summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2021-02-21 21:50:14 +0100
committerameerj <52414509+ameerj@users.noreply.github.com>2021-07-23 03:51:22 +0200
commit704c6f353f68745168902c6c66c04bb730bd30e6 (patch)
tree71ed9654de41b5828ae2613167537d39499d2f3b
parentshader: Add denorm flush support (diff)
downloadyuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar.gz
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar.bz2
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar.lz
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar.xz
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.tar.zst
yuzu-704c6f353f68745168902c6c66c04bb730bd30e6.zip
Diffstat (limited to '')
-rw-r--r--src/shader_recompiler/CMakeLists.txt3
-rw-r--r--src/shader_recompiler/backend/spirv/emit_spirv.h40
-rw-r--r--src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp58
-rw-r--r--src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp75
-rw-r--r--src/shader_recompiler/backend/spirv/emit_spirv_select.cpp4
-rw-r--r--src/shader_recompiler/frontend/ir/ir_emitter.cpp16
-rw-r--r--src/shader_recompiler/frontend/ir/pred.h4
-rw-r--r--src/shader_recompiler/frontend/maxwell/program.cpp2
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp20
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp2
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp4
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp2
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp17
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/impl.h7
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp4
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/move_predicate_to_register.cpp66
-rw-r--r--src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp12
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp4
18 files changed, 213 insertions, 127 deletions
diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt
index fbd4ec6dc..802527255 100644
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@@ -74,9 +74,10 @@ add_library(shader_recompiler STATIC
frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
frontend/maxwell/translate/impl/load_store_attribute.cpp
frontend/maxwell/translate/impl/load_store_memory.cpp
- frontend/maxwell/translate/impl/not_implemented.cpp
+ frontend/maxwell/translate/impl/move_predicate_to_register.cpp
frontend/maxwell/translate/impl/move_register.cpp
frontend/maxwell/translate/impl/move_special_register.cpp
+ frontend/maxwell/translate/impl/not_implemented.cpp
frontend/maxwell/translate/translate.cpp
frontend/maxwell/translate/translate.h
ir_opt/collect_shader_info_pass.cpp
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.h b/src/shader_recompiler/backend/spirv/emit_spirv.h
index de624a151..922e294a7 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.h
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.h
@@ -110,7 +110,7 @@ void EmitCompositeExtractF64x3(EmitContext& ctx);
void EmitCompositeExtractF64x4(EmitContext& ctx);
void EmitSelect8(EmitContext& ctx);
void EmitSelect16(EmitContext& ctx);
-void EmitSelect32(EmitContext& ctx);
+Id EmitSelect32(EmitContext& ctx, Id cond, Id true_value, Id false_value);
void EmitSelect64(EmitContext& ctx);
void EmitBitCastU16F16(EmitContext& ctx);
Id EmitBitCastU32F32(EmitContext& ctx, Id value);
