More type support for reverse-mode (#2551)

* Add vector arithmetic test. Make gradient accumulation work for any IRLoad

* Added support for general vector types, and split transposition into transpose & materialize to allow emitting the fully accumulated gradient for complex types.

* Several bug fixes + finished up support for vector & struct types + removed prop pass

* minor fixes (int/uint casts)

* Removed IRConstruct

* Added some type casts to prevent warnings

* minor fix for unused variable

1 files changed, 62 insertions, 0 deletions

diff --git a/tests/autodiff/reverse-struct-types.slang b/tests/autodiff/reverse-struct-types.slang
new file mode 100644
index 000000000..699e50480
--- /dev/null
+++ b/tests/autodiff/reverse-struct-types.slang
@@ -0,0 +1,62 @@
+
+//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:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+struct A : IDifferentiable
+{
+    float x;
+    float y;
+
+    [__unsafeForceInlineEarly]
+    static Differential dzero()
+    {
+        Differential b = {0.0, float.dzero()};
+        return b;
+    }
+
+    [__unsafeForceInlineEarly]
+    static Differential dadd(Differential a, Differential b)
+    {
+        Differential o = {a.x + b.x, 0.0};
+        return o;
+    }
+
+    [__unsafeForceInlineEarly]
+    static Differential dmul(This a, Differential b)
+    {
+        Differential o = {a.x * b.x, 0.0};
+        return o;
+    }
+};
+
+typedef DifferentialPair<A> dpA;
+
+[BackwardDifferentiable]
+A f(A a)
+{
+    A aout;
+    aout.y = 2 * a.x;
+    aout.x = 5 * a.x;
+
+    return aout;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    {
+        A a = {1.0, 2.0};
+        A.Differential azero = {0.0, 0.0};
+
+        dpA dpa = dpA(a, azero);
+
+        A.Differential dout = {1.0, 1.0};
+        
+        __bwd_diff(f)(dpa, dout);
+        outputBuffer[0] = dpa.d.x; // Expect: 10
+        outputBuffer[1] = dpa.d.y; // Expect: 0
+    }
+}