From cc88530a722cc2ce7a09f2a39dadeeb504e2f221 Mon Sep 17 00:00:00 2001
From: Jay Kwak <82421531+jkwak-work@users.noreply.github.com>
Date: Tue, 14 May 2024 20:28:28 -0700
Subject: Support combined textures for Metal target (#4169)

---
 source/slang/hlsl.meta.slang | 106 +++++++++++++++++++++++++++++++++----------
 source/slang/slang-emit.cpp  |  11 ++++-
 2 files changed, 92 insertions(+), 25 deletions(-)

(limited to 'source')

diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang
index f318eb79e..176c7c0e4 100644
--- a/source/slang/hlsl.meta.slang
+++ b/source/slang/hlsl.meta.slang
@@ -529,13 +529,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_querylod)]
+    [require(glsl_hlsl_metal_spirv, texture_querylod)]
     float CalculateLevelOfDetail(TextureCoord location)
     {
         __requireComputeDerivative();
         __target_switch
         {
         case hlsl:
+        case metal:
             {
                 return __getTexture().CalculateLevelOfDetail(__getSampler(), location);
             }
@@ -551,13 +552,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_querylod)]
+    [require(glsl_hlsl_metal_spirv, texture_querylod)]
     float CalculateLevelOfDetailUnclamped(TextureCoord location)
     {
         __requireComputeDerivative();
         __target_switch
         {
         case hlsl:
+        case metal:
             {
             return __getTexture().CalculateLevelOfDetailUnclamped(__getSampler(), location);
             }
@@ -573,7 +575,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_cuda_glsl_hlsl_spirv, texture_sm_4_1_fragment)]
+    [require(cpp_cuda_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
     T Sample(vector<float, Shape.dimensions+isArray> location)
     {
         __requireComputeDerivative();
@@ -581,6 +583,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().Sample(__getSampler(), location);
             case glsl:
                 __intrinsic_asm "$ctexture($0, $1)$z";
@@ -622,10 +625,34 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
         }
     }
 
+    [__readNone]
+    [ForceInline]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
+    T Sample(vector<float, Shape.dimensions+isArray> location, constexpr vector<int, Shape.planeDimensions> offset)
+    {
+        __requireComputeDerivative();
+        __target_switch
+        {
+            case cpp:
+            case hlsl:
+            case metal:
+                return __getTexture().Sample(__getSampler(), location, offset);
+            case glsl:
+                __intrinsic_asm "$ctextureOffsetClampARB($0, $1, $2)$z";
+            case spirv:
+                return spirv_asm
+                {
+                    OpCapability MinLod;
+                    %sampled : __sampledType(T) = OpImageSampleImplicitLod $this $location None|ConstOffset $offset;
+                    __truncate $$T result __sampledType(T) %sampled;
+                };
+        }
+    }
+
     [__readNone]
     [ForceInline]
     __glsl_extension(GL_ARB_sparse_texture_clamp)
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_fragment)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
     T Sample(vector<float, Shape.dimensions+isArray> location, vector<int, Shape.planeDimensions> offset, float clamp)
     {
         __requireComputeDerivative();
@@ -633,6 +660,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().Sample(__getSampler(), location, offset, clamp);
             case glsl:
                 __intrinsic_asm "$ctextureOffsetClampARB($0, $1, $2, $3)$z";
@@ -648,7 +676,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_fragment)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
     T Sample(vector<float, Shape.dimensions+isArray> location, vector<int, Shape.planeDimensions> offset, float clamp, out uint status)
     {
         __target_switch
@@ -662,7 +690,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_fragment)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
     T SampleBias(vector<float, Shape.dimensions+isArray> location, float bias)
     {
         __requireComputeDerivative();
@@ -670,6 +698,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleBias(__getSampler(), location, bias);
             case glsl:
                 __intrinsic_asm "$ctexture($0, $1, $2)$z";
@@ -685,7 +714,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_fragment)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_fragment)]
     T SampleBias(vector<float, Shape.dimensions+isArray> location, float bias, constexpr vector<int, Shape.planeDimensions> offset)
     {
         __requireComputeDerivative();
@@ -693,6 +722,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleBias(__getSampler(), location, bias, offset);
             case glsl:
                 __intrinsic_asm "$ctextureOffset($0, $1, $3, $2)$z";
@@ -707,7 +737,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_shadowlod)]
+    [require(glsl_hlsl_metal_spirv, texture_shadowlod)]
     float SampleCmp(vector<float, Shape.dimensions+isArray> location, float compareValue)
     {
         __target_switch
@@ -726,6 +756,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
                 return __glsl_texture(this, __makeVector(location, compareValue));
             }
         case hlsl:
+        case metal:
             return __getTexture().SampleCmp(__getComparisonSampler(), location, compareValue);
         case spirv:
             return spirv_asm
@@ -737,7 +768,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_shadowlod)]
+    [require(glsl_hlsl_metal_spirv, texture_shadowlod)]
     float SampleCmpLevelZero(vector<float, Shape.dimensions+isArray> location, float compareValue)
     {
         __target_switch
@@ -752,6 +783,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
                 return __glsl_texture_level_zero(this, __makeVector(location, compareValue));
             }
         case hlsl:
+        case metal:
             return __getTexture().SampleCmpLevelZero(__getComparisonSampler(), location, compareValue);
         case spirv:
             const float zeroFloat = 0.0f;
@@ -764,7 +796,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone] 
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_shadowlod)]
+    [require(glsl_hlsl_metal_spirv, texture_shadowlod)]
     float SampleCmp(vector<float, Shape.dimensions+isArray> location, float compareValue, constexpr vector<int, Shape.planeDimensions> offset)
     {
         __target_switch
@@ -779,6 +811,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
                 return __glsl_texture_offset(this, __makeVector(location, compareValue), offset);
             }
         case hlsl:
+        case metal:
             return __getTexture().SampleCmp(__getComparisonSampler(), location, compareValue, offset);
         case spirv:
             return spirv_asm
@@ -790,7 +823,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(glsl_hlsl_spirv, texture_shadowlod)]
+    [require(glsl_hlsl_metal_spirv, texture_shadowlod)]
     float SampleCmpLevelZero(vector<float, Shape.dimensions+isArray> location, float compareValue, constexpr vector<int, Shape.planeDimensions> offset)
     {
         __target_switch
@@ -805,6 +838,7 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
                 return __glsl_texture_offset_level_zero(this, __makeVector(location, compareValue), offset);
             }
         case hlsl:
+        case metal:
             return __getTexture().SampleCmpLevelZero(__getComparisonSampler(), location, compareValue, offset);
         case spirv:
             const float zeroFloat = 0.0f;
@@ -817,13 +851,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1)]
     T SampleGrad(vector<float, Shape.dimensions+isArray> location, vector<float, Shape.dimensions> gradX, vector<float, Shape.dimensions> gradY)
     {
         __target_switch
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleGrad(__getSampler(), location, gradX, gradY);
 