@@ -130,9 +130,9 @@ void EmitGetZeroFromOp(EmitContext& ctx);
void EmitGetSignFromOp(EmitContext& ctx);
void EmitGetCarryFromOp(EmitContext& ctx);
void EmitGetOverflowFromOp(EmitContext& ctx);
-void EmitFPAbs16(EmitContext& ctx);
-void EmitFPAbs32(EmitContext& ctx);
-void EmitFPAbs64(EmitContext& ctx);
+Id EmitFPAbs16(EmitContext& ctx, Id value);
+Id EmitFPAbs32(EmitContext& ctx, Id value);
+Id EmitFPAbs64(EmitContext& ctx, Id value);
Id EmitFPAdd16(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
Id EmitFPAdd32(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
Id EmitFPAdd64(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
@@ -146,9 +146,9 @@ void EmitFPMin64(EmitContext& ctx);
Id EmitFPMul16(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
Id EmitFPMul32(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
Id EmitFPMul64(EmitContext& ctx, IR::Inst* inst, Id a, Id b);
-void EmitFPNeg16(EmitContext& ctx);
-void EmitFPNeg32(EmitContext& ctx);
-void EmitFPNeg64(EmitContext& ctx);
+Id EmitFPNeg16(EmitContext& ctx, Id value);
+Id EmitFPNeg32(EmitContext& ctx, Id value);
+Id EmitFPNeg64(EmitContext& ctx, Id value);
void EmitFPRecip32(EmitContext& ctx);
void EmitFPRecip64(EmitContext& ctx);
void EmitFPRecipSqrt32(EmitContext& ctx);
@@ -161,9 +161,9 @@ void EmitFPExp2NotReduced(EmitContext& ctx);
void EmitFPCos(EmitContext& ctx);
void EmitFPCosNotReduced(EmitContext& ctx);
void EmitFPLog2(EmitContext& ctx);
-void EmitFPSaturate16(EmitContext& ctx);
-void EmitFPSaturate32(EmitContext& ctx);
-void EmitFPSaturate64(EmitContext& ctx);
+Id EmitFPSaturate16(EmitContext& ctx, Id value);
+Id EmitFPSaturate32(EmitContext& ctx, Id value);
+Id EmitFPSaturate64(EmitContext& ctx, Id value);
Id EmitFPRoundEven16(EmitContext& ctx, Id value);
Id EmitFPRoundEven32(EmitContext& ctx, Id value);
Id EmitFPRoundEven64(EmitContext& ctx, Id value);
@@ -186,21 +186,21 @@ void EmitIAbs32(EmitContext& ctx);
Id EmitShiftLeftLogical32(EmitContext& ctx, Id base, Id shift);
void EmitShiftRightLogical32(EmitContext& ctx);
void EmitShiftRightArithmetic32(EmitContext& ctx);
-void EmitBitwiseAnd32(EmitContext& ctx);
-void EmitBitwiseOr32(EmitContext& ctx);
-void EmitBitwiseXor32(EmitContext& ctx);
+Id EmitBitwiseAnd32(EmitContext& ctx, Id a, Id b);
+Id EmitBitwiseOr32(EmitContext& ctx, Id a, Id b);
+Id EmitBitwiseXor32(EmitContext& ctx, Id a, Id b);
void EmitBitFieldInsert(EmitContext& ctx);
void EmitBitFieldSExtract(EmitContext& ctx);
Id EmitBitFieldUExtract(EmitContext& ctx, Id base, Id offset, Id count);
Id EmitSLessThan(EmitContext& ctx, Id lhs, Id rhs);
-void EmitULessThan(EmitContext& ctx);
-void EmitIEqual(EmitContext& ctx);
-void EmitSLessThanEqual(EmitContext& ctx);
-void EmitULessThanEqual(EmitContext& ctx);
+Id EmitULessThan(EmitContext& ctx, Id lhs, Id rhs);
+Id EmitIEqual(EmitContext& ctx, Id lhs, Id rhs);
+Id EmitSLessThanEqual(EmitContext& ctx, Id lhs, Id rhs);
+Id EmitULessThanEqual(EmitContext& ctx, Id lhs, Id rhs);
Id EmitSGreaterThan(EmitContext& ctx, Id lhs, Id rhs);
-void EmitUGreaterThan(EmitContext& ctx);
-void EmitINotEqual(EmitContext& ctx);
-void EmitSGreaterThanEqual(EmitContext& ctx);
+Id EmitUGreaterThan(EmitContext& ctx, Id lhs, Id rhs);
+Id EmitINotEqual(EmitContext& ctx, Id lhs, Id rhs);
