shader: Address feedback

src/shader_recompiler/backend/spirv/emit_spirv_image.cpp

@@ -72,20 +72,19 @@ public:
     explicit ImageOperands(EmitContext& ctx, bool has_lod_clamp, Id derivates, u32 num_derivates,
                            Id offset, Id lod_clamp) {
         if (Sirit::ValidId(derivates)) {
-            boost::container::static_vector<Id, 3> deriv_x_accum;
-            boost::container::static_vector<Id, 3> deriv_y_accum;
-            for (size_t i = 0; i < num_derivates; i++) {
-                deriv_x_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2));
-                deriv_y_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2 + 1));
-            }
-            Id derivates_X = ctx.OpCompositeConstruct(
-                ctx.F32[num_derivates], std::span{deriv_x_accum.data(), deriv_x_accum.size()});
-            Id derivates_Y = ctx.OpCompositeConstruct(
-                ctx.F32[num_derivates], std::span{deriv_y_accum.data(), deriv_y_accum.size()});
-            Add(spv::ImageOperandsMask::Grad, derivates_X, derivates_Y);
-        } else {
             throw LogicError("Derivates must be present");
         }
+        boost::container::static_vector<Id, 3> deriv_x_accum;
+        boost::container::static_vector<Id, 3> deriv_y_accum;
+        for (size_t i = 0; i < num_derivates; i++) {
+            deriv_x_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2));
+            deriv_y_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2 + 1));
+        }
+        const Id derivates_X{ctx.OpCompositeConstruct(
+            ctx.F32[num_derivates], std::span{deriv_x_accum.data(), deriv_x_accum.size()})};
+        const Id derivates_Y{ctx.OpCompositeConstruct(
+            ctx.F32[num_derivates], std::span{deriv_y_accum.data(), deriv_y_accum.size()})};
+        Add(spv::ImageOperandsMask::Grad, derivates_X, derivates_Y);
         if (Sirit::ValidId(offset)) {
             Add(spv::ImageOperandsMask::Offset, offset);
         }
@@ -100,10 +99,10 @@ public:
         operands.push_back(value);
     }
 
-    void Add(spv::ImageOperandsMask new_mask, Id value, Id value_2) {
+    void Add(spv::ImageOperandsMask new_mask, Id value_1, Id value_2) {
         mask = static_cast<spv::ImageOperandsMask>(static_cast<unsigned>(mask) |
                                                    static_cast<unsigned>(new_mask));
-        operands.push_back(value);
+        operands.push_back(value_1);
         operands.push_back(value_2);
     }
 
@@ -345,7 +344,8 @@ Id EmitImageQueryLod(EmitContext& ctx, IR::Inst*, const IR::Value& index, Id coo
 Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
                      Id derivates, Id offset, Id lod_clamp) {
     const auto info{inst->Flags<IR::TextureInstInfo>()};
-    const ImageOperands operands(ctx, info.has_lod_clamp != 0, derivates, info.num_derivates, offset, lod_clamp);
+    const ImageOperands operands(ctx, info.has_lod_clamp != 0, derivates, info.num_derivates,
+                                 offset, lod_clamp);
     return Emit(&EmitContext::OpImageSparseSampleExplicitLod,
                 &EmitContext::OpImageSampleExplicitLod, ctx, inst, ctx.F32[4], Texture(ctx, index),
                 coords, operands.Mask(), operands.Span());

src/shader_recompiler/frontend/maxwell/translate/impl/texture_gradient.cpp

@@ -47,7 +47,7 @@ Shader::TextureType GetType(TextureType type, bool dc) {
 
 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;
+    const u32 base{has_lod_clamp ? 12U : 16U};
     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));
@@ -74,20 +74,21 @@ void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
     }
 
     IR::Value coords;
-    u32 num_derivates;
+    u32 num_derivates{};
-    IR::Reg base_reg = txd.coord_reg;
+    IR::Reg base_reg{txd.coord_reg};
     IR::Reg last_reg;
     IR::Value handle;
-    if (!is_bindless) {
+    if (is_bindless) {
-        handle = v.ir.Imm32(static_cast<u32>(txd.cbuf_offset.Value() * 4));
-    } else {
         handle = v.X(base_reg++);
+    } else {
+        handle = v.ir.Imm32(static_cast<u32>(txd.cbuf_offset.Value() * 4));
     }
 
     const auto read_array{[&]() -> IR::F32 {
-        return v.ir.ConvertUToF(32, 16,
+        const IR::U32 base{v.ir.Imm32(0)};
-                                v.ir.BitFieldExtract(v.X(last_reg), v.ir.Imm32(0),
+        const IR::U32 count{v.ir.Imm32(has_lod_clamp ? 12 : 16)};
-                                                     v.ir.Imm32(has_lod_clamp ? 12 : 16)));
+        const IR::U32 array_index{v.ir.BitFieldExtract(v.X(last_reg), base, count)};
+        return v.ir.ConvertUToF(32, 16, array_index);
     }};
     switch (txd.type) {
     case TextureType::_1D: {
@@ -141,19 +142,20 @@ void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
 
