Support cooperative vector including Vulkan-Header (#6228)

* Support cooperative vector including Vulkan-Header

Adding a Slang support for cooperative vector with vulkan-header update.

12 files changed, 286 insertions, 0 deletions

diff --git a/tests/cooperative-vector/glsl/cast.slang b/tests/cooperative-vector/glsl/cast.slang
new file mode 100644
index 000000000..263ad5026
--- /dev/null
+++ b/tests/cooperative-vector/glsl/cast.slang
@@ -0,0 +1,30 @@
+//DISABLE_TEST:CROSS_COMPILE: -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly
+
+RWStructuredBuffer<float> outputBuffer;
+
+StructuredBuffer<float> buf;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    coopvecNV<int, 8> r_int;
+    coopvecNV<float, 8> r = coopvecNV<float, 8>(r_int);
+    coopvecNV<int, 16> v = coopvecNV<int, 16>(1);
+    int offset = 0;
+    int layout = gl_CooperativeVectorMatrixLayoutRowMajorNV;
+    bool transpose = false;
+    int matrixStride = 4;
+    coopVecMatMulNV(
+        r,
+        v,
+        gl_ComponentTypeFloat32NV,
+        buf,
+        offset,
+        gl_ComponentTypeFloat32NV,
+        8,
+        16,
+        layout,
+        transpose,
+        matrixStride);
+    outputBuffer[dispatchThreadID.x] = r[0];
+}
diff --git a/tests/cooperative-vector/glsl/cast.slang.glsl b/tests/cooperative-vector/glsl/cast.slang.glsl
new file mode 100644
index 000000000..7194c02fc
--- /dev/null
+++ b/tests/cooperative-vector/glsl/cast.slang.glsl
@@ -0,0 +1,23 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 1) readonly buffer StructuredBuffer_float_t_0 {
+    float _data[];
+} buf_0;
+layout(std430, binding = 0) buffer StructuredBuffer_float_t_1 {
+    float _data[];
+} outputBuffer_0;
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    coopvecNV<int, 8 > r_int_0;
+    coopvecNV<float, 8 > _S1 = (coopvecNV<float, 8>((r_int_0)));
+    coopvecNV<float, 8 > r_0 = _S1;
+    coopvecNV<int, 16 > _S2 = (coopvecNV<int, 16>((1)));
+    coopVecMatMulNV((r_0), (_S2), (1), (buf_0)._data, (0U), (1), (8U), (16U), (0), (false), (4U));
+    float _S3 = r_0[0U];
+    outputBuffer_0._data[gl_GlobalInvocationID.x] = _S3;
+    return;
+}
+
diff --git a/tests/cooperative-vector/glsl/groupshared-sized.slang b/tests/cooperative-vector/glsl/groupshared-sized.slang
new file mode 100644
index 000000000..74ccac6a2
--- /dev/null
+++ b/tests/cooperative-vector/glsl/groupshared-sized.slang
@@ -0,0 +1,37 @@
+//DISABLE_TEST:CROSS_COMPILE: -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly
+
+RWStructuredBuffer<float> outputBuffer;
+
+groupshared float buf[100];
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    coopvecNV<float, 8> r;
+    coopvecNV<int, 16> v;
+    if(dispatchThreadID.x == 0)
+    {
+        for(int i = 0; i < 100; ++i)
+        {
+          buf[i] = float(i);
+        }
+    }
+    coopVecLoadNV(v, buf, 0);
+    int offset = 0;
+    int layout = gl_CooperativeVectorMatrixLayoutRowMajorNV;
+    bool transpose = false;
+    int matrixStride = 4;
+    coopVecMatMulNV(
+        r,
+        v,
+        gl_ComponentTypeFloat32NV,
+        buf,
+        offset,
+        gl_ComponentTypeFloat32NV,
+        8,
+        16,
+        layout,
+        transpose,
+        matrixStride);
+    outputBuffer[dispatchThreadID.x] = r[0];
+}
diff --git a/tests/cooperative-vector/glsl/groupshared-sized.slang.glsl b/tests/cooperative-vector/glsl/groupshared-sized.slang.glsl
new file mode 100644
index 000000000..f1f4b5a27
--- /dev/null
+++ b/tests/cooperative-vector/glsl/groupshared-sized.slang.glsl
@@ -0,0 +1,38 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 0) buffer StructuredBuffer_float_t_0 {
+    float _data[];
