Merge pull request #2220 from KhronosGroup/pr-2208

Land PR 2208 and fix more issues with sample mask

reference/opt/shaders-hlsl/frag/sample-mask-in-and-out.frag

@@ -1,5 +1,5 @@
-static int gl_SampleMaskIn;
+static int gl_SampleMaskIn[1];
-static int gl_SampleMask;
+static int gl_SampleMask[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Input
@@ -16,15 +16,15 @@ struct SPIRV_Cross_Output
 void frag_main()
 {
     FragColor = 1.0f.xxxx;
-    gl_SampleMask = gl_SampleMaskIn;
+    gl_SampleMask[0] = gl_SampleMaskIn[0];
 }
 
 SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
 {
-    gl_SampleMaskIn = stage_input.gl_SampleMaskIn;
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
     frag_main();
     SPIRV_Cross_Output stage_output;
-    stage_output.gl_SampleMask = gl_SampleMask;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
     stage_output.FragColor = FragColor;
     return stage_output;
 }

reference/opt/shaders-hlsl/frag/sample-mask-in.frag

@@ -1,5 +1,5 @@
 static int gl_SampleID;
-static int gl_SampleMaskIn;
+static int gl_SampleMaskIn[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Input
@@ -15,7 +15,7 @@ struct SPIRV_Cross_Output
 
 void frag_main()
 {
-    if ((gl_SampleMaskIn & (1 << gl_SampleID)) != 0)
+    if ((gl_SampleMaskIn[0] & (1 << gl_SampleID)) != 0)
     {
         FragColor = 1.0f.xxxx;
     }
@@ -24,7 +24,7 @@ void frag_main()
 SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
 {
     gl_SampleID = stage_input.gl_SampleID;
-    gl_SampleMaskIn = stage_input.gl_SampleMaskIn;
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
     frag_main();
     SPIRV_Cross_Output stage_output;
     stage_output.FragColor = FragColor;

reference/opt/shaders-hlsl/frag/sample-mask-out.frag

@@ -1,4 +1,4 @@
-static int gl_SampleMask;
+static int gl_SampleMask[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Output
@@ -10,14 +10,14 @@ struct SPIRV_Cross_Output
 void frag_main()
 {
     FragColor = 1.0f.xxxx;
-    gl_SampleMask = 0;
+    gl_SampleMask[0] = 0;
 }
 
 SPIRV_Cross_Output main()
 {
     frag_main();
     SPIRV_Cross_Output stage_output;
-    stage_output.gl_SampleMask = gl_SampleMask;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
     stage_output.FragColor = FragColor;
     return stage_output;
 }

reference/opt/shaders-msl/frag/post-depth-coverage.ios.msl2.frag

@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 

reference/opt/shaders-msl/frag/post-depth-coverage.msl23.frag

@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 

reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag

@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]], uint gl_SampleID
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22 & (1 << gl_SampleID));
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22 & (1 << gl_SampleID)));
     out.gl_SampleMask &= 0x22;
     return out;
 }

reference/opt/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag

@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22);
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22));
     out.gl_SampleMask &= 0x22;
     return out;
 }

reference/opt/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag

@@ -1,8 +1,49 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
 #include <metal_stdlib>
 #include <simd/simd.h>
 
 using namespace metal;
 
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+    
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+    
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+    
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+    
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
 struct type_View
 {
     float4x4 View_TranslatedWorldToClip;
@@ -489,12 +530,12 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     if (View.View_NumSceneColorMSAASamples > 1)
     {
         _268 = _255 * float4(float(View.View_NumSceneColorMSAASamples) * 0.25);
-        _269 = gl_SampleMaskIn & 15u;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0] & 15u;
     }
     else
     {
         _268 = _255;
-        _269 = gl_SampleMaskIn;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0];
     }
     out.out_var_SV_Target0 = _268;
     out.gl_SampleMask = _269;

