CUDA Wave intrinsic vector/matrix support (#1267)

* Distinguish between __activeMask and _getConvergedMask().
Remove need to pass in mask to CUDA wave impls.

* Add support for vector/matrix Wave intrinsics for CUDA.
Fix issue with CUDA parsing of errors.

* Fix typo.
Make WaveReadLineAt and WaveReadFirst work for vector/matrix types.

* Fix typo.

* Added equality wave intrinsic test.

* Fix some typos

* Added wave-lane-at.slang

4 files changed, 80 insertions, 0 deletions

diff --git a/tests/hlsl-intrinsic/wave-equality.slang b/tests/hlsl-intrinsic/wave-equality.slang
new file mode 100644
index 000000000..d12d8cfbc
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-equality.slang
@@ -0,0 +1,31 @@
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<int> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = int(dispatchThreadID.x);
+    
+    int value = 0;
+    
+    // Scalar
+    
+    value |= WaveActiveAllEqual(idx * 0 + 1)  ? 1 : 0;     // true
+    value |= WaveActiveAllEqual(idx & 2)      ? 2 : 0;     // false
+   
+    // Vector
+   
+    int2 v0 = int2(idx & 0xf0, (idx & 0xf00) + 1);      // (0, 1)
+    int2 v1 = int2(idx & 2, (idx & 2) + 1);
+    
+    value |= WaveActiveAllEqual(v0)           ? 0x10 : 0;     // true
+    value |= WaveActiveAllEqual(v1)           ? 0x20 : 0;     // false
+    
+    outputBuffer[idx] = value;
+}
\ No newline at end of file
diff --git a/tests/hlsl-intrinsic/wave-equality.slang.expected.txt b/tests/hlsl-intrinsic/wave-equality.slang.expected.txt
new file mode 100644
index 000000000..2bf571888
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-equality.slang.expected.txt
@@ -0,0 +1,4 @@
+11
+11
+11
+11
diff --git a/tests/hlsl-intrinsic/wave-lane-at.slang b/tests/hlsl-intrinsic/wave-lane-at.slang
new file mode 100644
index 000000000..ca05d985d
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-lane-at.slang
@@ -0,0 +1,41 @@
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+// Disabled because on GLSL targets the lane index *must* be a const expr - and in this test it is not.
+//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<int> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = int(dispatchThreadID.x);
+    
+    int value = 0;
+    
+    // Scalar
+    
+    value += WaveReadLaneAt(idx, (idx + 1) & 3);
+    
+    // vector
+    
+    {
+        float2 v = float2(idx + 1, idx + 2);
+        float2 readValue = WaveReadLaneAt(v, (idx - 1) & 3);
+        
+        value += int(readValue[0] + readValue[1]);
+    }
+        
+    // matrix
+    {
+        matrix<int, 2, 2> v = matrix<int, 2, 2>(idx, idx - 1, idx * 3, idx - 2);
+        
+        matrix<int, 2, 2> readValue = WaveReadLaneAt(v, (idx - 1) & 3);
+        
+        value += int(readValue[0][0] + readValue[0][1] + readValue[1][0] + readValue[1][1]);
+    }
+    
+    outputBuffer[idx] = value;
+}
\ No newline at end of file
diff --git a/tests/hlsl-intrinsic/wave-lane-at.slang.expected.txt b/tests/hlsl-intrinsic/wave-lane-at.slang.expected.txt
new file mode 100644
index 000000000..a327b0804
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-lane-at.slang.expected.txt
@@ -0,0 +1,4 @@
+19
+2
+B
+10