Initial copy elision pass regression fix (#8126)

when we have `GetElementPtr -> load -> GetElement` in our use-chain, the
final `GetElement` may use the `load` as a `Index`, not a base.

This is a non-issue with `getFieldExtract` since a field is a StructKey.
We will still add this check to ensure no bugs down the line.

---------

Co-authored-by: Harsh Aggarwal <haaggarwal@nvidia.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: slangbot <ellieh+slangbot@nvidia.com>
Co-authored-by: slangbot <186143334+slangbot@users.noreply.github.com>

1 files changed, 30 insertions, 0 deletions

diff --git a/tests/bugs/gh-8121.slang b/tests/bugs/gh-8121.slang
new file mode 100644
index 000000000..777e75b9e
--- /dev/null
+++ b/tests/bugs/gh-8121.slang
@@ -0,0 +1,30 @@
+//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHECK):-cuda -compute -shaderobj
+
+//TEST_INPUT:ubuffer(data=[1 1 1 1], stride=4),name=inputBuffer
+RWStructuredBuffer<int[2]> inputBuffer;
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<int> outputBuffer;
+
+void math(int data[2])
+{
+    // This test is to ensure that we do not break our optimization-pass which optimizes the IR of
+    // `getElement(load(getElementPtr(...)), ...)` into `load(getElementPtr(getElementPtr(...), ...))`.
+    //
+    // This test triggers the bug given the IR: `store(outputBuffer[0], getElement(Array<>, load(getElementPtr(data,0))))`.
+    // `Array<int,2>(5,6)` is not created as a local variable intentionally. We do this because we want to test the case
+    // where `Array<>` must be accessed as a value (`getElement`), not a pointer (`getElementPtr`).
+    //
+    // If this code fails, outputBuffer[0] will contain the value `1`. We expect `6` if the bug is fixed.
+    outputBuffer[0] = Array<int,2>(5,6)[data[0]];
+}
+
+[shader("compute")]
+[numthreads(1, 1, 1)]
+void computeMain(int3 id: SV_DispatchThreadID)
+{
+    math(inputBuffer[0]);
+}
+
+//CHECK: 6
+//CHECK-NEXT: 0