     IR::F32 lod_clamp;
     if (has_lod_clamp) {
-        const IR::F32 conv4_8fixp_f = v.ir.Imm32(Common::BitCast<f32>(0x3b800000U));
+        // Lod Clamp is a Fixed Point 4.8, we need to transform it to float.
-        const IR::F32 tmp = v.ir.ConvertUToF(
+        // to convert a fixed point, float(value) / float(1 << fixed_point)
-            32, 16, v.ir.BitFieldExtract(v.X(last_reg), v.ir.Imm32(20), v.ir.Imm32(12)));
+        // in this case the fixed_point is 8.
-        lod_clamp = v.ir.FPMul(tmp, conv4_8fixp_f);
+        const IR::F32 conv4_8fixp_f{v.ir.Imm32(static_cast<f32>(1U << 8))};
+        const IR::F32 fixp_lc{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(fixp_lc, 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 {
+    const IR::Value sample{v.ir.ImageGradient(handle, coords, derivates, offset, lod_clamp, info)};
-        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) {

src/shader_recompiler/frontend/maxwell/translate/impl/texture_load.cpp

@@ -117,10 +117,10 @@ void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
     IR::Value offset;
     IR::U32 lod;
     IR::U32 multisample;
-    if (!is_bindless) {
+    if (is_bindless) {
-        handle = v.ir.Imm32(static_cast<u32>(tld.cbuf_offset.Value() * 4));
-    } else {
         handle = v.X(meta_reg++);
+    } else {
+        handle = v.ir.Imm32(static_cast<u32>(tld.cbuf_offset.Value() * 4));
     }
     if (tld.lod != 0) {
         lod = v.X(meta_reg++);
@@ -138,9 +138,7 @@ void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
     }
     IR::TextureInstInfo info{};
     info.type.Assign(GetType(tld.type, false));
-    const IR::Value sample{[&]() -> IR::Value {
+    const IR::Value sample{v.ir.ImageFetch(handle, coords, offset, lod, multisample, info)};
-        return v.ir.ImageFetch(handle, coords, offset, lod, multisample, info);
-    }()};
 
     IR::Reg dest_reg{tld.dest_reg};
     for (size_t element = 0; element < 4; ++element) {

src/shader_recompiler/frontend/maxwell/translate/impl/texture_mipmap_level.cpp

@@ -81,39 +81,35 @@ void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
         BitField<36, 13, u64> cbuf_offset;
     } const tmml{insn};
 
-    if ((tmml.mask & 0xC) != 0) {
+    if ((tmml.mask & 0b1100) != 0) {
         throw NotImplementedException("TMML BA results are not implmented");
     }
 
-    IR::F32 transform_constant = v.ir.Imm32(256.0f);
+    IR::F32 transform_constant{v.ir.Imm32(256.0f)};
 
     const IR::Value coords{MakeCoords(v, tmml.coord_reg, tmml.type)};
 
     IR::U32 handle;
     IR::Reg meta_reg{tmml.meta_reg};
-    if (!is_bindless) {
+    if (is_bindless) {
-        handle = v.ir.Imm32(static_cast<u32>(tmml.cbuf_offset.Value() * 4));
-    } else {
         handle = v.X(meta_reg++);
+    } else {
+        handle = v.ir.Imm32(static_cast<u32>(tmml.cbuf_offset.Value() * 4));
     }
     IR::TextureInstInfo info{};
     info.type.Assign(GetType(tmml.type, false));
-    const IR::Value sample{
+    const IR::Value sample{v.ir.ImageQueryLod(handle, coords, info)};
-        [&]() -> IR::Value { return v.ir.ImageQueryLod(handle, coords, info); }()};
 
-    const IR::FpControl fp_control{
-        .no_contraction{false},
-        .rounding{IR::FpRounding::RP},
-        .fmz_mode{IR::FmzMode::FTZ},
-    };
     IR::Reg dest_reg{tmml.dest_reg};
     for (size_t element = 0; element < 4; ++element) {
         if (((tmml.mask >> element) & 1) == 0) {
             continue;
         }
-        IR::F32 value = IR::F32{v.ir.CompositeExtract(sample, element)};
+        IR::F32 value{v.ir.CompositeExtract(sample, element)};
-        v.F(dest_reg,
+        if (element < 2) {
-            element < 2 ? IR::F32{v.ir.FPMul(value, transform_constant, fp_control)} : value);
+            value = v.ir.FPMul(value, transform_constant);
+        }
+        v.F(dest_reg, value);
         ++dest_reg;
     }
 }

src/video_core/memory_manager.cpp

@@ -64,7 +64,7 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     }
     const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first);
     if (it != map_ranges.end()) {
-        // ASSERT(it->first == gpu_addr);
+        ASSERT(it->first == gpu_addr);
         map_ranges.erase(it);
     } else {
         UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr);