From f91859efd259995806c2944f7941b105b58300d3 Mon Sep 17 00:00:00 2001
From: ReinUsesLisp <reinuseslisp@airmail.cc>
Date: Sat, 20 Mar 2021 05:04:12 -0300
Subject: shader: Implement I2F
---
.../frontend/maxwell/translate/impl/impl.cpp | 21 +++
.../frontend/maxwell/translate/impl/impl.h | 2 +
.../impl/integer_floating_point_conversion.cpp | 173 +++++++++++++++++++++
.../maxwell/translate/impl/not_implemented.cpp | 12 --
.../maxwell/translate/impl/texture_fetch.cpp | 2 +-
.../translate/impl/texture_fetch_swizzled.cpp | 2 +-
6 files changed, 198 insertions(+), 14 deletions(-)
create mode 100644 src/shader_recompiler/frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
(limited to 'src/shader_recompiler/frontend/maxwell/translate/impl')
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
index e444dcd4fb..c9af830101 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp
@@ -121,6 +121,22 @@ IR::F64 TranslatorVisitor::GetDoubleCbuf(u64 insn) {
return ir.PackDouble2x32(ir.CompositeConstruct(lower_bits, value));
}
+IR::U64 TranslatorVisitor::GetPackedCbuf(u64 insn) {
+ union {
+ u64 raw;
+ BitField<20, 1, u64> unaligned;
+ } const cbuf{insn};
+
+ if (cbuf.unaligned != 0) {
+ throw NotImplementedException("Unaligned packed constant buffer read");
+ }
+ const auto [binding, lower_offset]{CbufAddr(insn)};
+ const IR::U32 upper_offset{ir.Imm32(lower_offset.U32() + 4)};
+ const IR::U32 lower_value{ir.GetCbuf(binding, lower_offset)};
+ const IR::U32 upper_value{ir.GetCbuf(binding, upper_offset)};
+ return ir.PackUint2x32(ir.CompositeConstruct(lower_value, upper_value));
+}
+
IR::U32 TranslatorVisitor::GetImm20(u64 insn) {
union {
u64 raw;
@@ -158,6 +174,11 @@ IR::F64 TranslatorVisitor::GetDoubleImm20(u64 insn) {
return ir.Imm64(Common::BitCast<f64>(value | sign_bit));
}
+IR::U64 TranslatorVisitor::GetPackedImm20(u64 insn) {
+ const s64 value{GetImm20(insn).U32()};
+ return ir.Imm64(static_cast<u64>(static_cast<s64>(value) << 32));
+}
+
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 ed81d9c369..cb66cca25f 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/impl.h
@@ -356,10 +356,12 @@ public:
[[nodiscard]] IR::U32 GetCbuf(u64 insn);
[[nodiscard]] IR::F32 GetFloatCbuf(u64 insn);
[[nodiscard]] IR::F64 GetDoubleCbuf(u64 insn);
+ [[nodiscard]] IR::U64 GetPackedCbuf(u64 insn);
[[nodiscard]] IR::U32 GetImm20(u64 insn);
[[nodiscard]] IR::F32 GetFloatImm20(u64 insn);
[[nodiscard]] IR::F64 GetDoubleImm20(u64 insn);
+ [[nodiscard]] IR::U64 GetPackedImm20(u64 insn);
[[nodiscard]] IR::U32 GetImm32(u64 insn);
[[nodiscard]] IR::F32 GetFloatImm32(u64 insn);
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
new file mode 100644
index 0000000000..e8b5ae1d2d
--- /dev/null
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
@@ -0,0 +1,173 @@
+// 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 "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class FloatFormat : u64 {
+ F16 = 1,
+ F32 = 2,
+ F64 = 3,
+};
+
+enum class IntFormat : u64 {
+ U8 = 0,
+ U16 = 1,
+ U32 = 2,
+ U64 = 3,