             case glsl:
@@ -839,13 +874,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1)]
     T SampleGrad(vector<float, Shape.dimensions+isArray> location, vector<float, Shape.dimensions> gradX, vector<float, Shape.dimensions> gradY, constexpr vector<int, Shape.dimensions> offset)
     {
         __target_switch
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleGrad(__getSampler(), location, gradX, gradY, offset);
             case glsl:
             __intrinsic_asm "$ctextureGradOffset($0, $1, $2, $3, $4)$z";
@@ -861,13 +897,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
     [__readNone]
     [ForceInline]
     __glsl_extension(GL_ARB_sparse_texture_clamp)
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1)]
     T SampleGrad(vector<float, Shape.dimensions+isArray> location, vector<float, Shape.dimensions> gradX, vector<float, Shape.dimensions> gradY, constexpr vector<int, Shape.dimensions> offset, float lodClamp)
     {
         __target_switch
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleGrad(__getSampler(), location, gradX, gradY, offset, lodClamp);
             case glsl:
             __intrinsic_asm "$ctextureGradOffsetClampARB($0, $1, $2, $3, $4, $5)$z";
@@ -883,13 +920,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_cuda_glsl_hlsl_spirv, texture_sm_4_1)]
+    [require(cpp_cuda_glsl_hlsl_metal_spirv, texture_sm_4_1)]
     T SampleLevel(vector<float, Shape.dimensions+isArray> location, float level)
     {
         __target_switch
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleLevel(__getSampler(), location, level);
             case glsl:
                 __intrinsic_asm "$ctextureLod($0, $1, $2)$z";
@@ -935,13 +973,14 @@ extension __TextureImpl<T,Shape,isArray,isMS,sampleCount,0,isShadow,1,format>
 
     [__readNone]
     [ForceInline]
-    [require(cpp_glsl_hlsl_spirv, texture_sm_4_1)]
+    [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1)]
     T SampleLevel(vector<float, Shape.dimensions+isArray> location, float level, constexpr vector<int, Shape.planeDimensions> offset)
     {
         __target_switch
         {
             case cpp:
             case hlsl:
+            case metal:
                 return __getTexture().SampleLevel(__getSampler(), location, level, offset);
 
             case glsl:
@@ -2109,15 +2148,29 @@ for (int isScalarTexture = 0; isScalarTexture <= 1; isScalarTexture++) {
 }}}}
 { // begin extension for gather
 ${{{{
+    if (isCombined)
+    {
+}}}}
+    __intrinsic_op($(kIROp_CombinedTextureSamplerGetTexture))
+    __TextureImpl<T, Shape, isArray, isMS, sampleCount, 0, isShadow, 0, format> __getTexture();
+
+    __intrinsic_op($(kIROp_CombinedTextureSamplerGetSampler))
+    SamplerState __getSampler();
+
+    __intrinsic_op($(kIROp_CombinedTextureSamplerGetSampler))
+    SamplerComparisonState __getComparisonSampler();
+${{{{
+    }
+
     // Gather component
     const char* samplerStateParam = isCombined ? "" : " s,";
-    const char* metalSupport = isCombined ? "" : "metal_";
-    const char* caseMetal = isCombined ? "" : "case metal:";
+    const char* getTexture = isCombined ? "__getTexture()" : "this";
     for (int isCmp = 0; isCmp < 2; ++isCmp) {
         const char* cmp = isCmp ? "Cmp" : "";
         const char* cmpParam = isCmp ? ", T compareValue" : "";
         const char* compareArg = isCmp ? ", compareValue" : "";
         const char* samplerStateType = isCombined ? "" : (isCmp ? "SamplerComparisonState" : "SamplerState");
+        const char* getSampler = isCombined ? (isCmp ? " __getComparisonSampler()," : " __getSampler(),") : samplerStateParam;
         const char* componentNames[] = { "", "Red", "Green", "Blue", "Alpha"};
         const char* glslComponentNames[] = { ", 0", ", 1", ", 2", ", 3" };
 
