5 files changed, 88 insertions, 2 deletions
diff --git a/prelude/slang-cpp-types.h b/prelude/slang-cpp-types.h
index c8f6357a2..67db607f6 100644
--- a/prelude/slang-cpp-types.h
+++ b/prelude/slang-cpp-types.h
@@ -85,6 +85,10 @@ struct Matrix
     Vector<T, COLS> rows[ROWS];
 };
 
+// We can just map `NonUniformResourceIndex` type directly to the index type on CPU, as CPU does not require
+// any special handling around such accesses.
+typedef size_t NonUniformResourceIndex;
+
 // ----------------------------- ResourceType -----------------------------------------
 
 // https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/sm5-object-structuredbuffer-getdimensions
diff --git a/tests/compute/unbounded-array-of-array-syntax.slang b/tests/compute/unbounded-array-of-array-syntax.slang
index fdc04c95e..df1074896 100644
--- a/tests/compute/unbounded-array-of-array-syntax.slang
+++ b/tests/compute/unbounded-array-of-array-syntax.slang
@@ -1,5 +1,7 @@
 //DISABLE_TEST:CPU_REFLECTION: -profile cs_5_0 -entry computeMain -target cpp 
 //TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
+//TEST:CROSS_COMPILE:-target dxbc-assembly -entry computeMain -profile cs_5_1
+//TEST:CROSS_COMPILE:-target spirv-assembly -entry computeMain -profile cs_5_1
 
 //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0], stride=4):dxbinding(0),glbinding(0),out,name outputBuffer
 RWStructuredBuffer<int> outputBuffer;
@@ -18,7 +20,7 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     int baseIndex = index >> 2;
     int innerIndex = index & 3;
 
-    RWStructuredBuffer<int> buffer = g_aoa[baseIndex];
+    RWStructuredBuffer<int> buffer = g_aoa[NonUniformResourceIndex(baseIndex)];
 
     // Get the size 
     uint bufferCount, bufferStride;
diff --git a/tests/compute/unbounded-array-of-array-syntax.slang.glsl b/tests/compute/unbounded-array-of-array-syntax.slang.glsl
new file mode 100644
index 000000000..675d00718
--- /dev/null
+++ b/tests/compute/unbounded-array-of-array-syntax.slang.glsl
@@ -0,0 +1,40 @@
+//TEST_IGNORE_FILE:
+
+#version 450
+layout(row_major) uniform;
+layout(row_major) buffer;
+#extension GL_EXT_nonuniform_qualifier : require
+
+layout(std430, binding = 1) buffer _S1 {
+    int _data[];
+} g_aoa_0[];
+layout(std430, binding = 0) buffer _S2 {
+    int _data[];
+} outputBuffer_0;
+
+layout(local_size_x = 8, local_size_y = 1, local_size_z = 1) in;
+void main()
+{
+    int innerIndex_0;
+
+    int index_0 = int(gl_GlobalInvocationID.x);
+
+    int innerIndex_1 = index_0 & 3;
+
+    uint _S3;
+    uint _S4;
+    (_S3) = (g_aoa_0[nonuniformEXT(index_0 >> 2)])._data.length(); (_S4) = 0;
+    uint bufferCount_0 = _S3;
+
+    if(uint(innerIndex_1) >= bufferCount_0)
+    {
+        innerIndex_0 = int(bufferCount_0 - uint(1));
+    }
+    else
+    {
+        innerIndex_0 = innerIndex_1;
+    }
+    uint _S5 = uint(innerIndex_0);
+    ((outputBuffer_0)._data[(uint(index_0))]) = ((g_aoa_0[nonuniformEXT(index_0 >> 2)])._data[(_S5)]);
+    return;
+}
diff --git a/tests/compute/unbounded-array-of-array-syntax.slang.hlsl b/tests/compute/unbounded-array-of-array-syntax.slang.hlsl
new file mode 100644
index 000000000..3fd9ee8ea
--- /dev/null
+++ b/tests/compute/unbounded-array-of-array-syntax.slang.hlsl
@@ -0,0 +1,40 @@
+//TEST_IGNORE_FILE:
+
+#pragma pack_matrix(column_major)
+
+RWStructuredBuffer<int >  g_aoa_0[] : register(u0, space1);
+
+RWStructuredBuffer<int > outputBuffer_0 : register(u0);
+
+[numthreads(8, 1, 1)]
+void computeMain(vector<uint,3> dispatchThreadID_0 : SV_DISPATCHTHREADID)
+{
+    int innerIndex_0;
+
+    int index_0 = (int) dispatchThreadID_0.x;
+
+    int innerIndex_1 = index_0 & 3;
+
+    RWStructuredBuffer<int > buffer_0 = g_aoa_0[NonUniformResourceIndex(index_0 >> 2)];
+
+    uint _S1;
+    uint _S2;
+
+    buffer_0.GetDimensions(_S1, _S2);
+
+    uint bufferCount_0 = _S1;
+
+    if((uint) innerIndex_1 >= bufferCount_0)
+    {
+        innerIndex_0 = (int) (bufferCount_0 - (uint) 1);
+    }
+    else
+    {
+        innerIndex_0 = innerIndex_1;
+    }
+
+    uint _S3 = (uint) innerIndex_0;
+    outputBuffer_0[(uint) index_0] = buffer_0[_S3];
+    return;
+}
+
diff --git a/tests/compute/unbounded-array-of-array.slang b/tests/compute/unbounded-array-of-array.slang
index 760740282..079808710 100644
--- a/tests/compute/unbounded-array-of-array.slang
+++ b/tests/compute/unbounded-array-of-array.slang
@@ -20,7 +20,7 @@ void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
     int baseIndex = index >> 2;
     int innerIndex = index & 3;
 
-    RWStructuredBuffer<int> buffer = g_aoa.array[baseIndex];
+    RWStructuredBuffer<int> buffer = g_aoa.array[NonUniformResourceIndex(baseIndex)];
 
     // Get the size 
     uint bufferCount, bufferStride;