reference/shaders-hlsl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,26 @@
+static uint gl_SampleMaskIn[1];
+static uint gl_SampleMask[1];
+struct SPIRV_Cross_Input
+{
+    uint gl_SampleMaskIn : SV_Coverage;
+};
+
+struct SPIRV_Cross_Output
+{
+    uint gl_SampleMask : SV_Coverage;
+};
+
+void frag_main()
+{
+    uint copy_sample_mask[1] = gl_SampleMaskIn;
+    gl_SampleMask = copy_sample_mask;
+}
+
+SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
+{
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
+    frag_main();
+    SPIRV_Cross_Output stage_output;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
+    return stage_output;
+}

reference/shaders-hlsl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,26 @@
+static int gl_SampleMaskIn[1];
+static int gl_SampleMask[1];
+struct SPIRV_Cross_Input
+{
+    uint gl_SampleMaskIn : SV_Coverage;
+};
+
+struct SPIRV_Cross_Output
+{
+    uint gl_SampleMask : SV_Coverage;
+};
+
+void frag_main()
+{
+    int copy_sample_mask[1] = gl_SampleMaskIn;
+    gl_SampleMask = copy_sample_mask;
+}
+
+SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
+{
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
+    frag_main();
+    SPIRV_Cross_Output stage_output;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
+    return stage_output;
+}

reference/shaders-hlsl/frag/sample-mask-in-and-out.frag

@@ -1,5 +1,5 @@
-static int gl_SampleMaskIn;
+static int gl_SampleMaskIn[1];
-static int gl_SampleMask;
+static int gl_SampleMask[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Input
@@ -16,15 +16,15 @@ struct SPIRV_Cross_Output
 void frag_main()
 {
     FragColor = 1.0f.xxxx;
-    gl_SampleMask = gl_SampleMaskIn;
+    gl_SampleMask[0] = gl_SampleMaskIn[0];
 }
 
 SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
 {
-    gl_SampleMaskIn = stage_input.gl_SampleMaskIn;
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
     frag_main();
     SPIRV_Cross_Output stage_output;
-    stage_output.gl_SampleMask = gl_SampleMask;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
     stage_output.FragColor = FragColor;
     return stage_output;
 }

reference/shaders-hlsl/frag/sample-mask-in.frag

@@ -1,5 +1,5 @@
 static int gl_SampleID;
-static int gl_SampleMaskIn;
+static int gl_SampleMaskIn[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Input
@@ -15,7 +15,7 @@ struct SPIRV_Cross_Output
 
 void frag_main()
 {
-    if ((gl_SampleMaskIn & (1 << gl_SampleID)) != 0)
+    if ((gl_SampleMaskIn[0] & (1 << gl_SampleID)) != 0)
     {
         FragColor = 1.0f.xxxx;
     }
@@ -24,7 +24,7 @@ void frag_main()
 SPIRV_Cross_Output main(SPIRV_Cross_Input stage_input)
 {
     gl_SampleID = stage_input.gl_SampleID;
-    gl_SampleMaskIn = stage_input.gl_SampleMaskIn;
+    gl_SampleMaskIn[0] = stage_input.gl_SampleMaskIn;
     frag_main();
     SPIRV_Cross_Output stage_output;
     stage_output.FragColor = FragColor;

reference/shaders-hlsl/frag/sample-mask-out.frag

@@ -1,4 +1,4 @@
-static int gl_SampleMask;
+static int gl_SampleMask[1];
 static float4 FragColor;
 
 struct SPIRV_Cross_Output
@@ -10,14 +10,14 @@ struct SPIRV_Cross_Output
 void frag_main()
 {
     FragColor = 1.0f.xxxx;
-    gl_SampleMask = 0;
+    gl_SampleMask[0] = 0;
 }
 
 SPIRV_Cross_Output main()
 {
     frag_main();
     SPIRV_Cross_Output stage_output;
-    stage_output.gl_SampleMask = gl_SampleMask;
+    stage_output.gl_SampleMask = gl_SampleMask[0];
     stage_output.FragColor = FragColor;
     return stage_output;
 }

reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,59 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+    
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+    
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+    
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+    
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
+struct main0_out
+{
+    uint gl_SampleMask [[sample_mask]];
+};
+
+fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
+{
+    main0_out out = {};
+    spvUnsafeArray<uint, 1> copy_sample_mask = spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) });
+    out.gl_SampleMask = copy_sample_mask[0];
+    return out;
+}
+