@@ -2126,26 +2179,28 @@ ${{{{
             auto glslComponent = (isCmp ? "" :glslComponentNames[componentId == 0 ? 0 : componentId - 1]);
     }}}}
     [ForceInline]
-    [require(glsl_hlsl_$(metalSupport)spirv, texture_gather)]
+    [require(glsl_hlsl_metal_spirv, texture_gather)]
     vector<T,4> Gather$(cmp)$(componentName)($(samplerStateType)$(samplerStateParam) vector<float, Shape.dimensions+isArray> location $(cmpParam))
     {
         __target_switch
         {
         case hlsl: __intrinsic_asm ".Gather$(cmp)$(componentName)";
-        $(caseMetal)
+        case metal:
+            return __texture_gather$(cmp)<T>($(getTexture),$(getSampler) location $(compareArg) $(glslComponent));
         case glsl:
         case spirv:
             return __texture_gather$(cmp)<T>(this,$(samplerStateParam) location $(compareArg) $(glslComponent));
         }
     }
     [ForceInline]
-    [require(glsl_hlsl_$(metalSupport)spirv, texture_gather)]
+    [require(glsl_hlsl_metal_spirv, texture_gather)]
     vector<T,4> Gather$(cmp)$(componentName)($(samplerStateType)$(samplerStateParam) vector<float, Shape.dimensions+isArray> location $(cmpParam), constexpr vector<int, Shape.planeDimensions> offset)
     {
         __target_switch
         {
         case hlsl: __intrinsic_asm ".Gather$(cmp)$(componentName)";
-        $(caseMetal)
+        case metal:
+            return __texture_gather$(cmp)_offset<T>($(getTexture),$(getSampler) location $(compareArg), offset $(glslComponent));
         case glsl:
         case spirv:
             return __texture_gather$(cmp)_offset<T>(this,$(samplerStateParam) location $(compareArg), offset $(glslComponent));
@@ -2347,7 +2402,7 @@ extension __TextureImpl<T,Shape,isArray,0,sampleCount,0,isShadow,isCombined,form
         __glsl_extension(GL_EXT_samplerless_texture_functions)
         [__readNone]
         [ForceInline]
-        [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_samplerless)]
+        [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_samplerless)]
         get
         {
             __target_switch
@@ -2355,6 +2410,8 @@ extension __TextureImpl<T,Shape,isArray,0,sampleCount,0,isShadow,isCombined,form
                 case cpp:
                 case hlsl:
                     __intrinsic_asm ".operator[]";
+                case metal:
+                    return Load(__makeVector(location, 0));
                 case glsl:
                     return __glsl_load(location);
                 case spirv:
@@ -2511,7 +2568,7 @@ extension __TextureImpl<T,Shape,isArray,1,sampleCount,0,isShadow,isCombined,form
         __glsl_extension(GL_EXT_samplerless_texture_functions)
         [__readNone]
         [ForceInline]
-        [require(cpp_glsl_hlsl_spirv, texture_sm_4_1_samplerless)]
+        [require(cpp_glsl_hlsl_metal_spirv, texture_sm_4_1_samplerless)]
         get
         {
             __target_switch
@@ -2519,6 +2576,7 @@ extension __TextureImpl<T,Shape,isArray,1,sampleCount,0,isShadow,isCombined,form
                 case cpp:
                 case hlsl:
                     __intrinsic_asm "($0).sample[$2][$1]";
+                case metal:
                 case glsl:
                 case spirv:
                     return Load(location, sampleIndex);
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index ab22e1f90..7cc056009 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -550,9 +550,18 @@ Result linkAndOptimizeIR(
         lowerAppendConsumeStructuredBuffers(targetProgram, irModule, sink);
     }
 
-    if (target == CodeGenTarget::HLSL || ArtifactDescUtil::isCpuLikeTarget(artifactDesc))
+    switch (target)
     {
+    default:
+        if (!ArtifactDescUtil::isCpuLikeTarget(artifactDesc))
+            break;
+        [[fallthrough]];
+    case CodeGenTarget::HLSL:
+    case CodeGenTarget::Metal:
+    case CodeGenTarget::MetalLib:
+    case CodeGenTarget::MetalLibAssembly:
         lowerCombinedTextureSamplers(irModule, sink);
+        break;
     }
 
     addUserTypeHintDecorations(irModule);
-- 
cgit v1.2.3