+};
+
+union Encoding {
+ u64 raw;
+ BitField<0, 8, IR::Reg> dest_reg;
+ BitField<8, 2, FloatFormat> float_format;
+ BitField<10, 2, IntFormat> int_format;
+ BitField<13, 1, u64> is_signed;
+ BitField<39, 2, FpRounding> fp_rounding;
+ BitField<41, 2, u64> selector;
+ BitField<47, 1, u64> cc;
+ BitField<45, 1, u64> neg;
+ BitField<49, 1, u64> abs;
+};
+
+bool Is64(u64 insn) {
+ return Encoding{insn}.int_format == IntFormat::U64;
+}
+
+int BitSize(FloatFormat format) {
+ switch (format) {
+ case FloatFormat::F16:
+ return 16;
+ case FloatFormat::F32:
+ return 32;
+ case FloatFormat::F64:
+ return 64;
+ }
+ throw NotImplementedException("Invalid float format {}", format);
+}
+
+IR::U32 SmallAbs(TranslatorVisitor& v, const IR::U32& value, int bitsize) {
+ const IR::U32 least_value{v.ir.Imm32(-(1 << (bitsize - 1)))};
+ const IR::U32 mask{v.ir.ShiftRightArithmetic(value, v.ir.Imm32(bitsize - 1))};
+ const IR::U32 absolute{v.ir.BitwiseXor(v.ir.IAdd(value, mask), mask)};
+ const IR::U1 is_least{v.ir.IEqual(value, least_value)};
+ return IR::U32{v.ir.Select(is_least, value, absolute)};
+}
+
+void I2F(TranslatorVisitor& v, u64 insn, IR::U32U64 src) {
+ const Encoding i2f{insn};
+ if (i2f.cc != 0) {
+ throw NotImplementedException("CC");
+ }
+ const bool is_signed{i2f.is_signed != 0};
+ int src_bitsize{};
+ switch (i2f.int_format) {
+ case IntFormat::U8:
+ src = v.ir.BitFieldExtract(src, v.ir.Imm32(static_cast<u32>(i2f.selector) * 8),
+ v.ir.Imm32(8), is_signed);
+ if (i2f.abs != 0) {
+ src = SmallAbs(v, src, 8);
+ }
+ src_bitsize = 8;
+ break;
+ case IntFormat::U16:
+ if (i2f.selector == 1 || i2f.selector == 3) {
+ throw NotImplementedException("Invalid U16 selector {}", i2f.selector.Value());
+ }
+ src = v.ir.BitFieldExtract(src, v.ir.Imm32(static_cast<u32>(i2f.selector) * 8),
+ v.ir.Imm32(16), is_signed);
+ if (i2f.abs != 0) {
+ src = SmallAbs(v, src, 16);
+ }
+ src_bitsize = 16;
+ break;
+ case IntFormat::U32:
+ case IntFormat::U64:
+ if (i2f.selector != 0) {
+ throw NotImplementedException("Unexpected selector {}", i2f.selector.Value());
+ }
+ if (i2f.abs != 0 && is_signed) {
+ src = v.ir.IAbs(src);
+ }
+ src_bitsize = i2f.int_format == IntFormat::U64 ? 64 : 32;
+ break;
+ }
+ const int conversion_src_bitsize{i2f.int_format == IntFormat::U64 ? 64 : 32};
+ const int dst_bitsize{BitSize(i2f.float_format)};
+ IR::F16F32F64 value{v.ir.ConvertIToF(dst_bitsize, conversion_src_bitsize, is_signed, src)};
+ if (i2f.neg != 0) {
+ if (i2f.abs != 0 || !is_signed) {
+ // We know the value is positive
+ value = v.ir.FPNeg(value);
+ } else {
+ // Only negate if the input isn't the lowest value
+ IR::U1 is_least;
+ if (src_bitsize == 64) {
+ is_least = v.ir.IEqual(src, v.ir.Imm64(std::numeric_limits<s64>::min()));
+ } else {
+ const IR::U32 least_value{v.ir.Imm32(-(1 << (src_bitsize - 1)))};
+ is_least = v.ir.IEqual(src, least_value);
+ }
+ value = IR::F16F32F64{v.ir.Select(is_least, value, v.ir.FPNeg(value))};
+ }
+ }
+ switch (i2f.float_format) {
+ case FloatFormat::F16: {