+Id EmitSGreaterThanEqual(EmitContext& ctx, Id lhs, Id rhs);
Id EmitUGreaterThanEqual(EmitContext& ctx, Id lhs, Id rhs);
void EmitLogicalOr(EmitContext& ctx);
void EmitLogicalAnd(EmitContext& ctx);
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp
index c9687de37..47f87054b 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp
@@ -12,37 +12,21 @@ Id Decorate(EmitContext& ctx, IR::Inst* inst, Id op) {
if (flags.no_contraction) {
ctx.Decorate(op, spv::Decoration::NoContraction);
}
- switch (flags.rounding) {
- case IR::FpRounding::DontCare:
- break;
- case IR::FpRounding::RN:
- ctx.Decorate(op, spv::Decoration::FPRoundingMode, spv::FPRoundingMode::RTE);
- break;
- case IR::FpRounding::RM:
- ctx.Decorate(op, spv::Decoration::FPRoundingMode, spv::FPRoundingMode::RTN);
- break;
- case IR::FpRounding::RP:
- ctx.Decorate(op, spv::Decoration::FPRoundingMode, spv::FPRoundingMode::RTP);
- break;
- case IR::FpRounding::RZ:
- ctx.Decorate(op, spv::Decoration::FPRoundingMode, spv::FPRoundingMode::RTZ);
- break;
- }
return op;
}
} // Anonymous namespace
-void EmitFPAbs16(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPAbs16(EmitContext& ctx, Id value) {
+ return ctx.OpFAbs(ctx.F16[1], value);
}
-void EmitFPAbs32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPAbs32(EmitContext& ctx, Id value) {
+ return ctx.OpFAbs(ctx.F32[1], value);
}
-void EmitFPAbs64(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPAbs64(EmitContext& ctx, Id value) {
+ return ctx.OpFAbs(ctx.F64[1], value);
}
Id EmitFPAdd16(EmitContext& ctx, IR::Inst* inst, Id a, Id b) {
@@ -97,16 +81,16 @@ Id EmitFPMul64(EmitContext& ctx, IR::Inst* inst, Id a, Id b) {
return Decorate(ctx, inst, ctx.OpFMul(ctx.F64[1], a, b));
}
-void EmitFPNeg16(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPNeg16(EmitContext& ctx, Id value) {
+ return ctx.OpFNegate(ctx.F16[1], value);
}
-void EmitFPNeg32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPNeg32(EmitContext& ctx, Id value) {
+ return ctx.OpFNegate(ctx.F32[1], value);
}
-void EmitFPNeg64(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPNeg64(EmitContext& ctx, Id value) {
+ return ctx.OpFNegate(ctx.F64[1], value);
}
void EmitFPRecip32(EmitContext&) {
@@ -157,16 +141,22 @@ void EmitFPLog2(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
-void EmitFPSaturate16(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPSaturate16(EmitContext& ctx, Id value) {
+ const Id zero{ctx.Constant(ctx.F16[1], u16{0})};
+ const Id one{ctx.Constant(ctx.F16[1], u16{0x3c00})};
+ return ctx.OpFClamp(ctx.F32[1], value, zero, one);
}
-void EmitFPSaturate32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPSaturate32(EmitContext& ctx, Id value) {
+ const Id zero{ctx.Constant(ctx.F32[1], f32{0.0})};
+ const Id one{ctx.Constant(ctx.F32[1], f32{1.0})};
+ return ctx.OpFClamp(ctx.F32[1], value, zero, one);
}
-void EmitFPSaturate64(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitFPSaturate64(EmitContext& ctx, Id value) {
+ const Id zero{ctx.Constant(ctx.F64[1], f64{0.0})};
+ const Id one{ctx.Constant(ctx.F64[1], f64{1.0})};
+ return ctx.OpFClamp(ctx.F64[1], value, zero, one);
}
Id EmitFPRoundEven16(EmitContext& ctx, Id value) {
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
index 22117a4ee..4c0b5990d 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_integer.cpp