+} outputBuffer_0;
+shared float  buf_0[100];
+
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    coopvecNV<float, 8 > r_0;
+    coopvecNV<int, 16 > v_0;
+    uint _S1 = gl_GlobalInvocationID.x;
+    if(_S1 == 0U)
+    {
+        int i_0 = 0;
+        for(;;)
+        {
+            if(i_0 < 100)
+            {
+            }
+            else
+            {
+                break;
+            }
+            buf_0[i_0] = float(i_0);
+            i_0 = i_0 + 1;
+        }
+    }
+    coopVecLoadNV((v_0), (buf_0), (0U));
+    coopvecNV<int, 16 > _S2 = v_0;
+    coopVecMatMulNV((r_0), (_S2), (1), (buf_0), (0U), (1), (8U), (16U), (0), (false), (4U));
+    float _S3 = r_0[0U];
+    outputBuffer_0._data[uint(_S1)] = _S3;
+    return;
+}
diff --git a/tests/cooperative-vector/glsl/load-store.slang b/tests/cooperative-vector/glsl/load-store.slang
new file mode 100644
index 000000000..04a0cd585
--- /dev/null
+++ b/tests/cooperative-vector/glsl/load-store.slang
@@ -0,0 +1,13 @@
+//DISABLE_TEST:CROSS_COMPILE: -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly
+
+RWStructuredBuffer<float> rwBuf;
+StructuredBuffer<float> buf;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    coopvecNV<float, 8> r;
+    coopVecLoadNV(r, rwBuf, 0);
+    coopVecLoadNV(r, buf, 1);
+    coopVecStoreNV(r, rwBuf, 2);
+}
diff --git a/tests/cooperative-vector/glsl/load-store.slang.glsl b/tests/cooperative-vector/glsl/load-store.slang.glsl
new file mode 100644
index 000000000..3844e2ff0
--- /dev/null
+++ b/tests/cooperative-vector/glsl/load-store.slang.glsl
@@ -0,0 +1,20 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 0) buffer StructuredBuffer_float_t_0 {
+    float _data[];
+} rwBuf_0;
+layout(std430, binding = 1) readonly buffer StructuredBuffer_float_t_1 {
+    float _data[];
+} buf_0;
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    coopvecNV<float, 8 > r_0;
+    coopVecLoadNV((r_0), (rwBuf_0)._data, (0U));
+    coopVecLoadNV((r_0), (buf_0)._data, (1U));
+    coopVecStoreNV((r_0), (rwBuf_0)._data, (2U));
+    return;
+}
+
diff --git a/tests/cooperative-vector/glsl/only-arith.slang b/tests/cooperative-vector/glsl/only-arith.slang
new file mode 100644
index 000000000..8a6d3d0d9
--- /dev/null
+++ b/tests/cooperative-vector/glsl/only-arith.slang
@@ -0,0 +1,18 @@
+//DISABLE_TEST:CROSS_COMPILE: -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly
+
+// Disabled because we don't output the extension because although the
+// coopvecNV type is used, it isn't *declared* here (and we don't attach
+// requirements to the add operations);
+
+RWStructuredBuffer<float> outputBuffer;
+
+StructuredBuffer<float> buf;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    coopvecNV<float, 8> r;
+    coopvecNV<float, 8> v = coopvecNV<float, 8>(1);
+    r = v + v;
+    outputBuffer[dispatchThreadID.x] = r[0];
+}
diff --git a/tests/cooperative-vector/glsl/only-arith.slang.glsl b/tests/cooperative-vector/glsl/only-arith.slang.glsl
new file mode 100644
index 000000000..db3372d70
--- /dev/null
+++ b/tests/cooperative-vector/glsl/only-arith.slang.glsl
@@ -0,0 +1,16 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 0) buffer _S1 {
+    float _data[];
+} outputBuffer_0;
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    const coopvecNV<float, 8 > v_0 = coopvecNV<float, 8 >(1.0);
+    const coopvecNV<float, 8 > r_0 = v_0 + v_0;
+    float _S2 = r_0[0U];
+    ((outputBuffer_0)._data[(gl_GlobalInvocationID.x)]) = _S2;
+    return;
+}
diff --git a/tests/cooperative-vector/glsl/outer-product-accumulate.slang b/tests/cooperative-vector/glsl/outer-product-accumulate.slang
new file mode 100644
