From 630273b6295f524401abf1c131dba09fdd055911 Mon Sep 17 00:00:00 2001
From: FernandoS27 <fsahmkow27@gmail.com>
Date: Mon, 29 Mar 2021 02:52:52 +0200
Subject: [PATCH] shader: Implement TXD

---
 src/shader_recompiler/CMakeLists.txt          |   1 +
 .../frontend/maxwell/maxwell.inc              |   4 +-
 .../translate/impl/not_implemented.cpp        |   8 -
 .../translate/impl/texture_gradient.cpp       | 180 ++++++++++++++++++
 4 files changed, 183 insertions(+), 10 deletions(-)
 create mode 100644 src/shader_recompiler/frontend/maxwell/translate/impl/texture_gradient.cpp

diff --git a/src/shader_recompiler/CMakeLists.txt b/src/shader_recompiler/CMakeLists.txt
index d9a2b9cb4..03a5793aa 100644
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@@ -133,6 +133,7 @@ add_library(shader_recompiler STATIC
     frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
     frontend/maxwell/translate/impl/texture_gather_swizzled.cpp
     frontend/maxwell/translate/impl/texture_gather.cpp
+    frontend/maxwell/translate/impl/texture_gradient.cpp
     frontend/maxwell/translate/impl/texture_load_swizzled.cpp
     frontend/maxwell/translate/impl/texture_load.cpp
     frontend/maxwell/translate/impl/texture_mipmap_level.cpp
diff --git a/src/shader_recompiler/frontend/maxwell/maxwell.inc b/src/shader_recompiler/frontend/maxwell/maxwell.inc
index b47fb9c2e..c759bd4d4 100644
--- a/src/shader_recompiler/frontend/maxwell/maxwell.inc
+++ b/src/shader_recompiler/frontend/maxwell/maxwell.inc
@@ -261,8 +261,8 @@ INST(TLDS,         "TLDS",           "1101 -01- ---- ----")
 INST(TMML,         "TMML",           "1101 1111 0101 1---")
 INST(TMML_b,       "TMML (b)",       "1101 1111 0110 0---")
 INST(TXA,          "TXA",            "1101 1111 0100 0---")
-INST(TXD,          "TXD",            "1101 1110 0011 10--")
-INST(TXD_b,        "TXD (b)",        "1101 1110 0111 10--")
+INST(TXD,          "TXD",            "1101 1110 00-- ----")
+INST(TXD_b,        "TXD (b)",        "1101 1110 01-- ----")
 INST(TXQ,          "TXQ",            "1101 1111 0100 1---")
 INST(TXQ_b,        "TXQ (b)",        "1101 1111 0101 0---")
 INST(VABSDIFF,     "VABSDIFF",       "0101 0100 ---- ----")
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 9f5ea7775..ba526817a 100644
--- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp
@@ -317,14 +317,6 @@ void TranslatorVisitor::TXA(u64) {
     ThrowNotImplemented(Opcode::TXA);
 }
 
