Implement MapElement for CoopMat (#7159)

With this PR, MapElement works for the following signatures:
 - CoopMat<...>::MapElement(functype(...));
 - CoopMat<...>::MapElement(capturing-lambda);
 - CoopMat<...>::MapElement(not-capturing-lambda);
 - Tuple<CoopMat<...>,...>::MapElement(functype(...));
 - Tuple<CoopMat<...>,...>::MapElement(capturing-lambda);
 - Tuple<CoopMat<...>,...>::MapElement(not-capturing-lambda);

4 files changed, 157 insertions, 13 deletions

diff --git a/tests/cooperative-matrix/map-element-single.slang b/tests/cooperative-matrix/map-element-single.slang
index 1661ee105..ecf35953e 100644
--- a/tests/cooperative-matrix/map-element-single.slang
+++ b/tests/cooperative-matrix/map-element-single.slang
@@ -1,12 +1,14 @@
-//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -render-feature cooperative-matrix-per-element-operations -Xslang -DTEST_MODE=0
-//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -render-feature cooperative-matrix-per-element-operations -Xslang -DTEST_MODE=1
-//DISABLE_TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -render-feature cooperative-matrix-per-element-operations -Xslang -DTEST_MODE=2
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=0 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=1 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=2 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=3 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=4 -render-feature cooperative-matrix-per-element-operations
 
 //CHECK: type: int32_t
-//CHECK-NEXT: 2
-//CHECK-NEXT: 4
-//CHECK-NEXT: 6
 //CHECK-NEXT: 8
+//CHECK-NEXT: 10
+//CHECK-NEXT: 12
+//CHECK-NEXT: 14
 
 //TEST_INPUT:ubuffer(data=[1 2 3 4], stride=4),name=input1
 StructuredBuffer<int> input1;
@@ -20,7 +22,7 @@ typealias CoopMatType = CoopMat<int, MemoryScope.Subgroup, 16, 16, CoopMatMatrix
 
 int MapOp(uint32_t row, uint32_t col, int value)
 {
-    return value * 2;
+    return value * 2 + 1 + 2 + 3;
 }
 
 [numthreads(32, 1, 1)]
@@ -29,21 +31,33 @@ void computeMain()
     let stride = 16;
     CoopMatType mat1 = CoopMatType.Load<CoopMatMatrixLayout.RowMajor>(input1, 0, stride);
 
+    // Testing the capturing lambda
+    int c0 = 1;
+    int c1 = 2;
+    int c2 = 3;
+
     CoopMatType result;
 
 #if TEST_MODE == 0
     result = mat1.MapElement(MapOp);
 
 #elif TEST_MODE == 1
-    // Lambda through IFunc.
-    // TODO: Not working due to issue #7024
-    IFunc<int, uint32_t, uint32_t, int> func = ((uint32_t row, uint32_t column, int value) => value * 2);
+    // Lambda via a temp variable (no capture)
+    let func = ((uint32_t row, uint32_t column, int value) => value * 2 + 1 + 2 + 3);
     result = mat1.MapElement(func);
 
 #elif TEST_MODE == 2
-    // Directly use lambda.
-    // TODO: Not working due to issue #7024
-    result = mat1.MapElement((uint32_t row, uint32_t column, int value) => (int)(value));
+    // Directly use lambda (no capture)
+    result = mat1.MapElement((uint32_t row, uint32_t column, int value) => value * 2 + 1 + 2 + 3);
+
+#elif TEST_MODE == 3
+    // Lambda via a temp variable (capture)
+    let func = ((uint32_t row, uint32_t column, int value) => value * 2 + c0 + c1 + c2);
+    result = mat1.MapElement(func);
+
+#elif TEST_MODE == 4
+    // Directly use lambda (capture)
+    result = mat1.MapElement((uint32_t row, uint32_t column, int value) => value * 2 + c0 + c1 + c2);
 #endif
 