@@ -7,10 +7,39 @@
namespace Shader::Backend::SPIRV {
Id EmitIAdd32(EmitContext& ctx, IR::Inst* inst, Id a, Id b) {
- if (inst->HasAssociatedPseudoOperation()) {
- throw NotImplementedException("Pseudo-operations on IAdd32");
+ Id result{};
+ if (IR::Inst* const carry{inst->GetAssociatedPseudoOperation(IR::Opcode::GetCarryFromOp)}) {
+ const Id carry_type{ctx.TypeStruct(ctx.U32[1], ctx.U32[1])};
+ const Id carry_result{ctx.OpIAddCarry(carry_type, a, b)};
+ result = ctx.OpCompositeExtract(ctx.U32[1], carry_result, 0U);
+
+ const Id carry_value{ctx.OpCompositeExtract(ctx.U32[1], carry_result, 1U)};
+ carry->SetDefinition(ctx.OpINotEqual(ctx.U1, carry_value, ctx.u32_zero_value));
+ carry->Invalidate();
+ } else {
+ result = ctx.OpIAdd(ctx.U32[1], a, b);
}
- return ctx.OpIAdd(ctx.U32[1], a, b);
+ if (IR::Inst* const zero{inst->GetAssociatedPseudoOperation(IR::Opcode::GetZeroFromOp)}) {
+ zero->SetDefinition(ctx.OpIEqual(ctx.U1, result, ctx.u32_zero_value));
+ zero->Invalidate();
+ }
+ if (IR::Inst* const sign{inst->GetAssociatedPseudoOperation(IR::Opcode::GetSignFromOp)}) {
+ sign->SetDefinition(ctx.OpSLessThan(ctx.U1, result, ctx.u32_zero_value));
+ sign->Invalidate();
+ }
+ if (IR::Inst * overflow{inst->GetAssociatedPseudoOperation(IR::Opcode::GetOverflowFromOp)}) {
+ // https://stackoverflow.com/questions/55468823/how-to-detect-integer-overflow-in-c
+ constexpr u32 s32_max{static_cast<u32>(std::numeric_limits<s32>::max())};
+ const Id is_positive{ctx.OpSGreaterThanEqual(ctx.U1, a, ctx.u32_zero_value)};
+ const Id sub_a{ctx.OpISub(ctx.U32[1], ctx.Constant(ctx.U32[1], s32_max), a)};
+
+ const Id positive_test{ctx.OpSGreaterThan(ctx.U1, b, sub_a)};
+ const Id negative_test{ctx.OpSLessThan(ctx.U1, b, sub_a)};
+ const Id carry_flag{ctx.OpSelect(ctx.U1, is_positive, positive_test, negative_test)};
+ overflow->SetDefinition(carry_flag);
+ overflow->Invalidate();
+ }
+ return result;
}
void EmitIAdd64(EmitContext&) {
@@ -49,16 +78,16 @@ void EmitShiftRightArithmetic32(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
-void EmitBitwiseAnd32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitBitwiseAnd32(EmitContext& ctx, Id a, Id b) {
+ return ctx.OpBitwiseAnd(ctx.U32[1], a, b);
}
-void EmitBitwiseOr32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitBitwiseOr32(EmitContext& ctx, Id a, Id b) {
+ return ctx.OpBitwiseOr(ctx.U32[1], a, b);
}
-void EmitBitwiseXor32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitBitwiseXor32(EmitContext& ctx, Id a, Id b) {
+ return ctx.OpBitwiseXor(ctx.U32[1], a, b);
}
void EmitBitFieldInsert(EmitContext&) {
@@ -77,36 +106,36 @@ Id EmitSLessThan(EmitContext& ctx, Id lhs, Id rhs) {
return ctx.OpSLessThan(ctx.U1, lhs, rhs);
}
-void EmitULessThan(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitULessThan(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpULessThan(ctx.U1, lhs, rhs);
}
-void EmitIEqual(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitIEqual(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpIEqual(ctx.U1, lhs, rhs);
}
-void EmitSLessThanEqual(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitSLessThanEqual(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpSLessThanEqual(ctx.U1, lhs, rhs);
}