reference/shaders-msl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,59 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+    
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+    
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+    
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+    
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
+struct main0_out
+{
+    uint gl_SampleMask [[sample_mask]];
+};
+
+fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
+{
+    main0_out out = {};
+    spvUnsafeArray<int, 1> copy_sample_mask = spvUnsafeArray<int, 1>({ int(gl_SampleMaskIn) });
+    out.gl_SampleMask = copy_sample_mask[0];
+    return out;
+}
+

reference/shaders-msl/frag/post-depth-coverage.ios.msl2.frag

@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 

reference/shaders-msl/frag/post-depth-coverage.msl23.frag

@@ -11,7 +11,7 @@ struct main0_out
 [[ early_fragment_tests ]] fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask, post_depth_coverage]])
 {
     main0_out out = {};
-    out.FragColor = float4(float(gl_SampleMaskIn));
+    out.FragColor = float4(float(int(gl_SampleMaskIn)));
     return out;
 }
 

reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.force-sample.frag

@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]], uint gl_SampleID
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22 & (1 << gl_SampleID));
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22 & (1 << gl_SampleID)));
     out.gl_SampleMask &= 0x22;
     return out;
 }

reference/shaders-msl/frag/sample-mask-in-and-out.fixed-sample-mask.frag

@@ -13,7 +13,7 @@ fragment main0_out main0(uint gl_SampleMaskIn [[sample_mask]])
 {
     main0_out out = {};
     out.FragColor = float4(1.0);
-    out.gl_SampleMask = (gl_SampleMaskIn & 0x22);
+    out.gl_SampleMask = int((gl_SampleMaskIn & 0x22));
     out.gl_SampleMask &= 0x22;
     return out;
 }

reference/shaders-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,16 @@
+#version 450
+
+void main()
+{
+    uint _16_unrolled[1];
+    for (int i = 0; i < int(1); i++)
+    {
+        _16_unrolled[i] = int(gl_SampleMaskIn[i]);
+    }
+    uint copy_sample_mask[1] = _16_unrolled;
+    for (int i = 0; i < int(1); i++)
+    {
+        gl_SampleMask[i] = int(copy_sample_mask[i]);
+    }
+}
+

reference/shaders-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,16 @@
+#version 450
+
+void main()
+{
+    int _21_unrolled[1];
+    for (int i = 0; i < int(1); i++)
+    {
+        _21_unrolled[i] = gl_SampleMaskIn[i];
+    }
+    int copy_sample_mask[1] = _21_unrolled;
+    for (int i = 0; i < int(1); i++)
+    {
+        gl_SampleMask[i] = copy_sample_mask[i];
+    }
+}
+

reference/shaders-ue4/asm/frag/sample-mask-not-array.asm.frag

@@ -1,8 +1,49 @@
+#pragma clang diagnostic ignored "-Wmissing-prototypes"
+#pragma clang diagnostic ignored "-Wmissing-braces"
+
 #include <metal_stdlib>
 #include <simd/simd.h>
 
 using namespace metal;
 
+template<typename T, size_t Num>
+struct spvUnsafeArray
+{
+    T elements[Num ? Num : 1];
+    
+    thread T& operator [] (size_t pos) thread
+    {
+        return elements[pos];
+    }
+    constexpr const thread T& operator [] (size_t pos) const thread
+    {
+        return elements[pos];
+    }
+    
+    device T& operator [] (size_t pos) device
+    {
+        return elements[pos];
+    }
+    constexpr const device T& operator [] (size_t pos) const device
+    {
+        return elements[pos];
+    }
+    
+    constexpr const constant T& operator [] (size_t pos) const constant
+    {
+        return elements[pos];
+    }
+    
+    threadgroup T& operator [] (size_t pos) threadgroup
+    {
+        return elements[pos];
+    }
+    constexpr const threadgroup T& operator [] (size_t pos) const threadgroup
+    {
+        return elements[pos];
+    }
+};
+
 struct type_View
 {
     float4x4 View_TranslatedWorldToClip;
@@ -489,12 +530,12 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     if (View.View_NumSceneColorMSAASamples > 1)
     {
         _268 = _255 * float4(float(View.View_NumSceneColorMSAASamples) * 0.25);
-        _269 = gl_SampleMaskIn & 15u;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0] & 15u;
     }
     else
     {
         _268 = _255;
-        _269 = gl_SampleMaskIn;
+        _269 = (spvUnsafeArray<uint, 1>({ uint(gl_SampleMaskIn) }))[0];
     }
     out.out_var_SV_Target0 = _268;
     out.gl_SampleMask = _269;

shaders-hlsl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,40 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %uint %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

shaders-hlsl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,45 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+        %int = OpTypeInt 32 1
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %int %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+      %int_0 = OpConstant %int 0
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Input_int = OpTypePointer Input %int
+%_ptr_Function_int = OpTypePointer Function %int
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+%_ptr_Output_int = OpTypePointer Output %int
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

shaders-msl-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,40 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %uint %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

shaders-msl-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,45 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+        %int = OpTypeInt 32 1
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %int %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+      %int_0 = OpConstant %int 0
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Input_int = OpTypePointer Input %int
+%_ptr_Function_int = OpTypePointer Function %int
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+%_ptr_Output_int = OpTypePointer Output %int
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

shaders-no-opt/asm/frag/sample-mask-load-store-array-uint.asm.frag

@@ -0,0 +1,40 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %uint %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

shaders-no-opt/asm/frag/sample-mask-load-store-array.asm.frag

@@ -0,0 +1,45 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 11
+; Bound: 30
+; Schema: 0
+               OpCapability Shader
+          %1 = OpExtInstImport "GLSL.std.450"
+               OpMemoryModel Logical GLSL450
+               OpEntryPoint Fragment %main "main" %gl_SampleMaskIn %gl_SampleMask
+               OpExecutionMode %main OriginUpperLeft
+               OpSource GLSL 450
+               OpName %main "main"
+               OpName %copy_sample_mask "copy_sample_mask"
+               OpName %gl_SampleMaskIn "gl_SampleMaskIn"
+               OpName %out_sample_mask "out_sample_mask"
+               OpName %gl_SampleMask "gl_SampleMask"
+               OpDecorate %gl_SampleMaskIn Flat
+               OpDecorate %gl_SampleMaskIn BuiltIn SampleMask
+               OpDecorate %gl_SampleMask BuiltIn SampleMask
+       %void = OpTypeVoid
+          %3 = OpTypeFunction %void
+        %int = OpTypeInt 32 1
+       %uint = OpTypeInt 32 0
+     %uint_1 = OpConstant %uint 1
+%_arr_int_uint_1 = OpTypeArray %int %uint_1
+%_ptr_Function__arr_int_uint_1 = OpTypePointer Function %_arr_int_uint_1
+      %int_0 = OpConstant %int 0
+%_ptr_Input__arr_int_uint_1 = OpTypePointer Input %_arr_int_uint_1
+%gl_SampleMaskIn = OpVariable %_ptr_Input__arr_int_uint_1 Input
+%_ptr_Input_int = OpTypePointer Input %int
+%_ptr_Function_int = OpTypePointer Function %int
+%_ptr_Output__arr_int_uint_1 = OpTypePointer Output %_arr_int_uint_1
+%gl_SampleMask = OpVariable %_ptr_Output__arr_int_uint_1 Output
+%_ptr_Output_int = OpTypePointer Output %int
+       %main = OpFunction %void None %3
+          %5 = OpLabel
+%copy_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+%out_sample_mask = OpVariable %_ptr_Function__arr_int_uint_1 Function
+
+         %loaded_sample_mask_in = OpLoad %_arr_int_uint_1 %gl_SampleMaskIn
+		   OpStore %copy_sample_mask %loaded_sample_mask_in
+		%loaded_copy = OpLoad %_arr_int_uint_1 %copy_sample_mask
+		   OpStore %gl_SampleMask %loaded_copy
+               OpReturn
+               OpFunctionEnd

spirv_glsl.cpp

@@ -10158,10 +10158,11 @@ string CompilerGLSL::access_chain_internal(uint32_t base, const uint32_t *indice
 				if (!pending_array_enclose)
 					expr += "]";
 			}
+			else if (index_is_literal || !builtin_translates_to_nonarray(BuiltIn(get_decoration(base, DecorationBuiltIn))))
+			{
 				// Some builtins are arrays in SPIR-V but not in other languages, e.g. gl_SampleMask[] is an array in SPIR-V but not in Metal.
 				// By throwing away the index, we imply the index was 0, which it must be for gl_SampleMask.
-			else if (!builtin_translates_to_nonarray(BuiltIn(get_decoration(base, DecorationBuiltIn))))
+				// For literal indices we are working on composites, so we ignore this since we have already converted to proper array.
-			{
 				append_index(index, is_literal);
 			}
 
@@ -17737,6 +17738,25 @@ void CompilerGLSL::cast_from_variable_load(uint32_t source_id, std::string &expr
 		expr = bitcast_expression(expr_type, expected_type, expr);
 }
 
+SPIRType::BaseType CompilerGLSL::get_builtin_basetype(BuiltIn builtin, SPIRType::BaseType default_type)
+{
+	// TODO: Fill in for more builtins.
+	switch (builtin)
+	{
+	case BuiltInLayer:
+	case BuiltInPrimitiveId:
+	case BuiltInViewportIndex:
+	case BuiltInFragStencilRefEXT:
+	case BuiltInSampleMask:
+	case BuiltInPrimitiveShadingRateKHR:
+	case BuiltInShadingRateKHR:
+		return SPIRType::Int;
+
+	default:
+		return default_type;
+	}
+}
+
 void CompilerGLSL::cast_to_variable_store(uint32_t target_id, std::string &expr, const SPIRType &expr_type)
 {
 	auto *var = maybe_get_backing_variable(target_id);
@@ -17748,24 +17768,7 @@ void CompilerGLSL::cast_to_variable_store(uint32_t target_id, std::string &expr,
 		return;
 
 	auto builtin = static_cast<BuiltIn>(get_decoration(target_id, DecorationBuiltIn));
-	auto expected_type = expr_type.basetype;
+	auto expected_type = get_builtin_basetype(builtin, expr_type.basetype);
-
-	// TODO: Fill in for more builtins.
-	switch (builtin)
-	{
-	case BuiltInLayer:
-	case BuiltInPrimitiveId:
-	case BuiltInViewportIndex:
-	case BuiltInFragStencilRefEXT:
-	case BuiltInSampleMask:
-	case BuiltInPrimitiveShadingRateKHR:
-	case BuiltInShadingRateKHR:
-		expected_type = SPIRType::Int;
-		break;
-
-	default:
-		break;
-	}
 
 	if (expected_type != expr_type.basetype)
 	{

spirv_glsl.hpp

@@ -993,6 +993,7 @@ protected:
 	// Builtins in GLSL are always specific signedness, but the SPIR-V can declare them
 	// as either unsigned or signed.
 	// Sometimes we will need to automatically perform casts on load and store to make this work.
+	virtual SPIRType::BaseType get_builtin_basetype(spv::BuiltIn builtin, SPIRType::BaseType default_type);
 	virtual void cast_to_variable_store(uint32_t target_id, std::string &expr, const SPIRType &expr_type);
 	virtual void cast_from_variable_load(uint32_t source_id, std::string &expr, const SPIRType &expr_type);
 	void unroll_array_from_complex_load(uint32_t target_id, uint32_t source_id, std::string &expr);

spirv_hlsl.cpp

@@ -1149,15 +1149,31 @@ void CompilerHLSL::emit_builtin_variables()
 	builtins.merge_or(active_output_builtins);
 
 	std::unordered_map<uint32_t, ID> builtin_to_initializer;
+
+	// We need to declare sample mask with the same type that module declares it.
+	// Sample mask is somewhat special in that SPIR-V has an array, and we can copy that array, so we need to
+	// match sign.
+	SPIRType::BaseType sample_mask_in_basetype = SPIRType::Void;
+	SPIRType::BaseType sample_mask_out_basetype = SPIRType::Void;
+
 	ir.for_each_typed_id<SPIRVariable>([&](uint32_t, SPIRVariable &var) {
-		if (!is_builtin_variable(var) || var.storage != StorageClassOutput || !var.initializer)
+		if (!is_builtin_variable(var))
 			return;
 
