Overhauled reverse-mode control flow handling (#2608)

* Added switch-case support; fixed non-diff parameter transposition

* Made region propagation much more robust. Partial loop unzip implementation

* WIP: Added most loop handling code, and a test. Still untested

* Added CFG Normalization pass + CFG Reversal Pass + Loop Unzipping + most loop transcription

* Add single-iter-loop test.

* proj files

* removed comments

* Update reverse-loop.slang

* Removed out-of-date code

* Disabled IR validation during constructSSA phase of normalizeCFG. constructSSA now reuses sharedBuilder

* Moved normalizeCFG() call to prepareFuncForBackwardDiff()

3 files changed, 146 insertions, 0 deletions

diff --git a/tests/autodiff/reverse-loop.slang b/tests/autodiff/reverse-loop.slang
new file mode 100644
index 000000000..6ae207080
--- /dev/null
+++ b/tests/autodiff/reverse-loop.slang
@@ -0,0 +1,40 @@
+//TEST_IGNORE_FILE:
+//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;
+
+typedef DifferentialPair<float> dpfloat;
+typedef float.Differential dfloat;
+
+[BackwardDifferentiable]
+float test_simple_loop(float y)
+{
+    float t = y;
+
+    for (int i = 0; i < 3; i++)
+    {
+        t = t * t;
+    }
+
+    return t;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    {
+        dpfloat dpa = dpfloat(1.0, 0.0);
+
+        __bwd_diff(test_simple_loop)(dpa, 1.0f);
+        outputBuffer[0] = dpa.d; // Expect: 2.0
+    }
+
+    {
+        dpfloat dpa = dpfloat(0.4, 0.0);
+        
+        __bwd_diff(test_simple_loop)(dpa, 1.0f);
+        outputBuffer[1] = dpa.d; // Expect: 1.0
+    }
+}
diff --git a/tests/autodiff/reverse-single-iter-loop.slang b/tests/autodiff/reverse-single-iter-loop.slang
new file mode 100644
index 000000000..47232147a
--- /dev/null
+++ b/tests/autodiff/reverse-single-iter-loop.slang
@@ -0,0 +1,99 @@
+//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 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+typedef DifferentialPair<float> dpfloat;
+typedef float.Differential dfloat;
+
+[BackwardDifferentiable]
+float test_simple_single_iter_loop(float y)
+{
+    float x = 0;
+
+    for (int i = 0; i < 1; i++)
+    {
+        if (y > 0.6)
+        {
+            x = y * 10.0f;
+            break;
+        }
+        else
+        {
+            x = y * 6.0f;
+            break;
+        }
+    }
+    
+    return x;
+}
+
+[BackwardDifferentiable]
+float test_nested_if_else_single_iter_loop(float y)
+{
+    float x = 0;
+
+    for (int i = 0; i < 1; i++)
+    {
+        if (y > 0.6)
+        {
+            if (y > 0.8)
+            {
+                x = y * 10.0f;
+                break;
+            }
+            else
+            {
+                x = y * 4.0f;
+                break;
+            }
+        }
+        else
+        {
+            x = y * 6.0f;
+            break;
+        }
+    }
+
+    return x;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    {
+        dpfloat dpa = dpfloat(1.0, 0.0);
+
+        __bwd_diff(test_simple_single_iter_loop)(dpa, 1.0f);
+        outputBuffer[0] = dpa.d; // Expect: 10.0
+    }
+
+    {
+        dpfloat dpa = dpfloat(0.4, 0.0);
+        
+        __bwd_diff(test_simple_single_iter_loop)(dpa, 1.0f);
+        outputBuffer[1] = dpa.d; // Expect: 6.0
+    }
+
+    {
+        dpfloat dpa = dpfloat(1.0, 0.0);
+
+        __bwd_diff(test_nested_if_else_single_iter_loop)(dpa, 1.0f);
+        outputBuffer[2] = dpa.d; // Expect: 10.0
+    }
+
+    {
+        dpfloat dpa = dpfloat(0.7, 0.0);
+
+        __bwd_diff(test_nested_if_else_single_iter_loop)(dpa, 1.0f);
+        outputBuffer[3] = dpa.d; // Expect: 4.0
+    }
+
+    {
+        dpfloat dpa = dpfloat(0.4, 0.0);
+
+        __bwd_diff(test_nested_if_else_single_iter_loop)(dpa, 1.0f);
+        outputBuffer[4] = dpa.d; // Expect: 6.0
+    }
+}
diff --git a/tests/autodiff/reverse-single-iter-loop.slang.expected.txt b/tests/autodiff/reverse-single-iter-loop.slang.expected.txt
new file mode 100644
index 000000000..229f05e65
--- /dev/null
+++ b/tests/autodiff/reverse-single-iter-loop.slang.expected.txt
@@ -0,0 +1,7 @@
+type: float
+10.000000
+6.000000
+10.000000
+4.000000
+6.000000
+0.000000
\ No newline at end of file