-void EmitULessThanEqual(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitULessThanEqual(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpULessThanEqual(ctx.U1, lhs, rhs);
}
Id EmitSGreaterThan(EmitContext& ctx, Id lhs, Id rhs) {
return ctx.OpSGreaterThan(ctx.U1, lhs, rhs);
}
-void EmitUGreaterThan(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitUGreaterThan(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpUGreaterThan(ctx.U1, lhs, rhs);
}
-void EmitINotEqual(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitINotEqual(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpINotEqual(ctx.U1, lhs, rhs);
}
-void EmitSGreaterThanEqual(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitSGreaterThanEqual(EmitContext& ctx, Id lhs, Id rhs) {
+ return ctx.OpSGreaterThanEqual(ctx.U1, lhs, rhs);
}
Id EmitUGreaterThanEqual(EmitContext& ctx, Id lhs, Id rhs) {
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_select.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_select.cpp
index 8d5062724..eb1926a4d 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_select.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_select.cpp
@@ -14,8 +14,8 @@ void EmitSelect16(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
-void EmitSelect32(EmitContext&) {
- throw NotImplementedException("SPIR-V Instruction");
+Id EmitSelect32(EmitContext& ctx, Id cond, Id true_value, Id false_value) {
+ return ctx.OpSelect(ctx.U32[1], cond, true_value, false_value);
}
void EmitSelect64(EmitContext&) {
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
index 8f120a2f6..34c2f67fb 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
@@ -468,11 +468,11 @@ F16F32F64 IREmitter::FPFma(const F16F32F64& a, const F16F32F64& b, const F16F32F
F16F32F64 IREmitter::FPAbs(const F16F32F64& value) {
switch (value.Type()) {
- case Type::U16:
+ case Type::F16:
return Inst<F16>(Opcode::FPAbs16, value);
- case Type::U32:
+ case Type::F32:
return Inst<F32>(Opcode::FPAbs32, value);
- case Type::U64:
+ case Type::F64:
return Inst<F64>(Opcode::FPAbs64, value);
default:
ThrowInvalidType(value.Type());
@@ -481,11 +481,11 @@ F16F32F64 IREmitter::FPAbs(const F16F32F64& value) {
F16F32F64 IREmitter::FPNeg(const F16F32F64& value) {
switch (value.Type()) {
- case Type::U16:
+ case Type::F16:
return Inst<F16>(Opcode::FPNeg16, value);
- case Type::U32:
+ case Type::F32:
return Inst<F32>(Opcode::FPNeg32, value);
- case Type::U64:
+ case Type::F64:
return Inst<F64>(Opcode::FPNeg64, value);
default:
ThrowInvalidType(value.Type());
@@ -495,10 +495,10 @@ F16F32F64 IREmitter::FPNeg(const F16F32F64& value) {
F16F32F64 IREmitter::FPAbsNeg(const F16F32F64& value, bool abs, bool neg) {
F16F32F64 result{value};
if (abs) {
- result = FPAbs(value);
+ result = FPAbs(result);
}
if (neg) {
- result = FPNeg(value);
+ result = FPNeg(result);
}
return result;
}
diff --git a/src/shader_recompiler/frontend/ir/pred.h b/src/shader_recompiler/frontend/ir/pred.h
index c6f2f82bf..4e7f32423 100644
--- a/src/shader_recompiler/frontend/ir/pred.h
+++ b/src/shader_recompiler/frontend/ir/pred.h
@@ -19,8 +19,8 @@ enum class Pred : u64 {
PT,
};
-constexpr size_t NUM_USER_PREDS = 6;
-constexpr size_t NUM_PREDS = 7;
+constexpr size_t NUM_USER_PREDS = 7;
+constexpr size_t NUM_PREDS = 8;
