From 8eacb88a865fd38d87ce1f2510ecd5e6b108ae27 Mon Sep 17 00:00:00 2001 From: Vladislav Mikhalin Date: Sat, 7 Dec 2024 02:20:09 +0300 Subject: [PATCH] recompiler: fixed fragment shader built-in attribute access (#1676) * recompiler: fixed fragment shader built-in attribute access * handle en/addr separately * handle other registers as well --- .../frontend/translate/translate.cpp | 75 +++++++++++++++++-- src/shader_recompiler/runtime_info.h | 4 + src/video_core/amdgpu/liverpool.h | 29 ++++++- .../renderer_vulkan/vk_pipeline_cache.cpp | 2 + 4 files changed, 101 insertions(+), 9 deletions(-) diff --git a/src/shader_recompiler/frontend/translate/translate.cpp b/src/shader_recompiler/frontend/translate/translate.cpp index 68625a12b..b3a47fde8 100644 --- a/src/shader_recompiler/frontend/translate/translate.cpp +++ b/src/shader_recompiler/frontend/translate/translate.cpp @@ -53,15 +53,74 @@ void Translator::EmitPrologue() { } break; case Stage::Fragment: - // https://github.com/chaotic-cx/mesa-mirror/blob/72326e15/src/amd/vulkan/radv_shader_args.c#L258 - // The first two VGPRs are used for i/j barycentric coordinates. In the vast majority of - // cases it will be only those two, but if shader is using both e.g linear and perspective - // inputs it can be more For now assume that this isn't the case. - dst_vreg = IR::VectorReg::V2; - for (u32 i = 0; i < 4; i++) { - ir.SetVectorReg(dst_vreg++, ir.GetAttribute(IR::Attribute::FragCoord, i)); + dst_vreg = IR::VectorReg::V0; + if (runtime_info.fs_info.addr_flags.persp_sample_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.persp_center_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.persp_centroid_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.persp_pull_model_ena) { + ++dst_vreg; // I/W + ++dst_vreg; // J/W + ++dst_vreg; // 1/W + } + if (runtime_info.fs_info.addr_flags.linear_sample_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.linear_center_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.linear_centroid_ena) { + ++dst_vreg; // I + ++dst_vreg; // J + } + if (runtime_info.fs_info.addr_flags.line_stipple_tex_ena) { + ++dst_vreg; + } + if (runtime_info.fs_info.addr_flags.pos_x_float_ena) { + if (runtime_info.fs_info.en_flags.pos_x_float_ena) { + ir.SetVectorReg(dst_vreg++, ir.GetAttribute(IR::Attribute::FragCoord, 0)); + } else { + ir.SetVectorReg(dst_vreg++, ir.Imm32(0.0f)); + } + } + if (runtime_info.fs_info.addr_flags.pos_y_float_ena) { + if (runtime_info.fs_info.en_flags.pos_y_float_ena) { + ir.SetVectorReg(dst_vreg++, ir.GetAttribute(IR::Attribute::FragCoord, 1)); + } else { + ir.SetVectorReg(dst_vreg++, ir.Imm32(0.0f)); + } + } + if (runtime_info.fs_info.addr_flags.pos_z_float_ena) { + if (runtime_info.fs_info.en_flags.pos_z_float_ena) { + ir.SetVectorReg(dst_vreg++, ir.GetAttribute(IR::Attribute::FragCoord, 2)); + } else { + ir.SetVectorReg(dst_vreg++, ir.Imm32(0.0f)); + } + } + if (runtime_info.fs_info.addr_flags.pos_w_float_ena) { + if (runtime_info.fs_info.en_flags.pos_w_float_ena) { + ir.SetVectorReg(dst_vreg++, ir.GetAttribute(IR::Attribute::FragCoord, 3)); + } else { + ir.SetVectorReg(dst_vreg++, ir.Imm32(0.0f)); + } + } + if (runtime_info.fs_info.addr_flags.front_face_ena) { + if (runtime_info.fs_info.en_flags.front_face_ena) { + ir.SetVectorReg(dst_vreg++, ir.GetAttributeU32(IR::Attribute::IsFrontFace)); + } else { + ir.SetVectorReg(dst_vreg++, ir.Imm32(0)); + } } - ir.SetVectorReg(dst_vreg++, ir.GetAttributeU32(IR::Attribute::IsFrontFace)); break; case