Various fixes in GLSL emit. (#3074)

* Fix name mangling of modified types.

* Add `InterlockedAdd(__ref uint, int)` overload.

* Fix.

* Fix type error in ImageStore legalization.

---------

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

6 files changed, 59 insertions, 11 deletions

diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang
index 1ab046b19..56ad32326 100644
--- a/source/slang/hlsl.meta.slang
+++ b/source/slang/hlsl.meta.slang
@@ -2366,6 +2366,10 @@ __target_intrinsic(glsl, "$atomicAdd($A, $1)")
 __target_intrinsic(cuda, "atomicAdd((uint*)$0, $1)")
 void InterlockedAdd(__ref uint dest, uint value);
 
+__target_intrinsic(glsl, "$atomicAdd($A, $1)")
+__target_intrinsic(cuda, "atomicAdd((uint*)$0, $1)")
+void InterlockedAdd(__ref uint dest, int value);
+
 __target_intrinsic(glsl, "($2 = $atomicAdd($A, $1))")
 __target_intrinsic(cuda, "(*$2 = atomicAdd($0, $1))")
 void InterlockedAdd(__ref  int dest,  int value, out  int original_value);
@@ -7770,4 +7774,4 @@ ${{{{
 
 ${{{{
 }
-}}}}
\ No newline at end of file
+}}}}
diff --git a/source/slang/slang-ir-glsl-legalize.cpp b/source/slang/slang-ir-glsl-legalize.cpp
index dfe370d2b..a5bb299c5 100644
--- a/source/slang/slang-ir-glsl-legalize.cpp
+++ b/source/slang/slang-ir-glsl-legalize.cpp
@@ -79,26 +79,20 @@ void legalizeImageSubscriptStoreForGLSL(IRBuilder& builder, IRInst* storeInst)
             // Here we assume the imageElementType is already lowered into float4/uint4 types from any
             // user-defined type.
             assert(imageElementType->getOp() == kIROp_VectorType);
-            auto originalValue = builder.emitImageLoad(imageElementType, imageSubscript->getImage(), legalizedCoord);
+            auto vectorBaseType = getIRVectorBaseType(imageElementType);
+            IRType* vector4Type = builder.getVectorType(vectorBaseType, 4);
+            auto originalValue = builder.emitImageLoad(vector4Type, imageSubscript->getImage(), legalizedCoord);
             Array<IRInst*, 4> indices;
             for (UInt i = 0; i < swizzledStore->getElementCount(); i++)
             {
                 indices.add(swizzledStore->getElementIndex(i));
             }
             auto newValue = builder.emitSwizzleSet(
-                imageElementType,
+                vector4Type,
                 originalValue,
                 swizzledStore->getSource(),
                 swizzledStore->getElementCount(),
                 indices.getBuffer());
-            if (getIRVectorElementSize(imageElementType) != 4)
-            {
-                auto vectorBaseType = getIRVectorBaseType(imageElementType);
-                newValue = builder.emitVectorReshape(
-                    builder.getVectorType(
-                        vectorBaseType, builder.getIntValue(builder.getIntType(), 4)),
-                    newValue);
-            }
             auto imageStore = builder.emitImageStore(
                 builder.getVoidType(), imageSubscript->getImage(), legalizedCoord, newValue);
             storeInst->replaceUsesWith(imageStore);
diff --git a/source/slang/slang-mangle.cpp b/source/slang/slang-mangle.cpp
index b27a45484..6a3a17caa 100644
--- a/source/slang/slang-mangle.cpp
+++ b/source/slang/slang-mangle.cpp
@@ -251,6 +251,15 @@ namespace Slang
             for(Index i = 0; i < n; ++i)
                 emitType(context, tupleType->getMember(i));
         }
+        else if (auto modifiedType = dynamicCast<ModifiedType>(type))
+        {
+            emitRaw(context, "Tm");
+            emitType(context, modifiedType->getBase());
+            auto n = modifiedType->getModifierCount();
+            emit(context, n);
+            for (Index i = 0; i < n; ++i)
+                emitVal(context, modifiedType->getModifier(i));
+        }
         else
         {
             SLANG_UNEXPECTED("unimplemented case in type mangling");
@@ -336,6 +345,10 @@ namespace Slang
             emitVal(context, typecastIntVal->getType());
             emitVal(context, typecastIntVal->getBase());
         }
+        else if (auto modifier = as<ModifierVal>(val))
+        {
+            emitNameImpl(context, UnownedStringSlice(modifier->getClassInfo().m_name));
+        }
         else
         {
             SLANG_UNEXPECTED("unimplemented case in val mangling");
diff --git a/tests/bugs/buffer-swizzle-store.slang b/tests/bugs/buffer-swizzle-store.slang
new file mode 100644
index 000000000..ca09cded9
--- /dev/null
+++ b/tests/bugs/buffer-swizzle-store.slang
@@ -0,0 +1,16 @@
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
+
+//TEST_INPUT: RWTexture2D(format=R16G16_FLOAT, size=4, content = one, mipMaps = 1):name g_test
+[format("rg16f")]
+RWTexture2D<float2> g_test;
+
+//TEST_INPUT:ubuffer(data=[0], stride=4):out,name outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+[numthreads(1,1,1)]
+void computeMain( uint2 dispatchThreadID : SV_DispatchThreadID )
+{
+    g_test[dispatchThreadID].xy = float2(0.0, 1.0);
+    outputBuffer[dispatchThreadID.x] = g_test[dispatchThreadID].y;
+}
\ No newline at end of file
diff --git a/tests/bugs/buffer-swizzle-store.slang.expected.txt b/tests/bugs/buffer-swizzle-store.slang.expected.txt
new file mode 100644
index 000000000..eddb06a2a
--- /dev/null
+++ b/tests/bugs/buffer-swizzle-store.slang.expected.txt
@@ -0,0 +1,2 @@
+type: float
+1.0
\ No newline at end of file
diff --git a/tests/bugs/interlocked-add-uint-int.slang b/tests/bugs/interlocked-add-uint-int.slang
new file mode 100644
index 000000000..10baa7f17
--- /dev/null
+++ b/tests/bugs/interlocked-add-uint-int.slang
@@ -0,0 +1,19 @@
+//TEST:SIMPLE(filecheck=CHECK): -target spirv -profile glsl_450 -stage compute -entry MainCs -line-directive-mode none
+//CHECK: {{.*}} AtomicIAdd
+RWBuffer<uint> g_InterlockTest;
+
+[numthreads(1,1,1)]
+void MainCs( uint2 dispatchThreadID : SV_DispatchThreadID, uint2 groupThreadID : SV_GroupThreadID )
+{
+	uint nVertexCount = 41;
+	uint nVertexBufferSize = 40;
+	uint nVertex = 3;
+
+	InterlockedAdd( g_InterlockTest[ 0 ], nVertexCount, nVertex );
+
+	// Did we overflow?
+	if ( nVertex + nVertexCount > nVertexBufferSize )
+	{
+		InterlockedAdd( g_InterlockTest[ 0 ], -int(nVertexCount) );
+	}
+}
\ No newline at end of file