[[nodiscard]] constexpr size_t PredIndex(Pred pred) noexcept {
return static_cast<size_t>(pred);
diff --git a/src/shader_recompiler/frontend/maxwell/program.cpp b/src/shader_recompiler/frontend/maxwell/program.cpp
index 16cdc12e2..ed5dbf41f 100644
--- a/src/shader_recompiler/frontend/maxwell/program.cpp
+++ b/src/shader_recompiler/frontend/maxwell/program.cpp
@@ -56,12 +56,12 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
.post_order_blocks{},
});
}
+ fmt::print(stdout, "{}\n", IR::DumpProgram(program));
Optimization::LowerFp16ToFp32(program);
for (IR::Function& function : functions) {
function.post_order_blocks = PostOrder(function.blocks);
Optimization::SsaRewritePass(function.post_order_blocks);
}
- fmt::print(stdout, "{}\n", IR::DumpProgram(program));
Optimization::GlobalMemoryToStorageBufferPass(program);
for (IR::Function& function : functions) {
Optimization::PostOrderInvoke(Optimization::ConstantPropagationPass, function);
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
index cb3a326cf..219ffcc6a 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp
@@ -9,7 +9,6 @@
namespace Shader::Maxwell {
namespace {
-
void FADD(TranslatorVisitor& v, u64 insn, bool sat, bool cc, bool ftz, FpRounding fp_rounding,
const IR::F32& src_b, bool abs_a, bool neg_a, bool abs_b, bool neg_b) {
union {
@@ -18,9 +17,6 @@ void FADD(TranslatorVisitor& v, u64 insn, bool sat, bool cc, bool ftz, FpRoundin
BitField<8, 8, IR::Reg> src_a;
} const fadd{insn};
- if (sat) {
- throw NotImplementedException("FADD SAT");
- }
if (cc) {
throw NotImplementedException("FADD CC");
}
@@ -31,7 +27,11 @@ void FADD(TranslatorVisitor& v, u64 insn, bool sat, bool cc, bool ftz, FpRoundin
.rounding{CastFpRounding(fp_rounding)},
.fmz_mode{ftz ? IR::FmzMode::FTZ : IR::FmzMode::None},
};
- v.F(fadd.dest_reg, v.ir.FPAdd(op_a, op_b, control));
+ IR::F32 value{v.ir.FPAdd(op_a, op_b, control)};
+ if (sat) {
+ value = v.ir.FPSaturate(value);
+ }
+ v.F(fadd.dest_reg, value);
}
void FADD(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
@@ -53,15 +53,15 @@ void FADD(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
} // Anonymous namespace
void TranslatorVisitor::FADD_reg(u64 insn) {
- FADD(*this, insn, GetReg20F(insn));
+ FADD(*this, insn, GetRegFloat20(insn));
}
-void TranslatorVisitor::FADD_cbuf(u64) {
- throw NotImplementedException("FADD (cbuf)");
+void TranslatorVisitor::FADD_cbuf(u64 insn) {
+ FADD(*this, insn, GetFloatCbuf(insn));
}
-void TranslatorVisitor::FADD_imm(u64) {
- throw NotImplementedException("FADD (imm)");
+void TranslatorVisitor::FADD_imm(u64 insn) {
+ FADD(*this, insn, GetFloatImm20(insn));
}
void TranslatorVisitor::FADD32I(u64) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp
index 4d82a0009..81175627f 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp
@@ -158,7 +158,7 @@ void TranslatorVisitor::F2I_cbuf(u64 insn) {
case SrcFormat::F16:
return IR::F16{ir.CompositeExtract(ir.UnpackFloat2x16(GetCbuf(insn)), f2i.half)};
case SrcFormat::F32:
- return GetCbufF(insn);
+ return GetFloatCbuf(insn);
case SrcFormat::F64: {
return UnpackCbuf(*this, insn);
}
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
index 1464f2807..758700d3c 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
@@ -51,7 +51,7 @@ void FFMA(TranslatorVisitor& v, u64 insn, const IR::F32& src_b, const IR::F32& s