Stage::Compute: ir.SetVectorReg(dst_vreg++, ir.GetAttributeU32(IR::Attribute::LocalInvocationId, 0)); diff --git a/src/shader_recompiler/runtime_info.h b/src/shader_recompiler/runtime_info.h index 4662def93..4c779a368 100644 --- a/src/shader_recompiler/runtime_info.h +++ b/src/shader_recompiler/runtime_info.h @@ -7,6 +7,7 @@ #include #include #include "common/types.h" +#include "video_core/amdgpu/liverpool.h" #include "video_core/amdgpu/types.h" namespace Shader { @@ -105,6 +106,8 @@ struct FragmentRuntimeInfo { auto operator<=>(const PsInput&) const noexcept = default; }; + AmdGpu::Liverpool::PsInput en_flags; + AmdGpu::Liverpool::PsInput addr_flags; u32 num_inputs; std::array inputs; struct PsColorBuffer { @@ -117,6 +120,7 @@ struct FragmentRuntimeInfo { bool operator==(const FragmentRuntimeInfo& other) const noexcept { return std::ranges::equal(color_buffers, other.color_buffers) && + en_flags.raw == other.en_flags.raw && addr_flags.raw == other.addr_flags.raw && num_inputs == other.num_inputs && std::ranges::equal(inputs.begin(), inputs.begin() + num_inputs, other.inputs.begin(), other.inputs.begin() + num_inputs); diff --git a/src/video_core/amdgpu/liverpool.h b/src/video_core/amdgpu/liverpool.h index 2b2f2c00a..ca3b01612 100644 --- a/src/video_core/amdgpu/liverpool.h +++ b/src/video_core/amdgpu/liverpool.h @@ -1071,6 +1071,28 @@ struct Liverpool { BitField<27, 1, u32> enable_postz_overrasterization; }; + union PsInput { + u32 raw; + struct { + u32 persp_sample_ena : 1; + u32 persp_center_ena : 1; + u32 persp_centroid_ena : 1; + u32 persp_pull_model_ena : 1; + u32 linear_sample_ena : 1; + u32 linear_center_ena : 1; + u32 linear_centroid_ena : 1; + u32 line_stipple_tex_ena : 1; + u32 pos_x_float_ena : 1; + u32 pos_y_float_ena : 1; + u32 pos_z_float_ena : 1; + u32 pos_w_float_ena : 1; + u32 front_face_ena : 1; + u32 ancillary_ena : 1; + u32 sample_coverage_ena : 1; + u32 pos_fixed_pt_ena : 1; + }; + }; + union Regs { struct { INSERT_PADDING_WORDS(0x2C08); @@ -1126,7 +1148,10 @@ struct Liverpool { INSERT_PADDING_WORDS(0xA191 - 0xA187); std::array ps_inputs; VsOutputConfig vs_output_config; - INSERT_PADDING_WORDS(4); + INSERT_PADDING_WORDS(1); + PsInput ps_input_ena; + PsInput ps_input_addr; + INSERT_PADDING_WORDS(1); BitField<0, 6, u32> num_interp; INSERT_PADDING_WORDS(0xA1C3 - 0xA1B6 - 1); ShaderPosFormat shader_pos_format; @@ -1388,6 +1413,8 @@ static_assert(GFX6_3D_REG_INDEX(viewports) == 0xA10F); static_assert(GFX6_3D_REG_INDEX(clip_user_data) == 0xA16F); static_assert(GFX6_3D_REG_INDEX(ps_inputs) == 0xA191); static_assert(GFX6_3D_REG_INDEX(vs_output_config) == 0xA1B1); +static_assert(GFX6_3D_REG_INDEX(ps_input_ena) == 0xA1B3); +static_assert(GFX6_3D_REG_INDEX(ps_input_addr) == 0xA1B4); static_assert(GFX6_3D_REG_INDEX(num_interp) == 0xA1B6); static_assert(GFX6_3D_REG_INDEX(shader_pos_format) == 0xA1C3); static_assert(GFX6_3D_REG_INDEX(z_export_format) == 0xA1C4); diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index 82a029b95..53bdc79a6 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -123,6 +123,8 @@ Shader::RuntimeInfo PipelineCache::BuildRuntimeInfo(Shader::Stage stage) { } case Shader::Stage::Fragment: { BuildCommon(regs.ps_program); + info.fs_info.en_flags = regs.ps_input_ena; + info.fs_info.addr_flags = regs.ps_input_addr; const auto& ps_inputs = regs.ps_inputs; info.fs_info.num_inputs = regs.num_interp; for (u32 i = 0; i < regs.num_interp; i++) {