FP16 atomics for RWByteAddresBuffer, fp32 atomics for images. (#3536)

* FP16 atomics for RWByteAddresBuffer, fp32 atomics for images.

* Fix spelling.

* Add overload.

* Fix test failures.

---------

Co-authored-by: Yong He <yhe@nvidia.com>

1 files changed, 34 insertions, 0 deletions

diff --git a/tests/hlsl-intrinsic/byte-address-buffer/byte-address-half-atomics.slang b/tests/hlsl-intrinsic/byte-address-buffer/byte-address-half-atomics.slang
new file mode 100644
index 000000000..a89c7c3fa
--- /dev/null
+++ b/tests/hlsl-intrinsic/byte-address-buffer/byte-address-half-atomics.slang
@@ -0,0 +1,34 @@
+// byte-address-half-atomics.slang
+// test the atomics on half types.
+
+// Disabled because our current driver doesn't support half atomics yet.
+//DISABLED_TEST(compute, vulkan):COMPARE_COMPUTE(filecheck-buffer=CHECK): -vk -compute -profile cs_6_2 -render-features half -shaderobj
+
+//TEST:SIMPLE(filecheck=SPIRV):-target spirv -entry computeMain -stage compute
+//TEST:SIMPLE(filecheck=SPIRV):-target spirv -entry computeMain -stage compute -emit-spirv-directly
+
+//TEST_INPUT:set tmpBuffer = ubuffer(data=[0 0 0 0], stride=4)
+RWByteAddressBuffer tmpBuffer;
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+[numthreads(1, 1, 1)]
+void computeMain(int3 dispatchThreadID : SV_DispatchThreadID)
+{
+    half originalValue;
+    
+    // SPIRV: OpAtomicFAddEXT
+
+    tmpBuffer.InterlockedAddF16(0, 1.0h, originalValue);
+    tmpBuffer.InterlockedAddF16(2, 2.0h, originalValue);
+    
+    half2 v = tmpBuffer.Load<half2>(0);
+    
+    // CHECK: 1.0
+    outputBuffer[0] = v.x;
+    // CHECK: 2.0
+    outputBuffer[1] = v.y;
+    // CHECK: 0.0
+    outputBuffer[3] = originalValue;
+}