Use SPIRV integer vector dot product instructions (#6141)

* Use SPIRV integer vector dot product instructions

* fix test

---------

Co-authored-by: Yong He <yonghe@outlook.com>

Author: fairywreath

source/slang/hlsl.meta.slang

@@ -8292,6 +8292,30 @@ T dot(vector<T, N> x, vector<T, N> y)
     {
     case hlsl: __intrinsic_asm "dot";
     case wgsl: __intrinsic_asm "dot";
+    case spirv:
+    {
+        spirv_asm
+        {
+            OpCapability DotProduct;
+            OpCapability DotProductInputAll;
+            OpExtension "SPV_KHR_integer_dot_product";
+        };
+
+        if (__isSignedInt<T>())
+        {
+            return spirv_asm
+            {
+                result:$$T = OpSDot $x $y;
+            };
+        }
+        else
+        {
+            return spirv_asm
+            {
+                result:$$T = OpUDot $x $y;
+            };
+        }
+    }
     default:
         T result = T(0);
         for(int i = 0; i < N; ++i)

tests/hlsl-intrinsic/vector-dot-int.slang

@@ -1,23 +1,44 @@
-//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -shaderobj
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj
-//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -shaderobj
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj
-//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj
+//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -shaderobj -output-using-type
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -shaderobj -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj -output-using-type
+
+// No 16-bit and 64-bit integer support on DX11.
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type -xslang -DDX11
+
+//TEST(compute, vulkan):SIMPLE(filecheck=CHECK_SPV): -stage compute -entry computeMain -target spirv
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
 
 [numthreads(1, 1, 1)]
 void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
 {
-    int idx = int(dispatchThreadID.x);
+    int index = int(dispatchThreadID.x);
+    uint outIndex = 0;
 
-    float tmp = dot(float3(idx), float3(1));
-    
-    int3 a = { idx + 1, idx + 2, idx + 3};
+    // CHECK_SPV: OpSDot
+    int3 a = { index - 1, index - 2, index - 3};
     int3 b = { 1, 2, 3};
+    outputBuffer[outIndex++] = dot(a, b);
+
+    // CHECK_SPV: OpUDot
+    uint3 c = { index + 1, index + 2, index + 3};
+    uint3 d = { 2, 4, 6};
+    outputBuffer[outIndex++] = int(dot(c, d));
 
-    int result = dot(a, b);
+#if !defined(DX11)
+    // CHECK_SPV: OpUDot
+    uint64_t2 e = { index + 1, index + 2};
+    uint64_t2 f = { 4, 8};
+    outputBuffer[outIndex++] = int(dot(e, f));
 
-    outputBuffer[idx] = result;
-}
+    // CHECK_SPV: OpSDot
+    int16_t4 g = { int16_t(index + 1), int16_t(index + 2), int16_t(index + 3), int16_t(index + 4)};
+    int16_t4 h = { -1, 2, 2, -1};
+    outputBuffer[outIndex++] = int(dot(g, h));
+#else
+    outputBuffer[outIndex++] = 20;
+    outputBuffer[outIndex++] = 5;
+#endif
+}

tests/hlsl-intrinsic/vector-dot-int.slang.expected.txt

@@ -1,4 +1,5 @@
-E
-0
-0
-0
+type: int32_t
+-14
+28
+20
+5