} // Anonymous namespace
void TranslatorVisitor::FFMA_reg(u64 insn) {
- FFMA(*this, insn, GetReg20F(insn), GetReg39F(insn));
+ FFMA(*this, insn, GetRegFloat20(insn), GetRegFloat39(insn));
}
void TranslatorVisitor::FFMA_rc(u64) {
@@ -59,7 +59,7 @@ void TranslatorVisitor::FFMA_rc(u64) {
}
void TranslatorVisitor::FFMA_cr(u64 insn) {
- FFMA(*this, insn, GetCbufF(insn), GetReg39F(insn));
+ FFMA(*this, insn, GetFloatCbuf(insn), GetRegFloat39(insn));
}
void TranslatorVisitor::FFMA_imm(u64) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
index 1b1d38be7..5c38d3fc1 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp
@@ -91,7 +91,7 @@ void FMUL(TranslatorVisitor& v, u64 insn, const IR::F32& src_b) {
} // Anonymous namespace
void TranslatorVisitor::FMUL_reg(u64 insn) {
- return FMUL(*this, insn, GetReg20F(insn));
+ return FMUL(*this, insn, GetRegFloat20(insn));
}
void TranslatorVisitor::FMUL_cbuf(u64) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
index 079e3497f..be17bb0d9 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
@@ -48,11 +48,11 @@ IR::U32 TranslatorVisitor::GetReg39(u64 insn) {
return X(reg.index);
}
-IR::F32 TranslatorVisitor::GetReg20F(u64 insn) {
+IR::F32 TranslatorVisitor::GetRegFloat20(u64 insn) {
return ir.BitCast<IR::F32>(GetReg20(insn));
}
-IR::F32 TranslatorVisitor::GetReg39F(u64 insn) {
+IR::F32 TranslatorVisitor::GetRegFloat39(u64 insn) {
return ir.BitCast<IR::F32>(GetReg39(insn));
}
@@ -73,7 +73,7 @@ IR::U32 TranslatorVisitor::GetCbuf(u64 insn) {
return ir.GetCbuf(binding, byte_offset);
}
-IR::F32 TranslatorVisitor::GetCbufF(u64 insn) {
+IR::F32 TranslatorVisitor::GetFloatCbuf(u64 insn) {
return ir.BitCast<IR::F32>(GetCbuf(insn));
}
@@ -88,6 +88,17 @@ IR::U32 TranslatorVisitor::GetImm20(u64 insn) {
return ir.Imm32(value);
}
+IR::F32 TranslatorVisitor::GetFloatImm20(u64 insn) {
+ union {
+ u64 raw;
+ BitField<20, 19, u64> value;
+ BitField<56, 1, u64> is_negative;
+ } const imm{insn};
+ const f32 positive_value{Common::BitCast<f32>(static_cast<u32>(imm.value) << 12)};
+ const f32 value{imm.is_negative != 0 ? -positive_value : positive_value};
+ return ir.Imm32(value);
+}
+
IR::U32 TranslatorVisitor::GetImm32(u64 insn) {
union {
u64 raw;
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
index 27aba2cf8..4d4cf2ebf 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
@@ -304,13 +304,14 @@ public:
[[nodiscard]] IR::U32 GetReg8(u64 insn);
[[nodiscard]] IR::U32 GetReg20(u64 insn);
[[nodiscard]] IR::U32 GetReg39(u64 insn);
- [[nodiscard]] IR::F32 GetReg20F(u64 insn);
- [[nodiscard]] IR::F32 GetReg39F(u64 insn);
+ [[nodiscard]] IR::F32 GetRegFloat20(u64 insn);
+ [[nodiscard]] IR::F32 GetRegFloat39(u64 insn);
[[nodiscard]] IR::U32 GetCbuf(u64 insn);
- [[nodiscard]] IR::F32 GetCbufF(u64 insn);
+ [[nodiscard]] IR::F32 GetFloatCbuf(u64 insn);
[[nodiscard]] IR::U32 GetImm20(u64 insn);
+ [[nodiscard]] IR::F32 GetFloatImm20(u64 insn);
[[nodiscard]] IR::U32 GetImm32(u64 insn);
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp
index 623e78ff8..1493e1815 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp
@@ -84,8 +84,8 @@ void TranslatorVisitor::IADD_cbuf(u64 insn) {
IADD(*this, insn, GetCbuf(insn));
}
-void TranslatorVisitor::IADD_imm(u64) {
- throw NotImplementedException("IADD (imm)");
