Implement subgroup quad operations for Metal (#6745)

2 files changed, 33 insertions, 13 deletions

diff --git a/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-quad.slang b/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-quad.slang
index b847cf460..e5d4c9de0 100644
--- a/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-quad.slang
+++ b/tests/glsl-intrinsic/shader-subgroup/shader-subgroup-quad.slang
@@ -10,6 +10,7 @@
 //TEST(compute, vulkan):COMPARE_COMPUTE(filecheck-buffer=BUF):-vk -compute -entry computeMain -allow-glsl
 //TEST(compute, vulkan):COMPARE_COMPUTE(filecheck-buffer=BUF):-vk -compute -entry computeMain -allow-glsl -emit-spirv-directly
 //TEST(compute):COMPARE_COMPUTE(filecheck-buffer=BUF):-wgpu -compute -entry computeMain -allow-glsl -xslang -DWGPU
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=BUF):-metal -compute -entry computeMain -allow-glsl -xslang -DMETAL
 
 #version 430
 
@@ -27,7 +28,7 @@ bool test1QuadX() {
         & subgroupQuadSwapHorizontal(T(2)) == T(2)
         & subgroupQuadSwapVertical(T(2)) == T(2)
         & subgroupQuadSwapDiagonal(T(3)) == T(3)
-        // subgroupQuadBroadcast is not implemented for WGSL as the WGSL intrinsic only accepts const integers expressions, as of time of writing.
+        // subgroupQuadBroadcast is not implemented for WGSL and Metal as their intrinsics only accepts const integers expressions.
 #if !defined(WGPU)
         & subgroupQuadBroadcast(T(1), 1) == T(1)
 #endif
@@ -41,7 +42,7 @@ bool testVQuadX() {
         & subgroupQuadSwapHorizontal(gvec(T(2))) == gvec(T(2))
         & subgroupQuadSwapVertical(gvec(T(2))) == gvec(T(2))
         & subgroupQuadSwapDiagonal(gvec(T(3))) == gvec(T(3))
-        // subgroupQuadBroadcast is not implemented for WGSL as the WGSL intrinsic only accepts const integers expressions, as of time of writing.
+        // subgroupQuadBroadcast is not implemented for WGSL and Metal as their intrinsics only accepts const integers expressions.
 #if !defined(WGPU)
         & subgroupQuadBroadcast(gvec(T(1)), 1) == gvec(T(1))
 #endif
@@ -54,7 +55,7 @@ bool test1QuadX() {
         & subgroupQuadSwapHorizontal(T(2)) == T(2)
         & subgroupQuadSwapVertical(T(2)) == T(2)
         & subgroupQuadSwapDiagonal(T(3)) == T(3)
-        // subgroupQuadBroadcast is not implemented for WGSL as the WGSL intrinsic only accepts const integers expressions, as of time of writing.
+        // subgroupQuadBroadcast is not implemented for WGSL and Metal as their intrinsics only accepts const integers expressions.
 #if !defined(WGPU)
         & subgroupQuadBroadcast(T(1), 1) == T(1)
 #endif
@@ -68,7 +69,7 @@ bool testVQuadX() {
         & subgroupQuadSwapHorizontal(gvec(T(2))) == gvec(T(2))
         & subgroupQuadSwapVertical(gvec(T(2))) == gvec(T(2))
         & subgroupQuadSwapDiagonal(gvec(T(3))) == gvec(T(3))
-        // subgroupQuadBroadcast is not implemented for WGSL as the WGSL intrinsic only accepts const integers expressions, as of time of writing.
+        // subgroupQuadBroadcast is not implemented for WGSL and Metal as their intrinsics only accepts const integers expressions.
 #if !defined(WGPU)
         & subgroupQuadBroadcast(gvec(T(1)), 1) == gvec(T(1))
 #endif
@@ -93,8 +94,8 @@ bool testQuadSwapX() {
         & testVQuadX<uint, 3>()
         & testVQuadX<uint, 4>()
 
-        // Disabled on WGPU as these built-in types are not supported as of time of writing.
-#if !defined (WGPU)
+        // Disabled on WGSL and Metal as these built-in types are not supported.
+#if !defined(WGPU) && !defined(METAL)
         & test1QuadX<double>() // WARNING: intel GPU's lack FP64 support
         & testVQuadX<double, 2>()
         & testVQuadX<double, 3>()
@@ -115,19 +116,25 @@ bool testQuadSwapX() {
         & testVQuadX<uint8_t, 2>()
         & testVQuadX<uint8_t, 3>()
         & testVQuadX<uint8_t, 4>()
-        & test1QuadX<uint16_t>() 
-        & testVQuadX<uint16_t, 2>()
-        & testVQuadX<uint16_t, 3>()
-        & testVQuadX<uint16_t, 4>()
         & test1QuadX<uint64_t>() 
         & testVQuadX<uint64_t, 2>()
         & testVQuadX<uint64_t, 3>()
-        & testVQuadX<uint64_t, 4>()
-        & test1QuadX<bool>()
-        & testVQuadX<bool, 2>()
+        & testVQuadX<uint64_t, 4>() & test1QuadX<bool>() & testVQuadX<bool, 2>()
         & testVQuadX<bool, 3>()
         & testVQuadX<bool, 4>()
 #endif
+
+
+#if !defined(WGPU)
+        & test1QuadX<int16_t>() 
+        & testVQuadX<int16_t, 2>()
+        & testVQuadX<int16_t, 3>()
+        & testVQuadX<int16_t, 4>()
+        & test1QuadX<uint16_t>() 
+        & testVQuadX<uint16_t, 2>()
+        & testVQuadX<uint16_t, 3>()
+        & testVQuadX<uint16_t, 4>()
+#endif
         ;
 }
 
diff --git a/tests/hlsl-intrinsic/subgroup-quad.slang b/tests/hlsl-intrinsic/subgroup-quad.slang
index 928431a45..1cfbffb49 100644
--- a/tests/hlsl-intrinsic/subgroup-quad.slang
+++ b/tests/hlsl-intrinsic/subgroup-quad.slang
@@ -1,6 +1,7 @@
 //TEST:SIMPLE(filecheck=SPIRV): -entry main -stage compute -target spirv
 //TEST:SIMPLE(filecheck=SPIRV): -entry main -stage compute -target spirv -emit-spirv-directly
 //TEST:SIMPLE(filecheck=HLSL): -entry main -stage compute -target hlsl
+//TEST:SIMPLE(filecheck=METAL): -entry main -stage compute -target metal
 
 RWStructuredBuffer<float> output;
 
@@ -38,4 +39,16 @@ void main()
     // SPIRV: OpGroupNonUniformQuadSwap {{.*}} %{{u?int_3}} {{.*}} %{{u?int_1}}
     // SPIRV: OpGroupNonUniformQuadSwap {{.*}} %{{u?int_3}} {{.*}} %{{u?int_2}}
     // SPIRV: OpGroupNonUniformQuadSwap {{.*}} %{{u?int_3}} {{.*}} %{{u?int_2}}
+
+    // METAL: quad_shuffle
+    // METAL: quad_shuffle
+    // METAL: ^ 1
+    // METAL: quad_shuffle
+    // METAL: quad_shuffle
+    // METAL: ^ 2
+    // METAL: quad_shuffle
+    // METAL: quad_shuffle
+    // METAL: ^ 3
+    // METAL: quad_shuffle
+    // METAL: quad_shuffle
 }