Implement SnormNz conversion (#2841)

* +

* +

* Unpack Snorm 2x16

* +

* SintToSnormNz

* all is broken ig....

* review changes

* my stupid ass messed all while trying to resolve the conflicts..

* +

* +

* fix rebase

* clang-format fix (1)

* clang-format fix (2)

---------

Co-authored-by: squidbus <175574877+squidbus@users.noreply.github.com>

src/shader_recompiler/ir/passes/lower_buffer_format_to_raw.cpp

@@ -206,7 +206,8 @@ static void LowerBufferFormatInst(IR::Block& block, IR::Inst& inst, Info& info)
         .swizzle = is_inst_typed
                        ? AmdGpu::RemapSwizzle(flags.inst_data_fmt.Value(), AmdGpu::IdentityMapping)
                        : buffer.DstSelect(),
-        .num_conversion = is_inst_typed ? AmdGpu::MapNumberConversion(flags.inst_num_fmt.Value())
+        .num_conversion = is_inst_typed ? AmdGpu::MapNumberConversion(flags.inst_num_fmt.Value(),
+                                                                      flags.inst_data_fmt.Value())
                                         : buffer.GetNumberConversion(),
         .num_components = AmdGpu::NumComponents(data_format),
     };

src/shader_recompiler/ir/reinterpret.h

@@ -34,6 +34,18 @@ inline F32 ApplyReadNumberConversion(IREmitter& ir, const F32& value,
     case AmdGpu::NumberConversion::UnormToUbnorm:
         // Convert 0...1 to -1...1
         return ir.FPSub(ir.FPMul(value, ir.Imm32(2.f)), ir.Imm32(1.f));
+    case AmdGpu::NumberConversion::Sint8ToSnormNz: {
+        const IR::U32 additon = ir.IAdd(ir.IMul(ir.BitCast<U32>(value), ir.Imm32(2)), ir.Imm32(1));
+        const IR::F32 left = ir.ConvertSToF(32, 32, additon);
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u8>::max()));
+        return ir.FPDiv(left, max);
+    }
+    case AmdGpu::NumberConversion::Sint16ToSnormNz: {
+        const IR::U32 additon = ir.IAdd(ir.IMul(ir.BitCast<U32>(value), ir.Imm32(2)), ir.Imm32(1));
+        const IR::F32 left = ir.ConvertSToF(32, 32, additon);
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u16>::max()));
+        return ir.FPDiv(left, max);
+    }
     default:
         UNREACHABLE();
     }
@@ -66,6 +78,20 @@ inline F32 ApplyWriteNumberConversion(IREmitter& ir, const F32& value,
     case AmdGpu::NumberConversion::UnormToUbnorm:
         // Convert -1...1 to 0...1
         return ir.FPDiv(ir.FPAdd(value, ir.Imm32(1.f)), ir.Imm32(2.f));
+    case AmdGpu::NumberConversion::Sint8ToSnormNz: {
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u8>::max()));
+        const IR::F32 mul = ir.FPMul(ir.FPClamp(value, ir.Imm32(-1.f), ir.Imm32(1.f)), max);
+        const IR::F32 left = ir.FPSub(mul, ir.Imm32(1.f));
+        const IR::U32 raw = ir.ConvertFToS(32, ir.FPDiv(left, ir.Imm32(2.f)));
+        return ir.BitCast<F32>(raw);
+    }
+    case AmdGpu::NumberConversion::Sint16ToSnormNz: {
+        const IR::F32 max = ir.Imm32(float(std::numeric_limits<u16>::max()));
+        const IR::F32 mul = ir.FPMul(ir.FPClamp(value, ir.Imm32(-1.f), ir.Imm32(1.f)), max);
+        const IR::F32 left = ir.FPSub(mul, ir.Imm32(1.f));
+        const IR::U32 raw = ir.ConvertFToS(32, ir.FPDiv(left, ir.Imm32(2.f)));
+        return ir.BitCast<F32>(raw);
+    }
     default:
         UNREACHABLE();
     }

src/video_core/amdgpu/liverpool.h

@@ -928,7 +928,7 @@ struct Liverpool {
         }
 
         [[nodiscard]] NumberConversion GetNumberConversion() const {
-            return MapNumberConversion(GetFixedNumberFormat());
+            return MapNumberConversion(GetFixedNumberFormat(), info.format);
         }
 
         [[nodiscard]] CompMapping Swizzle() const {

src/video_core/amdgpu/resource.h

@@ -68,7 +68,7 @@ struct Buffer {
     }
 
     NumberConversion GetNumberConversion() const noexcept {
-        return MapNumberConversion(NumberFormat(num_format));
+        return MapNumberConversion(NumberFormat(num_format), DataFormat(data_format));
     }
 
     u32 GetStride() const noexcept {
@@ -292,7 +292,7 @@ struct Image {
     }
 
     NumberConversion GetNumberConversion() const noexcept {
-        return MapNumberConversion(NumberFormat(num_format));
+        return MapNumberConversion(NumberFormat(num_format), DataFormat(data_format));
     }
 
     TilingMode GetTilingMode() const {

src/video_core/amdgpu/types.h

@@ -197,6 +197,8 @@ enum class NumberConversion : u32 {
     UintToUscaled = 1,
     SintToSscaled = 2,
     UnormToUbnorm = 3,
+    Sint8ToSnormNz = 5,
+    Sint16ToSnormNz = 6,
 };
 
 struct CompMapping {
@@ -287,6 +289,7 @@ inline NumberFormat RemapNumberFormat(const NumberFormat format, const DataForma
     case NumberFormat::Uscaled:
         return NumberFormat::Uint;
     case NumberFormat::Sscaled:
+    case NumberFormat::SnormNz:
         return NumberFormat::Sint;
     case NumberFormat::Ubnorm:
         return NumberFormat::Unorm;
@@ -336,14 +339,28 @@ inline CompMapping RemapSwizzle(const DataFormat format, const CompMapping swizz
     }
 }
 
-inline NumberConversion MapNumberConversion(const NumberFormat format) {
+inline NumberConversion MapNumberConversion(const NumberFormat num_fmt, const DataFormat data_fmt) {
-    switch (format) {
+    switch (num_fmt) {
     case NumberFormat::Uscaled:
         return NumberConversion::UintToUscaled;
     case NumberFormat::Sscaled:
         return NumberConversion::SintToSscaled;
     case NumberFormat::Ubnorm:
         return NumberConversion::UnormToUbnorm;
+    case NumberFormat::SnormNz: {
+        switch (data_fmt) {
+        case DataFormat::Format8:
+        case DataFormat::Format8_8:
+        case DataFormat::Format8_8_8_8:
+            return NumberConversion::Sint8ToSnormNz;
+        case DataFormat::Format16:
+        case DataFormat::Format16_16:
+        case DataFormat::Format16_16_16_16:
+            return NumberConversion::Sint16ToSnormNz;
+        default:
+            UNREACHABLE_MSG("data_fmt = {}", u32(data_fmt));
+        }
+    }
     default:
         return NumberConversion::None;
     }