen_compare_op[d->stencil_op.back.compare_op], + }; - GEN9_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, dwords, &wm_depth_stencil); + GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dwords, &wm_depth_stencil); anv_batch_emit_merge(&cmd_buffer->batch, dwords, pipeline->gen9.wm_depth_stencil); - } -#endif - - if (cmd_buffer->state.dirty & (ANV_CMD_DIRTY_PIPELINE | - ANV_CMD_DIRTY_INDEX_BUFFER)) { - anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), - .IndexedDrawCutIndexEnable = pipeline->primitive_restart, - .CutIndex = cmd_buffer->state.restart_index, - ); - } - - cmd_buffer->state.vb_dirty &= ~vb_emit; - cmd_buffer->state.dirty = 0; -} - -void genX(CmdBindIndexBuffer)( - VkCommandBuffer commandBuffer, - VkBuffer _buffer, - VkDeviceSize offset, - VkIndexType indexType) -{ - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_buffer, buffer, _buffer); - - static const uint32_t vk_to_gen_index_type[] = { - [VK_INDEX_TYPE_UINT16] = INDEX_WORD, - [VK_INDEX_TYPE_UINT32] = INDEX_DWORD, - }; - - static const uint32_t restart_index_for_type[] = { - [VK_INDEX_TYPE_UINT16] = UINT16_MAX, - [VK_INDEX_TYPE_UINT32] = UINT32_MAX, - }; - cmd_buffer->state.restart_index = restart_index_for_type[indexType]; - - anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), - .IndexFormat = vk_to_gen_index_type[indexType], - .MemoryObjectControlState = GENX(MOCS), - .BufferStartingAddress = { buffer->bo, buffer->offset + offset }, - .BufferSize = buffer->size - offset); - - cmd_buffer->state.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER; -} - -static VkResult -flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer) -{ - struct anv_device *device = cmd_buffer->device; - struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline; - struct anv_state surfaces = { 0, }, samplers = { 0, }; - VkResult result; - - result = anv_cmd_buffer_emit_samplers(cmd_buffer, - MESA_SHADER_COMPUTE, &samplers); - if (result != VK_SUCCESS) - return result; - result = anv_cmd_buffer_emit_binding_table(cmd_buffer, - MESA_SHADER_COMPUTE, &surfaces); - if (result != VK_SUCCESS) - return result; - - struct anv_state push_state = anv_cmd_buffer_cs_push_constants(cmd_buffer); - - const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data; - const struct brw_stage_prog_data *prog_data = &cs_prog_data->base; - - unsigned local_id_dwords = cs_prog_data->local_invocation_id_regs * 8; - unsigned push_constant_data_size = - (prog_data->nr_params + local_id_dwords) * 4; - unsigned reg_aligned_constant_size = ALIGN(push_constant_data_size, 32); - unsigned push_constant_regs = reg_aligned_constant_size / 32; - - if (push_state.alloc_size) { - anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), - .CURBETotalDataLength = push_state.alloc_size, - .CURBEDataStartAddress = push_state.offset); + genX(cmd_buffer_enable_pma_fix)(cmd_buffer, + want_stencil_pma_fix(cmd_buffer)); } +#endif - assert(prog_data->total_shared <= 64 * 1024); - uint32_t slm_size = 0; - if (prog_data->total_shared > 0) { - /* slm_size is in 4k increments, but must be a power of 2. */ - slm_size = 4 * 1024; - while (slm_size < prog_data->total_shared) - slm_size <<= 1; - slm_size /= 4 * 1024; +#if GEN_GEN >= 12 + if(cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS | + ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE)) { + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_DEPTH_BOUNDS), db) { + db.DepthBoundsTestValueModifyDisable = false; + db.DepthBoundsTestEnableModifyDisable = false; + db.DepthBoundsTestEnable = d->depth_bounds_test_enable; + db.DepthBoundsTestMinValue = d->depth_bounds.min; + db.DepthBoundsTestMaxValue = d->depth_bounds.max; + } } - - struct anv_state state = - anv_state_pool_emit(&device->dynamic_state_pool, - GENX(INTERFACE_DESCRIPTOR_DATA), 64, - .KernelStartPointer = pipeline->cs_simd, - .KernelStartPointerHigh = 0, - .BindingTablePointer = surfaces.offset, - .BindingTableEntryCount = 0, - .SamplerStatePointer = samplers.offset, - .SamplerCount = 0, - .ConstantIndirectURBEntryReadLength = push_constant_regs, - .ConstantURBEntryReadOffset = 0, - .BarrierEnable = cs_prog_data->uses_barrier, - .SharedLocalMemorySize = slm_size, - .NumberofThreadsinGPGPUThreadGroup = - pipeline->cs_thread_width_max); - - uint32_t size = GENX(INTERFACE_DESCRIPTOR_DATA_length) * sizeof(uint32_t); - anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_INTERFACE_DESCRIPTOR_LOAD), - .InterfaceDescriptorTotalLength = size, - .InterfaceDescriptorDataStartAddress = state.offset); - - return VK_SUCCESS; -} - -void -genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer) -{ - struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline; - VkResult result; - - assert(pipeline->active_stages == VK_SHADER_STAGE_COMPUTE_BIT); - - bool needs_slm = pipeline->cs_prog_data.base.total_shared > 0; - config_l3(cmd_buffer, needs_slm); - - if (cmd_buffer->state.current_pipeline != GPGPU) { -#if GEN_GEN < 10 - /* From the Broadwell PRM, Volume 2a: Instructions, PIPELINE_SELECT: - * - * Software must clear the COLOR_CALC_STATE Valid field in - * 3DSTATE_CC_STATE_POINTERS command prior to send a PIPELINE_SELECT - * with Pipeline Select set to GPGPU. - * - * The internal hardware docs recommend the same workaround for Gen9 - * hardware too. - */ - anv_batch_emit(&cmd_buffer->batch, - GENX(3DSTATE_CC_STATE_POINTERS)); #endif - anv_batch_emit(&cmd_buffer->batch, GENX(PIPELINE_SELECT), -#if GEN_GEN >= 9 - .MaskBits = 3, -#endif - .PipelineSelection = GPGPU); - cmd_buffer->state.current_pipeline = GPGPU; + if (cmd_buffer->state.gfx.dirty & ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE) { + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_LINE_STIPPLE), ls) { + ls.LineStipplePattern = d->line_stipple.pattern; + ls.LineStippleInverseRepeatCount = + 1.0f / MAX2(1, d->line_stipple.factor); + ls.LineStippleRepeatCount = d->line_stipple.factor; + } } - if (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE) - anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch); - - if ((cmd_buffer->state.descriptors_dirty & VK_SHADER_STAGE_COMPUTE_BIT) || - (cmd_buffer->state.compute_dirty & ANV_CMD_DIRTY_PIPELINE)) { - result = flush_compute_descriptor_set(cmd_buffer); - assert(result == VK_SUCCESS); - cmd_buffer->state.descriptors_dirty &= ~VK_SHADER_STAGE_COMPUTE_BIT; + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_INDEX_BUFFER)) { + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF), vf) { + vf.IndexedDrawCutIndexEnable = pipeline->primitive_restart; + vf.CutIndex = cmd_buffer->state.restart_index; + } } - cmd_buffer->state.compute_dirty = 0; -} - -static void -emit_ps_depth_count(struct anv_batch *batch, - struct anv_bo *bo, uint32_t offset) -{ - anv_batch_emit(batch, GENX(PIPE_CONTROL), - .DestinationAddressType = DAT_PPGTT, - .PostSyncOperation = WritePSDepthCount, - .DepthStallEnable = true, - .Address = { bo, offset }); -} + if (cmd_buffer->state.gfx.dirty & (ANV_CMD_DIRTY_PIPELINE | + ANV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY)) { + uint32_t topology; + if (anv_pipeline_has_stage(pipeline, MESA_SHADER_TESS_EVAL)) + topology = d->primitive_topology; + else + topology = vk_to_gen_primitive_type[d->primitive_topology]; -static void -emit_query_availability(struct anv_batch *batch, - struct anv_bo *bo, uint32_t offset) -{ - anv_batch_emit(batch, GENX(PIPE_CONTROL), - .DestinationAddressType = DAT_PPGTT, - .PostSyncOperation = WriteImmediateData, - .Address = { bo, offset }, - .ImmediateData = 1); -} + cmd_buffer->state.gfx.primitive_topology = topology; -void genX(CmdBeginQuery)( - VkCommandBuffer commandBuffer, - VkQueryPool queryPool, - uint32_t query, - VkQueryControlFlags flags) -{ - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_query_pool, pool, queryPool); - - /* Workaround: When meta uses the pipeline with the VS disabled, it seems - * that the pipelining of the depth write breaks. What we see is that - * samples from the render pass clear leaks into the first query - * immediately after the clear. Doing a pipecontrol with a post-sync - * operation and DepthStallEnable seems to work around the issue. - */ - if (cmd_buffer->state.need_query_wa) { - cmd_buffer->state.need_query_wa = false; - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .DepthCacheFlushEnable = true, - .DepthStallEnable = true); + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_VF_TOPOLOGY), vft) { + vft.PrimitiveTopologyType = topology; + } } - switch (pool->type) { - case VK_QUERY_TYPE_OCCLUSION: - emit_ps_depth_count(&cmd_buffer->batch, &pool->bo, - query * sizeof(struct anv_query_pool_slot)); - break; - - case VK_QUERY_TYPE_PIPELINE_STATISTICS: - default: - unreachable(""); - } + cmd_buffer->state.gfx.dirty = 0; } -void genX(CmdEndQuery)( - VkCommandBuffer commandBuffer, - VkQueryPool queryPool, - uint32_t query) +static uint32_t vk_to_gen_index_type(VkIndexType type) { - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_query_pool, pool, queryPool); - - switch (pool->type) { - case VK_QUERY_TYPE_OCCLUSION: - emit_ps_depth_count(&cmd_buffer->batch, &pool->bo, - query * sizeof(struct anv_query_pool_slot) + 8); - - emit_query_availability(&cmd_buffer->batch, &pool->bo, - query * sizeof(struct anv_query_pool_slot) + 16); - break; - - case VK_QUERY_TYPE_PIPELINE_STATISTICS: + switch (type) { + case VK_INDEX_TYPE_UINT8_EXT: + return INDEX_BYTE; + case VK_INDEX_TYPE_UINT16: + return INDEX_WORD; + case VK_INDEX_TYPE_UINT32: + return INDEX_DWORD; default: - unreachable(""); + unreachable("invalid index type"); } } -#define TIMESTAMP 0x2358 - -void genX(CmdWriteTimestamp)( - VkCommandBuffer commandBuffer, - VkPipelineStageFlagBits pipelineStage, - VkQueryPool queryPool, - uint32_t query) +static uint32_t restart_index_for_type(VkIndexType type) { - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_query_pool, pool, queryPool); - uint32_t offset = query * sizeof(struct anv_query_pool_slot); - - assert(pool->type == VK_QUERY_TYPE_TIMESTAMP); - - switch (pipelineStage) { - case VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT: - anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_REGISTER_MEM), - .RegisterAddress = TIMESTAMP, - .MemoryAddress = { &pool->bo, offset }); - anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_REGISTER_MEM), - .RegisterAddress = TIMESTAMP + 4, - .MemoryAddress = { &pool->bo, offset + 4 }); - break; - + switch (type) { + case VK_INDEX_TYPE_UINT8_EXT: + return UINT8_MAX; + case VK_INDEX_TYPE_UINT16: + return UINT16_MAX; + case VK_INDEX_TYPE_UINT32: + return UINT32_MAX; default: - /* Everything else is bottom-of-pipe */ - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .DestinationAddressType = DAT_PPGTT, - .PostSyncOperation = WriteTimestamp, - .Address = { &pool->bo, offset }); - break; + unreachable("invalid index type"); } - - emit_query_availability(&cmd_buffer->batch, &pool->bo, query + 16); } -#define alu_opcode(v) __gen_uint((v), 20, 31) -#define alu_operand1(v) __gen_uint((v), 10, 19) -#define alu_operand2(v) __gen_uint((v), 0, 9) -#define alu(opcode, operand1, operand2) \ - alu_opcode(opcode) | alu_operand1(operand1) | alu_operand2(operand2) - -#define OPCODE_NOOP 0x000 -#define OPCODE_LOAD 0x080 -#define OPCODE_LOADINV 0x480 -#define OPCODE_LOAD0 0x081 -#define OPCODE_LOAD1 0x481 -#define OPCODE_ADD 0x100 -#define OPCODE_SUB 0x101 -#define OPCODE_AND 0x102 -#define OPCODE_OR 0x103 -#define OPCODE_XOR 0x104 -#define OPCODE_STORE 0x180 -#define OPCODE_STOREINV 0x580 - -#define OPERAND_R0 0x00 -#define OPERAND_R1 0x01 -#define OPERAND_R2 0x02 -#define OPERAND_R3 0x03 -#define OPERAND_R4 0x04 -#define OPERAND_SRCA 0x20 -#define OPERAND_SRCB 0x21 -#define OPERAND_ACCU 0x31 -#define OPERAND_ZF 0x32 -#define OPERAND_CF 0x33 - -#define CS_GPR(n) (0x2600 + (n) * 8) - -static void -emit_load_alu_reg_u64(struct anv_batch *batch, uint32_t reg, - struct anv_bo *bo, uint32_t offset) -{ - anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_MEM), - .RegisterAddress = reg, - .MemoryAddress = { bo, offset }); - anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_MEM), - .RegisterAddress = reg + 4, - .MemoryAddress = { bo, offset + 4 }); -} - -static void -store_query_result(struct anv_batch *batch, uint32_t reg, - struct anv_bo *bo, uint32_t offset, VkQueryResultFlags flags) -{ - anv_batch_emit(batch, GENX(MI_STORE_REGISTER_MEM), - .RegisterAddress = reg, - .MemoryAddress = { bo, offset }); - - if (flags & VK_QUERY_RESULT_64_BIT) - anv_batch_emit(batch, GENX(MI_STORE_REGISTER_MEM), - .RegisterAddress = reg + 4, - .MemoryAddress = { bo, offset + 4 }); -} - -void genX(CmdCopyQueryPoolResults)( - VkCommandBuffer commandBuffer, - VkQueryPool queryPool, - uint32_t firstQuery, - uint32_t queryCount, - VkBuffer destBuffer, - VkDeviceSize destOffset, - VkDeviceSize destStride, - VkQueryResultFlags flags) -{ - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_query_pool, pool, queryPool); - ANV_FROM_HANDLE(anv_buffer, buffer, destBuffer); - uint32_t slot_offset, dst_offset; - - if (flags & VK_QUERY_RESULT_WAIT_BIT) - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .CommandStreamerStallEnable = true, - .StallAtPixelScoreboard = true); - - dst_offset = buffer->offset + destOffset; - for (uint32_t i = 0; i < queryCount; i++) { - - slot_offset = (firstQuery + i) * sizeof(struct anv_query_pool_slot); - switch (pool->type) { - case VK_QUERY_TYPE_OCCLUSION: - emit_load_alu_reg_u64(&cmd_buffer->batch, - CS_GPR(0), &pool->bo, slot_offset); - emit_load_alu_reg_u64(&cmd_buffer->batch, - CS_GPR(1), &pool->bo, slot_offset + 8); - - /* FIXME: We need to clamp the result for 32 bit. */ - - uint32_t *dw = anv_batch_emitn(&cmd_buffer->batch, 5, GENX(MI_MATH)); - dw[1] = alu(OPCODE_LOAD, OPERAND_SRCA, OPERAND_R1); - dw[2] = alu(OPCODE_LOAD, OPERAND_SRCB, OPERAND_R0); - dw[3] = alu(OPCODE_SUB, 