+ const IR::F16 zero{v.ir.FPConvert(16, v.ir.Imm32(0.0f))};
+ v.X(i2f.dest_reg, v.ir.PackFloat2x16(v.ir.CompositeConstruct(value, zero)));
+ break;
+ }
+ case FloatFormat::F32:
+ v.F(i2f.dest_reg, value);
+ break;
+ case FloatFormat::F64: {
+ if (!IR::IsAligned(i2f.dest_reg, 2)) {
+ throw NotImplementedException("Unaligned destination {}", i2f.dest_reg.Value());
+ }
+ const IR::Value vector{v.ir.UnpackDouble2x32(value)};
+ for (int i = 0; i < 2; ++i) {
+ v.X(i2f.dest_reg + i, IR::U32{v.ir.CompositeExtract(vector, i)});
+ }
+ break;
+ }
+ default:
+ throw NotImplementedException("Invalid float format {}", i2f.float_format.Value());
+ }
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::I2F_reg(u64 insn) {
+ if (Is64(insn)) {
+ union {
+ u64 raw;
+ BitField<20, 8, IR::Reg> reg;
+ } const value{insn};
+ const IR::Value regs{ir.CompositeConstruct(ir.GetReg(value.reg), ir.GetReg(value.reg + 1))};
+ I2F(*this, insn, ir.PackUint2x32(regs));
+ } else {
+ I2F(*this, insn, GetReg20(insn));
+ }
+}
+
+void TranslatorVisitor::I2F_cbuf(u64 insn) {
+ if (Is64(insn)) {
+ I2F(*this, insn, GetPackedCbuf(insn));
+ } else {
+ I2F(*this, insn, GetCbuf(insn));
+ }
+}
+
+void TranslatorVisitor::I2F_imm(u64 insn) {
+ if (Is64(insn)) {
+ I2F(*this, insn, GetPackedImm20(insn));
+ } else {
+ I2F(*this, insn, GetImm20(insn));
+ }
+}
+
+} // namespace Shader::Maxwell
\ No newline at end of file
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 a4367fc5a6..4078feafa8 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@@ -241,18 +241,6 @@ void TranslatorVisitor::HSETP2_imm(u64) {
ThrowNotImplemented(Opcode::HSETP2_imm);
}
-void TranslatorVisitor::I2F_reg(u64) {
- ThrowNotImplemented(Opcode::I2F_reg);
-}
-
-void TranslatorVisitor::I2F_cbuf(u64) {
- ThrowNotImplemented(Opcode::I2F_cbuf);
-}
-
-void TranslatorVisitor::I2F_imm(u64) {
- ThrowNotImplemented(Opcode::I2F_imm);
-}
-
void TranslatorVisitor::IDE(u64) {
ThrowNotImplemented(Opcode::IDE);
}
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch.cpp
index 0fbb87ec43..b691b4d1fd 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch.cpp
@@ -56,7 +56,7 @@ Shader::TextureType GetType(TextureType type, bool dc) {
}
IR::Value MakeCoords(TranslatorVisitor& v, IR::Reg reg, TextureType type) {
- const auto read_array{[&]() -> IR::F32 { return v.ir.ConvertUToF(32, v.X(reg)); }};
+ const auto read_array{[&]() -> IR::F32 { return v.ir.ConvertUToF(32, 16, v.X(reg)); }};
switch (type) {
case TextureType::_1D:
return v.F(reg);
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
index 54f0df7547..d5fda20f42 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
@@ -65,7 +65,7 @@ IR::Value Composite(TranslatorVisitor& v, Args... regs) {
}
IR::F32 ReadArray(TranslatorVisitor& v, const IR::U32& value) {
- return v.ir.ConvertUToF(32, v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(16)));
+ return v.ir.ConvertUToF(32, 16, v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(16)));
}
IR::Value Sample(TranslatorVisitor& v, u64 insn) {
--
cgit v1.2.3-70-g09d2