Overhaul `transposeParameterBlock` to support `inout` params. (#2621)

* Overhaul `transposeParameterBlock` to support `inout` params.

* Small bug fixes.

* Bug fix on differentiable intrinsic specialization.

* Fixes.

* Run autodiff tests on CPU.

* Clean up.

* More bug fixes.,

* Add test coverage on inout param.

* Fix language server hinting for transcribed mutable params.

---------

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

1 files changed, 31 insertions, 0 deletions

diff --git a/tests/autodiff/reverse-inout-param-0.slang b/tests/autodiff/reverse-inout-param-0.slang
new file mode 100644
index 000000000..13bfb6277
--- /dev/null
+++ b/tests/autodiff/reverse-inout-param-0.slang
@@ -0,0 +1,31 @@
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
+//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -output-using-type -shaderobj
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+[BackwardDifferentiable]
+void g(float x, out float y)
+{
+    y = (x + 1) * (x + 1);
+}
+
+[BackwardDifferentiable]
+void f(float x, out float y)
+{
+    g(x, y);
+    g(x, y);
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    var x = diffPair(5.0);
+    float yDiffOut = 1.0;
+
+    __bwd_diff(f)(x, yDiffOut);
+
+    outputBuffer[0] = x.p; // should be 5, since bwd_diff does not write back new primal val.
+    outputBuffer[1] = x.d; // 12.0
+}
\ No newline at end of file