+		auto &type = this->get<SPIRType>(var.basetype);
+		auto builtin = BuiltIn(get_decoration(var.self, DecorationBuiltIn));
+
+		if (var.storage == StorageClassInput && builtin == BuiltInSampleMask)
+			sample_mask_in_basetype = type.basetype;
+		else if (var.storage == StorageClassOutput && builtin == BuiltInSampleMask)
+			sample_mask_out_basetype = type.basetype;
+
+		if (var.initializer && var.storage == StorageClassOutput)
+		{
 			auto *c = this->maybe_get<SPIRConstant>(var.initializer);
 			if (!c)
 				return;
 
-		auto &type = this->get<SPIRType>(var.basetype);
 			if (type.basetype == SPIRType::Struct)
 			{
 				uint32_t member_count = uint32_t(type.member_types.size());
@@ -1171,7 +1187,10 @@ void CompilerHLSL::emit_builtin_variables()
 				}
 			}
 			else if (has_decoration(var.self, DecorationBuiltIn))
-			builtin_to_initializer[get_decoration(var.self, DecorationBuiltIn)] = var.initializer;
+			{
+				builtin_to_initializer[builtin] = var.initializer;
+			}
+		}
 	});
 
 	// Emit global variables for the interface variables which are statically used by the shader.
@@ -1288,7 +1307,11 @@ void CompilerHLSL::emit_builtin_variables()
 			break;
 
 		case BuiltInSampleMask:
-			type = "int";
+			if (active_input_builtins.get(BuiltInSampleMask))
+				type = sample_mask_in_basetype == SPIRType::UInt ? "uint" : "int";
+			else
+				type = sample_mask_out_basetype == SPIRType::UInt ? "uint" : "int";
+			array_size = 1;
 			break;
 
 		case BuiltInPrimitiveId:
@@ -1322,6 +1345,10 @@ void CompilerHLSL::emit_builtin_variables()
 		// declared the input variable and we need to add the output one now.
 		if (builtin == BuiltInSampleMask && storage == StorageClassInput && this->active_output_builtins.get(i))
 		{
+			type = sample_mask_out_basetype == SPIRType::UInt ? "uint" : "int";
+			if (array_size)
+				statement("static ", type, " ", this->builtin_to_glsl(builtin, StorageClassOutput), "[", array_size, "]", init_expr, ";");
+			else
 				statement("static ", type, " ", this->builtin_to_glsl(builtin, StorageClassOutput), init_expr, ";");
 		}
 	});
@@ -1536,6 +1563,18 @@ void CompilerHLSL::replace_illegal_names()
 	CompilerGLSL::replace_illegal_names();
 }
 
+SPIRType::BaseType CompilerHLSL::get_builtin_basetype(BuiltIn builtin, SPIRType::BaseType default_type)
+{
+	switch (builtin)
+	{
+	case BuiltInSampleMask:
+		// We declare sample mask array with module type, so always use default_type here.
+		return default_type;
+	default:
+		return CompilerGLSL::get_builtin_basetype(builtin, default_type);
+	}
+}
+
 void CompilerHLSL::emit_resources()
 {
 	auto &execution = get_entry_point();
@@ -3121,6 +3160,10 @@ void CompilerHLSL::emit_hlsl_entry_point()
 			statement(builtin, " = int(stage_input.", builtin, ");");
 			break;
 
+		case BuiltInSampleMask:
+			statement(builtin, "[0] = stage_input.", builtin, ";");
+			break;
+
 		case BuiltInNumWorkgroups:
 		case BuiltInPointCoord:
 		case BuiltInSubgroupSize:
@@ -3295,6 +3338,10 @@ void CompilerHLSL::emit_hlsl_entry_point()
 					          cull, "];");
 				break;
 
+			case BuiltInSampleMask:
+				statement("stage_output.gl_SampleMask = gl_SampleMask[0];");
+				break;
+
 			default:
 			{
 				auto builtin_expr = builtin_to_glsl(static_cast<BuiltIn>(i), StorageClassOutput);
@@ -6751,11 +6798,6 @@ void CompilerHLSL::set_hlsl_force_storage_buffer_as_uav(uint32_t desc_set, uint3
 	force_uav_buffer_bindings.insert(pair);
 }
 