     result.Store<CoopMatMatrixLayout.RowMajor>(outputBuffer, 0, stride);
diff --git a/tests/cooperative-matrix/map-element-tuple.slang b/tests/cooperative-matrix/map-element-tuple.slang
new file mode 100644
index 000000000..06ab99d8f
--- /dev/null
+++ b/tests/cooperative-matrix/map-element-tuple.slang
@@ -0,0 +1,68 @@
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=0 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=1 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=2 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=3 -render-feature cooperative-matrix-per-element-operations
+//TEST(compute):COMPARE_COMPUTE(filecheck-buffer=CHECK):-vk -output-using-type -emit-spirv-directly -Xslang -DTEST_MODE=4 -render-feature cooperative-matrix-per-element-operations
+
+//CHECK:type: int32_t
+//CHECK-NEXT:9
+//CHECK-NEXT:12
+//CHECK-NEXT:15
+//CHECK-NEXT:14
+
+//TEST_INPUT:ubuffer(data=[1 2 3 4], stride=4),name=input1
+StructuredBuffer<int> input1;
+
+//TEST_INPUT:ubuffer(data=[0 1 2 3], stride=4),name=input2
+StructuredBuffer<int> input2;
+
+//TEST_INPUT:ubuffer(data=[2 3 4 1], stride=4),name=input3
+StructuredBuffer<int> input3;
+
+//TEST_INPUT:ubuffer(stride=4, count=256):out,name=outputBuffer
+RWStructuredBuffer<int32_t> outputBuffer;
+
+using namespace linalg;
+
+typealias CoopMatType = CoopMat<int, MemoryScope.Subgroup, 16, 16, CoopMatMatrixUse.MatrixAccumulator>;
+
+int MapOp(uint32_t row, uint32_t col, int a, int b, int c)
+{
+    return a + b + c + 1 + 2 + 3;
+}
+
+[numthreads(32, 1, 1)]
+void computeMain()
+{
+    let stride = 16;
+    let mat1 = CoopMatType.Load<CoopMatMatrixLayout.RowMajor>(input1, 0, stride);
+    let mat2 = CoopMatType.Load<CoopMatMatrixLayout.RowMajor>(input2, 0, stride);
+    let mat3 = CoopMatType.Load<CoopMatMatrixLayout.RowMajor>(input3, 0, stride);
+
+    // Testing the capturing lambda
+    int c0 = 1;
+    int c1 = 2;
+    int c2 = 3;
+
+    CoopMatType result;
+
+#if TEST_MODE == 0
+    result = makeTuple(mat1, mat2, mat3).MapElement(MapOp);
+
+#elif TEST_MODE == 1
+    let f = ((uint32_t x, uint32_t y, int a, int b, int c) => a + b + c + 1 + 2 + 3);
+    result = makeTuple(mat1, mat2, mat3).MapElement(f);
+
+#elif TEST_MODE == 2
+    result = makeTuple(mat1, mat2, mat3).MapElement((uint32_t x, uint32_t y, int a, int b, int c) => a + b + c + 1 + 2 + 3);
+
+#elif TEST_MODE == 3
+    let f = ((uint32_t x, uint32_t y, int a, int b, int c) => a + b + c + c0 + c1 + c2);
+    result = makeTuple(mat1, mat2, mat3).MapElement(f);
+
+#elif TEST_MODE == 4
+    result = makeTuple(mat1, mat2, mat3).MapElement((uint32_t x, uint32_t y, int a, int b, int c) => a + b + c + c0 + c1 + c2);
+#endif
+
+    result.Store<CoopMatMatrixLayout.RowMajor>(outputBuffer, 0, stride);
+}
diff --git a/tests/language-feature/tuple/tuple-expand-multiple.slang b/tests/language-feature/tuple/tuple-expand-multiple.slang
new file mode 100644
index 000000000..9392e2d66
--- /dev/null
+++ b/tests/language-feature/tuple/tuple-expand-multiple.slang
@@ -0,0 +1,37 @@
+//TEST:COMPARE_COMPUTE(filecheck-buffer=CHECK): -output-using-type
+
+//TEST_INPUT: set outputBuffer = out ubuffer(data=[0 0 0 0 0 0], stride=4)
+RWStructuredBuffer<int> outputBuffer;
+
+extension<
+    T0 : __BuiltinArithmeticType,
+    each Ts0 : __BuiltinArithmeticType,
+    each Ts1 : __BuiltinArithmeticType
+> Tuple<T0, T0, expand each Ts0, expand each Ts1>
+{
+    static int getSize_Ts0() { return countof(Ts0); }
+    static int getSize_Ts1() { return countof(Ts1); }
+}
+
+[numthreads(1,1,1)]
+void computeMain()
+{
+    int i = 2;
+    float f0 = 3, f1 = 5;
+    uint ui0 = 4, ui1 = 6;
+
+    let s0 = makeTuple(i, i); // T, T
+    let s1 = makeTuple(i, i, f0, ui0); // T, T, Ts0, Ts1
+    let s2 = makeTuple(i, i, f0, ui0, f1, ui1); // T, T, Ts0, Ts0, Ts1, Ts1
+
+    outputBuffer[0] = s0.getSize_Ts0();
+    outputBuffer[1] = s0.getSize_Ts1();
+    outputBuffer[2] = s1.getSize_Ts0();
+    outputBuffer[3] = s1.getSize_Ts1();
+    outputBuffer[4] = s2.getSize_Ts0();
+    outputBuffer[5] = s2.getSize_Ts1();
+
+    // CHECK-COUNT-2:0
+    // CHECK-COUNT-2:1
+    // CHECK-COUNT-2:2
+}
diff --git a/tests/language-feature/tuple/tuple-expand.slang b/tests/language-feature/tuple/tuple-expand.slang
new file mode 100644
index 000000000..c9ea26161
--- /dev/null
+++ b/tests/language-feature/tuple/tuple-expand.slang
@@ -0,0 +1,25 @@
+//TEST:COMPARE_COMPUTE(filecheck-buffer=CHECK): -output-using-type
+
+//TEST_INPUT: set outputBuffer = out ubuffer(data=[0 0], stride=4)
+RWStructuredBuffer<int> outputBuffer;
+
+struct X<B, each T, each U>
+{
+    int getTSize() { return countof(T); }
+    int getUSize() { return countof(U); }
+}
+
+func foo<each T, each U>() -> X<bool, expand Ptr<each T>, int, expand Ptr<each U>, float> // unify
+{
+    return {};
+}
+
+[numthreads(1,1,1)]
+void computeMain()
+{
+    let x = foo<int, float>();
+
+    outputBuffer[0] = x.getTSize();
+    outputBuffer[1] = x.getUSize();
+    // CHECK-COUNT-2: 2
+}