-void TranslatorVisitor::TXD(u64) {
-    ThrowNotImplemented(Opcode::TXD);
-}
-
-void TranslatorVisitor::TXD_b(u64) {
-    ThrowNotImplemented(Opcode::TXD_b);
-}
-
 void TranslatorVisitor::VABSDIFF(u64) {
     ThrowNotImplemented(Opcode::VABSDIFF);
 }
diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_gradient.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_gradient.cpp
new file mode 100644
index 000000000..00768e167
--- /dev/null
+++ b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_gradient.cpp
@@ -0,0 +1,180 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <optional>
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/ir/modifiers.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+
+enum class TextureType : u64 {
+    _1D,
+    ARRAY_1D,
+    _2D,
+    ARRAY_2D,
+    _3D,
+    ARRAY_3D,
+    CUBE,
+    ARRAY_CUBE,
+};
+
+Shader::TextureType GetType(TextureType type, bool dc) {
+    switch (type) {
+    case TextureType::_1D:
+        return dc ? Shader::TextureType::Shadow1D : Shader::TextureType::Color1D;
+    case TextureType::ARRAY_1D:
+        return dc ? Shader::TextureType::ShadowArray1D : Shader::TextureType::ColorArray1D;
+    case TextureType::_2D:
+        return dc ? Shader::TextureType::Shadow2D : Shader::TextureType::Color2D;
+    case TextureType::ARRAY_2D:
+        return dc ? Shader::TextureType::ShadowArray2D : Shader::TextureType::ColorArray2D;
+    case TextureType::_3D:
+        return dc ? Shader::TextureType::Shadow3D : Shader::TextureType::Color3D;
+    case TextureType::ARRAY_3D:
+        throw NotImplementedException("3D array texture type");
+    case TextureType::CUBE:
+        return dc ? Shader::TextureType::ShadowCube : Shader::TextureType::ColorCube;
+    case TextureType::ARRAY_CUBE:
+        return dc ? Shader::TextureType::ShadowArrayCube : Shader::TextureType::ColorArrayCube;
+    }
+    throw NotImplementedException("Invalid texture type {}", type);
+}
+
+IR::Value MakeOffset(TranslatorVisitor& v, IR::Reg reg, bool has_lod_clamp) {
+    const IR::U32 value{v.X(reg)};
+    const u32 base = has_lod_clamp ? 12 : 16;
+    return v.ir.CompositeConstruct(
+        v.ir.BitFieldExtract(value, v.ir.Imm32(base), v.ir.Imm32(4), true),
+        v.ir.BitFieldExtract(value, v.ir.Imm32(base + 4), v.ir.Imm32(4), true));
+}
+
+void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
+    union {
+        u64 raw;
+        BitField<49, 1, u64> nodep;
+        BitField<35, 1, u64> aoffi;
+        BitField<50, 1, u64> lc;
+        BitField<51, 3, IR::Pred> sparse_pred;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> coord_reg;
+        BitField<20, 8, IR::Reg> derivate_reg;
+        BitField<28, 3, TextureType> type;
+        BitField<31, 4, u64> mask;
+        BitField<36, 13, u64> cbuf_offset;
+    } const txd{insn};
+
+    const bool has_lod_clamp = txd.lc != 0;
+    if (has_lod_clamp) {
+        throw NotImplementedException("TXD.LC - CLAMP is not implemented");
+    }
+
+    IR::Value coords;
+    u32 num_derivates;
+    IR::Reg base_reg = txd.coord_reg;
+    IR::Reg last_reg;
+    IR::Value handle;
+    if (!is_bindless) {
+        handle = v.ir.Imm32(static_cast<u32>(txd.cbuf_offset.Value() * 4));
+    } else {
+        handle = v.X(base_reg++);
+    }
+
+    const auto read_array{[&]() -> IR::F32 {
+        return v.ir.ConvertUToF(32, 16,
+                                v.ir.BitFieldExtract(v.X(last_reg), v.ir.Imm32(0),
+                                                     v.ir.Imm32(has_lod_clamp ? 12 : 16)));
+    }};
+    switch (txd.type) {
+    case TextureType::_1D: {
+        coords = v.F(base_reg);
+        num_derivates = 1;
+        last_reg = base_reg + 1;
+        break;
+    }
+    case TextureType::ARRAY_1D: {
+        last_reg = base_reg + 1;
+        coords = v.ir.CompositeConstruct(v.F(base_reg), read_array());
+        num_derivates = 1;
+        break;
+    }
+    case TextureType::_2D: {
+        last_reg = base_reg + 2;
+        coords = v.ir.CompositeConstruct(v.F(base_reg), v.F(base_reg + 1));
+        num_derivates = 2;
+        break;
+    }
+    case TextureType::ARRAY_2D: {
+        last_reg = base_reg + 2;
+        coords = v.ir.CompositeConstruct(v.F(base_reg), v.F(base_reg + 1), read_array());
+        num_derivates = 2;
+        break;
+    }
+    default:
+        throw NotImplementedException("Invalid texture type");
+    }
+
+    const IR::Reg derivate_reg{txd.derivate_reg};
+    IR::Value derivates;
+    switch (num_derivates) {
+    case 1: {
+        derivates = v.ir.CompositeConstruct(v.F(derivate_reg), v.F(derivate_reg + 1));
+        break;
+    }
+    case 2: {
+        derivates = v.ir.CompositeConstruct(v.F(derivate_reg), v.F(derivate_reg + 1),
+                                            v.F(derivate_reg + 2), v.F(derivate_reg + 3));
+        break;
+    }
+    default:
+        throw NotImplementedException("Invalid texture type");
+    }
+
+    IR::Value offset;
+    if (txd.aoffi != 0) {
+        offset = MakeOffset(v, last_reg, has_lod_clamp);
+    }
+
+    IR::F32 lod_clamp;
+    if (has_lod_clamp) {
+        const IR::F32 conv4_8fixp_f = v.ir.Imm32(Common::BitCast<f32>(0x3b800000U));
+        const IR::F32 tmp = v.ir.ConvertUToF(
+            32, 16, v.ir.BitFieldExtract(v.X(last_reg), v.ir.Imm32(20), v.ir.Imm32(12)));
+        lod_clamp = v.ir.FPMul(tmp, conv4_8fixp_f);
+    }
+
+    IR::TextureInstInfo info{};
+    info.type.Assign(GetType(txd.type, false));
+    info.num_derivates.Assign(num_derivates);
+    info.has_lod_clamp.Assign(has_lod_clamp ? 1 : 0);
+    const IR::Value sample{[&]() -> IR::Value {
+        return v.ir.ImageGradient(handle, coords, derivates, offset, lod_clamp, info);
+    }()};
+
+    IR::Reg dest_reg{txd.dest_reg};
+    for (size_t element = 0; element < 4; ++element) {
+        if (((txd.mask >> element) & 1) == 0) {
+            continue;
+        }
+        v.F(dest_reg, IR::F32{v.ir.CompositeExtract(sample, element)});
+        ++dest_reg;
+    }
+    if (txd.sparse_pred != IR::Pred::PT) {
+        v.ir.SetPred(txd.sparse_pred, v.ir.LogicalNot(v.ir.GetSparseFromOp(sample)));
+    }
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::TXD(u64 insn) {
+    Impl(*this, insn, false);
+}
+
+void TranslatorVisitor::TXD_b(u64 insn) {
+    Impl(*this, insn, true);
+}
+
+} // namespace Shader::Maxwell