index 000000000..1db30cb0d
--- /dev/null
+++ b/tests/cooperative-vector/glsl/outer-product-accumulate.slang
@@ -0,0 +1,19 @@
+//DISABLE_TEST:CROSS_COMPILE: -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly -emit-spirv-via-glsl
+
+RWByteAddressBuffer buf;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    coopvecNV<float, 8> op0 = coopvecNV<float, 8>(1);
+    coopvecNV<float, 16> op1 = coopvecNV<float, 16>(2);
+
+    coopVecOuterProductAccumulateNV(
+        op0,
+        op1,
+        buf,
+        0, // firstElement
+        4, // matrixStride
+        gl_CooperativeVectorMatrixLayoutRowMajorNV,
+        gl_ComponentTypeFloat32NV);
+}
diff --git a/tests/cooperative-vector/glsl/outer-product-accumulate.slang.glsl b/tests/cooperative-vector/glsl/outer-product-accumulate.slang.glsl
new file mode 100644
index 000000000..09d781824
--- /dev/null
+++ b/tests/cooperative-vector/glsl/outer-product-accumulate.slang.glsl
@@ -0,0 +1,17 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 0) buffer _S1
+{
+    uint _data[];
+} buf_0;
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    coopvecNV<float, 8 > _S2 = (coopvecNV<float, 8>((1.0)));
+    coopvecNV<float, 16 > _S3 = (coopvecNV<float, 16>((2.0)));
+    coopVecOuterProductAccumulateNV((_S2), (_S3), (buf_0)._data, (0U), (4U), (0), (1));
+    return;
+}
+
diff --git a/tests/cooperative-vector/glsl/simple.slang b/tests/cooperative-vector/glsl/simple.slang
new file mode 100644
index 000000000..4128f6827
--- /dev/null
+++ b/tests/cooperative-vector/glsl/simple.slang
@@ -0,0 +1,35 @@
+//DISABLE_TEST:CROSS_COMPILE(filecheck=CHECK): -profile glsl_450+spirv_1_4 -stage compute -entry computeMain -target spirv-assembly -emit-spirv-via-glsl
+
+RWStructuredBuffer<float> outputBuffer;
+
+StructuredBuffer<float> buf;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    // CHECK: %[[VAR1:[a-zA-Z0-9_]+]] = OpTypeCooperativeVectorNV %int %uint_16
+    // CHECK: OpConstantCompositeReplicateEXT %9 %int_1
+    coopvecNV<int, 16> v = coopvecNV<int, 16>(1);
+
+    // CHECK: %[[VAR2:[a-zA-Z0-9_]+]] = OpTypeCooperativeVectorNV %int %uint_8
+    coopvecNV<int, 8> r;
+    int offset = 0;
+    int layout = gl_CooperativeVectorMatrixLayoutRowMajorNV;
+    bool transpose = false;
+    int matrixStride = 4;
+
+    // CHECK: OpCooperativeVectorMatrixMulNV %[[VAR2]] %{{.*}} %int_1 %{{.*}} %uint_0 %int_1 %uint_8 %uint_16 %int_0 %false %uint_4 MatrixBSignedComponentsKHR|MatrixResultSignedComponentsKHR
+    coopVecMatMulNV(
+        r,
+        v,
+        gl_ComponentTypeFloat32NV,
+        buf,
+        offset,
+        gl_ComponentTypeFloat32NV,
+        8,
+        16,
+        layout,
+        transpose,
+        matrixStride);
+    outputBuffer[dispatchThreadID.x] = r[0];
+}
diff --git a/tests/cooperative-vector/glsl/simple.slang.glsl b/tests/cooperative-vector/glsl/simple.slang.glsl
new file mode 100644
index 000000000..cf5f0566e
--- /dev/null
+++ b/tests/cooperative-vector/glsl/simple.slang.glsl
@@ -0,0 +1,20 @@
+#version 450
+#extension GL_NV_cooperative_vector : require
+layout(row_major) uniform;
+layout(row_major) buffer;
+layout(std430, binding = 1) readonly buffer StructuredBuffer_float_t_0 {
+    float _data[];
+} buf_0;
+layout(std430, binding = 0) buffer StructuredBuffer_float_t_1 {
+    float _data[];
+} outputBuffer_0;
+layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    coopvecNV<int, 16 > _S1 = (coopvecNV<int, 16>((1)));
+    coopvecNV<int, 8 > r_0;
+    coopVecMatMulNV((r_0), (_S1), (1), (buf_0)._data, (0U), (1), (8U), (16U), (0), (false), (4U));
+    float _S2 = r_0[0U];
+    outputBuffer_0._data[gl_GlobalInvocationID.x] = _S2;
+    return;
+}