0, 0); - dw[4] = alu(OPCODE_STORE, OPERAND_R2, OPERAND_ACCU); - break; - - case VK_QUERY_TYPE_TIMESTAMP: - emit_load_alu_reg_u64(&cmd_buffer->batch, - CS_GPR(2), &pool->bo, slot_offset); - break; - - default: - unreachable("unhandled query type"); - } - - store_query_result(&cmd_buffer->batch, - CS_GPR(2), buffer->bo, dst_offset, flags); - - if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) { - emit_load_alu_reg_u64(&cmd_buffer->batch, CS_GPR(0), - &pool->bo, slot_offset + 16); - if (flags & VK_QUERY_RESULT_64_BIT) - store_query_result(&cmd_buffer->batch, - CS_GPR(0), buffer->bo, dst_offset + 8, flags); - else - store_query_result(&cmd_buffer->batch, - CS_GPR(0), buffer->bo, dst_offset + 4, flags); - } - - dst_offset += destStride; - } -} - -void genX(CmdSetEvent)( +void genX(CmdBindIndexBuffer)( VkCommandBuffer commandBuffer, - VkEvent _event, - VkPipelineStageFlags stageMask) + VkBuffer _buffer, + VkDeviceSize offset, + VkIndexType indexType) { ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_event, event, _event); - - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .DestinationAddressType = DAT_PPGTT, - .PostSyncOperation = WriteImmediateData, - .Address = { - &cmd_buffer->device->dynamic_state_block_pool.bo, - event->state.offset - }, - .ImmediateData = VK_EVENT_SET); -} + ANV_FROM_HANDLE(anv_buffer, buffer, _buffer); -void genX(CmdResetEvent)( - VkCommandBuffer commandBuffer, - VkEvent _event, - VkPipelineStageFlags stageMask) -{ - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - ANV_FROM_HANDLE(anv_event, event, _event); - - anv_batch_emit(&cmd_buffer->batch, GENX(PIPE_CONTROL), - .DestinationAddressType = DAT_PPGTT, - .PostSyncOperation = WriteImmediateData, - .Address = { - &cmd_buffer->device->dynamic_state_block_pool.bo, - event->state.offset - }, - .ImmediateData = VK_EVENT_RESET); -} + cmd_buffer->state.restart_index = restart_index_for_type(indexType); -void genX(CmdWaitEvents)( - VkCommandBuffer commandBuffer, - uint32_t eventCount, - const VkEvent* pEvents, - VkPipelineStageFlags srcStageMask, - VkPipelineStageFlags destStageMask, - uint32_t memoryBarrierCount, - const VkMemoryBarrier* pMemoryBarriers, - uint32_t bufferMemoryBarrierCount, - const VkBufferMemoryBarrier* pBufferMemoryBarriers, - uint32_t imageMemoryBarrierCount, - const VkImageMemoryBarrier* pImageMemoryBarriers) -{ - ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer); - for (uint32_t i = 0; i < eventCount; i++) { - ANV_FROM_HANDLE(anv_event, event, pEvents[i]); - - anv_batch_emit(&cmd_buffer->batch, GENX(MI_SEMAPHORE_WAIT), - .WaitMode = PollingMode, - .CompareOperation = COMPARE_SAD_EQUAL_SDD, - .SemaphoreDataDword = VK_EVENT_SET, - .SemaphoreAddress = { - &cmd_buffer->device->dynamic_state_block_pool.bo, - event->state.offset - }); + anv_batch_emit(&cmd_buffer->batch, GENX(3DSTATE_INDEX_BUFFER), ib) { + ib.IndexFormat = vk_to_gen_index_type(indexType); + ib.MOCS = anv_mocs_for_bo(cmd_buffer->device, + buffer->address.bo); + ib.BufferStartingAddress = anv_address_add(buffer->address, offset); + ib.BufferSize = buffer->size - offset; } - genX(CmdPipelineBarrier)(commandBuffer, srcStageMask, destStageMask, - false, /* byRegion */ - memoryBarrierCount, pMemoryBarriers, - bufferMemoryBarrierCount, pBufferMemoryBarriers, - imageMemoryBarrierCount, pImageMemoryBarriers); + cmd_buffer->state.gfx.dirty |= ANV_CMD_DIRTY_INDEX_BUFFER; }