+void TranslatorVisitor::IADD_imm(u64 insn) {
+ IADD(*this, insn, GetImm20(insn));
}
void TranslatorVisitor::IADD32I(u64 insn) {
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/move_predicate_to_register.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/move_predicate_to_register.cpp
new file mode 100644
index 000000000..4324fd443
--- /dev/null
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/move_predicate_to_register.cpp
@@ -0,0 +1,66 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "shader_recompiler/exception.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class Mode : u64 {
+ PR,
+ CC,
+};
+} // Anonymous namespace
+
+void TranslatorVisitor::P2R_reg(u64) {
+ throw NotImplementedException("P2R (reg)");
+}
+
+void TranslatorVisitor::P2R_cbuf(u64) {
+ throw NotImplementedException("P2R (cbuf)");
+}
+
+void TranslatorVisitor::P2R_imm(u64 insn) {
+ union {
+ u64 raw;
+ BitField<0, 8, IR::Reg> dest_reg;
+ BitField<8, 8, IR::Reg> src;
+ BitField<40, 1, Mode> mode;
+ BitField<41, 2, u64> byte_selector;
+ } const p2r{insn};
+
+ const u32 mask{GetImm20(insn).U32()};
+ const bool pr_mode{p2r.mode == Mode::PR};
+ const u32 num_items{pr_mode ? 7U : 4U};
+ const u32 offset{static_cast<u32>(p2r.byte_selector) * 8};
+ IR::U32 insert{ir.Imm32(0)};
+ for (u32 index = 0; index < num_items; ++index) {
+ if (((mask >> index) & 1) == 0) {
+ continue;
+ }
+ const IR::U1 cond{[this, index, pr_mode] {
+ if (pr_mode) {
+ return ir.GetPred(IR::Pred{index});
+ }
+ switch (index) {
+ case 0:
+ return ir.GetZFlag();
+ case 1:
+ return ir.GetSFlag();
+ case 2:
+ return ir.GetCFlag();
+ case 3:
+ return ir.GetOFlag();
+ }
+ throw LogicError("Unreachable P2R index");
+ }()};
+ const IR::U32 bit{ir.Select(cond, ir.Imm32(1U << (index + offset)), ir.Imm32(0))};
+ insert = ir.BitwiseOr(insert, bit);
+ }
+ const IR::U32 masked_out{ir.BitwiseAnd(X(p2r.src), ir.Imm32(~(mask << offset)))};
+ X(p2r.dest_reg, ir.BitwiseOr(masked_out, insert));
+}
+
+} // namespace Shader::Maxwell
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
index 6b2a1356b..628cf1c14 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@@ -633,18 +633,6 @@ void TranslatorVisitor::OUT_imm(u64) {
ThrowNotImplemented(Opcode::OUT_imm);
}
-void TranslatorVisitor::P2R_reg(u64) {
- ThrowNotImplemented(Opcode::P2R_reg);
-}
-
-void TranslatorVisitor::P2R_cbuf(u64) {
- ThrowNotImplemented(Opcode::P2R_cbuf);
-}
-
-void TranslatorVisitor::P2R_imm(u64) {
- ThrowNotImplemented(Opcode::P2R_imm);
-}
-
void TranslatorVisitor::PBK() {
// PBK is a no-op
}
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 49ff911d6..b25af6cd3 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -191,12 +191,12 @@ ComputePipeline PipelineCache::CreateComputePipeline(ShaderInfo* shader_info) {
.support_fp32_denorm_flush = float_control.shaderDenormFlushToZeroFloat32 != VK_FALSE,
};
const auto [info, code]{Shader::RecompileSPIRV(profile, env, qmd.program_start)};
-
+ /*
FILE* file = fopen("D:\\shader.spv", "wb");
fwrite(code.data(), 4, code.size(), file);
fclose(file);
std::system("spirv-dis D:\\shader.spv");
-
+ */
shader_info->unique_hash = env.ComputeHash();
shader_info->size_bytes = env.ShaderSize();
return ComputePipeline{device, descriptor_pool, update_descriptor_queue, info,