-bool CompilerHLSL::builtin_translates_to_nonarray(spv::BuiltIn builtin) const
-{
-	return (builtin == BuiltInSampleMask);
-}
-
 bool CompilerHLSL::is_user_type_structured(uint32_t id) const
 {
 	if (hlsl_options.preserve_structured_buffers)

spirv_hlsl.hpp

@@ -290,6 +290,8 @@ private:
 	const char *to_storage_qualifiers_glsl(const SPIRVariable &var) override;
 	void replace_illegal_names() override;
 
+	SPIRType::BaseType get_builtin_basetype(spv::BuiltIn builtin, SPIRType::BaseType default_type) override;
+
 	bool is_hlsl_force_storage_buffer_as_uav(ID id) const;
 
 	Options hlsl_options;
@@ -400,9 +402,6 @@ private:
 		bool used = false;
 	} base_vertex_info;
 
-	// Returns true for BuiltInSampleMask because gl_SampleMask[] is an array in SPIR-V, but SV_Coverage is a scalar in HLSL.
-	bool builtin_translates_to_nonarray(spv::BuiltIn builtin) const override;
-
 	// Returns true if the specified ID has a UserTypeGOOGLE decoration for StructuredBuffer or RWStructuredBuffer resources.
 	bool is_user_type_structured(uint32_t id) const override;
 

spirv_msl.cpp

@@ -4911,9 +4911,18 @@ void CompilerMSL::emit_store_statement(uint32_t lhs_expression, uint32_t rhs_exp
 
 	bool transpose = lhs_e && lhs_e->need_transpose;
 
-	// No physical type remapping, and no packed type, so can just emit a store directly.
+	if (has_decoration(lhs_expression, DecorationBuiltIn) &&
-	if (!lhs_remapped_type && !lhs_packed_type)
+	    BuiltIn(get_decoration(lhs_expression, DecorationBuiltIn)) == BuiltInSampleMask &&
+	    type_is_top_level_array(type))
 	{
+		// Storing an array to SampleMask, have to remove the array-ness before storing.
+		statement(to_expression(lhs_expression), " = ", to_enclosed_unpacked_expression(rhs_expression), "[0];");
+		register_write(lhs_expression);
+	}
+	else if (!lhs_remapped_type && !lhs_packed_type)
+	{
+		// No physical type remapping, and no packed type, so can just emit a store directly.
+
 		// We might not be dealing with remapped physical types or packed types,
 		// but we might be doing a clean store to a row-major matrix.
 		// In this case, we just flip transpose states, and emit the store, a transpose must be in the RHS expression, if any.
@@ -17178,6 +17187,7 @@ void CompilerMSL::cast_from_variable_load(uint32_t source_id, std::string &expr,
 	case BuiltInInstanceIndex:
 	case BuiltInBaseInstance:
 	case BuiltInBaseVertex:
+	case BuiltInSampleMask:
 		expected_type = SPIRType::UInt;
 		expected_width = 32;
 		break;
@@ -17195,9 +17205,17 @@ void CompilerMSL::cast_from_variable_load(uint32_t source_id, std::string &expr,
 		break;
 	}
 
-	if (expected_type != expr_type.basetype)
+	if (type_is_top_level_array(expr_type) && builtin == BuiltInSampleMask)
 	{
-		if (!expr_type.array.empty() && (builtin == BuiltInTessLevelInner || builtin == BuiltInTessLevelOuter))
+		// Needs special handling.
+		auto wrap_expr = join(type_to_glsl(expr_type), "({ ");
+		wrap_expr += join(type_to_glsl(get<SPIRType>(expr_type.parent_type)), "(", expr, ")");
+		wrap_expr += " })";
+		expr = std::move(wrap_expr);
+	}
+	else if (expected_type != expr_type.basetype)
+	{
+		if (type_is_top_level_array(expr_type) && (builtin == BuiltInTessLevelInner || builtin == BuiltInTessLevelOuter))
 		{
 			// Triggers when loading TessLevel directly as an array.
 			// Need explicit padding + cast.