From cd009cfec6719b3edf5616736b0efd85225ecff5 Mon Sep 17 00:00:00 2001 From: IndecisiveTurtle <47210458+raphaelthegreat@users.noreply.github.com> Date: Wed, 17 Jul 2024 16:49:45 +0300 Subject: [PATCH] shader_recompiler: Normal gathers --- .../backend/spirv/emit_spirv_image.cpp | 35 +++++++++++++------ .../backend/spirv/emit_spirv_instructions.h | 6 ++-- .../frontend/translate/translate.cpp | 2 ++ .../frontend/translate/vector_memory.cpp | 1 + .../ir/passes/resource_tracking_pass.cpp | 4 ++- src/shader_recompiler/ir/reg.h | 1 + .../renderer_vulkan/liverpool_to_vk.cpp | 4 +++ 7 files changed, 38 insertions(+), 15 deletions(-) diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp index 2d35b97c..b7249100 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp @@ -9,6 +9,9 @@ namespace Shader::Backend::SPIRV { struct ImageOperands { void Add(spv::ImageOperandsMask new_mask, Id value) { + if (!Sirit::ValidId(value)) { + return; + } mask = static_cast(static_cast(mask) | static_cast(new_mask)); operands.push_back(value); @@ -25,9 +28,7 @@ Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id c const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]); const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler); ImageOperands operands; - if (Sirit::ValidId(offset)) { - operands.Add(spv::ImageOperandsMask::ConstOffset, offset); - } + operands.Add(spv::ImageOperandsMask::Offset, offset); return ctx.OpImageSampleImplicitLod(ctx.F32[4], sampled_image, coords, operands.mask, operands.operands); } @@ -61,18 +62,30 @@ Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f)); } -Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, - const IR::Value& offset, const IR::Value& offset2) { - UNREACHABLE_MSG("SPIR-V Instruction"); -} - -Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, - const IR::Value& offset, const IR::Value& offset2, Id dref) { +Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, + Id offset, Id offset2) { const auto& texture = ctx.images[handle & 0xFFFF]; const Id image = ctx.OpLoad(texture.image_type, texture.id); const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]); const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler); - return ctx.OpImageDrefGather(ctx.F32[4], sampled_image, coords, dref); + const u32 comp = inst->Flags().gather_comp.Value(); + ImageOperands operands; + operands.Add(spv::ImageOperandsMask::Offset, offset); + operands.Add(spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f)); + return ctx.OpImageGather(ctx.F32[4], sampled_image, coords, ctx.ConstU32(comp), + operands.mask, operands.operands); +} + +Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, + Id offset, Id offset2, Id dref) { + const auto& texture = ctx.images[handle & 0xFFFF]; + const Id image = ctx.OpLoad(texture.image_type, texture.id); + const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]); + const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler); + ImageOperands operands; + operands.Add(spv::ImageOperandsMask::Offset, offset); + return ctx.OpImageDrefGather(ctx.F32[4], sampled_image, coords, dref, + operands.mask, operands.operands); } Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id offset, Id lod, diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h index f43ea3b3..1432041c 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h +++ b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h @@ -358,10 +358,10 @@ Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id bias_lc, const IR::Value& offset); Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref, Id bias_lc, Id offset); -Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, - const IR::Value& offset, const IR::Value& offset2); +Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, + Id offset, Id offset2); Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, - const IR::Value& offset, const IR::Value& offset2, Id dref); + Id offset, Id offset2, Id dref); Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id offset, Id lod, Id ms); Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, u32 handle, Id lod, bool skip_mips); diff --git a/src/shader_recompiler/frontend/translate/translate.cpp b/src/shader_recompiler/frontend/translate/translate.cpp index bc2e0bf2..6e5f7f8b 100644 --- a/src/shader_recompiler/frontend/translate/translate.cpp +++ b/src/shader_recompiler/frontend/translate/translate.cpp @@ -456,6 +456,8 @@ void Translate(IR::Block* block, u32 block_base, std::span inst_l translator.IMAGE_GET_LOD(inst); break; case Opcode::IMAGE_GATHER4_C: + case Opcode::IMAGE_GATHER4_LZ: + case Opcode::IMAGE_GATHER4_LZ_O: translator.IMAGE_GATHER(inst); break; case Opcode::IMAGE_STORE: diff --git a/src/shader_recompiler/frontend/translate/vector_memory.cpp b/src/shader_recompiler/frontend/translate/vector_memory.cpp index 1ddee523..f4383c61 100644 --- a/src/shader_recompiler/frontend/translate/vector_memory.cpp +++ b/src/shader_recompiler/frontend/translate/vector_memory.cpp @@ -158,6 +158,7 @@ void Translator::IMAGE_GATHER(const GcnInst& inst) { info.has_lod_clamp.Assign(flags.test(MimgModifier::LodClamp)); info.force_level0.Assign(flags.test(MimgModifier::Level0)); info.explicit_lod.Assign(explicit_lod); + info.gather_comp.Assign(std::bit_width(mimg.dmask) - 1); // Issue IR instruction, leaving unknown fields blank to patch later. const IR::Value texel = [&]() -> IR::Value { diff --git a/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp b/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp index f58b4d96..6526ece6 100644 --- a/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp +++ b/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp @@ -543,7 +543,9 @@ void PatchImageInstruction(IR::Block& block, IR::Inst& inst, Info& info, Descrip if (inst_info.has_offset) { // The offsets are six-bit signed integers: X=[5:0], Y=[13:8], and Z=[21:16]. - const u32 arg_pos = inst_info.is_depth ? 4 : 3; + const bool is_gather = inst.GetOpcode() == IR::Opcode::ImageGather || + inst.GetOpcode() == IR::Opcode::ImageGatherDref; + const u32 arg_pos = is_gather ? 2 : (inst_info.is_depth ? 4 : 3); const IR::Value arg = inst.Arg(arg_pos); ASSERT_MSG(arg.Type() == IR::Type::U32, "Unexpected offset type"); const auto sign_ext = [&](u32 value) { return ir.Imm32(s32(value << 24) >> 24); }; diff --git a/src/shader_recompiler/ir/reg.h b/src/shader_recompiler/ir/reg.h index ae38ecf3..d9e9b030 100644 --- a/src/shader_recompiler/ir/reg.h +++ b/src/shader_recompiler/ir/reg.h @@ -39,6 +39,7 @@ union TextureInstInfo { BitField<3, 1, u32> force_level0; BitField<4, 1, u32> explicit_lod; BitField<5, 1, u32> has_offset; + BitField<6, 2, u32> gather_comp; }; union BufferInstInfo { diff --git a/src/video_core/renderer_vulkan/liverpool_to_vk.cpp b/src/video_core/renderer_vulkan/liverpool_to_vk.cpp index 0cbe0de8..d9ca8b1f 100644 --- a/src/video_core/renderer_vulkan/liverpool_to_vk.cpp +++ b/src/video_core/renderer_vulkan/liverpool_to_vk.cpp @@ -420,6 +420,10 @@ vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat nu num_format == AmdGpu::NumberFormat::Uint) { return vk::Format::eR32G32B32A32Uint; } + if (data_format == AmdGpu::DataFormat::Format32_32_32_32 && + num_format == AmdGpu::NumberFormat::Sint) { + return vk::Format::eR32G32B32A32Sint; + } if (data_format == AmdGpu::DataFormat::Format8 && num_format == AmdGpu::NumberFormat::Sint) { return vk::Format::eR8Sint; }