Fix issue with nested loop unrolling (#3100)

* Do not eliminate single-iter-loops that have inner loops using their break label.

* Add test

* Delete out-old.hlsl

* Update slang-ir-autodiff-cfg-norm.cpp

* Fix whitespace

2 files changed, 49 insertions, 0 deletions

diff --git a/tests/autodiff/nested-loop-unroll.slang b/tests/autodiff/nested-loop-unroll.slang
new file mode 100644
index 000000000..026764f8f
--- /dev/null
+++ b/tests/autodiff/nested-loop-unroll.slang
@@ -0,0 +1,44 @@
+//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], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+static const uint levels = 8;
+
+[Differentiable]
+void eval(float3 p, out float4 output[levels])
+{
+    [ForceUnroll] for (int level = 0; level < 3; ++level) 
+    {
+        float4 f = 0.f;
+
+        // tri-linear time!
+        [ForceUnroll] for (int z = 0; z < 2; ++z)
+        {
+            float wx = 0;
+            if (z != 0)
+                wx = p.x;
+            else
+                wx = p.y;
+
+            f += wx;
+        }
+
+        output[level] = f;
+    }
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID: SV_DispatchThreadID)
+{
+    float3 p = float3(2.0, 3.0, 0);
+
+    float4 output[levels];
+    eval(p, output);
+    DifferentialPair<float3> dp = DifferentialPair<float3>(p, 0);
+    __bwd_diff(eval)(dp, output);
+
+    // Write output
+    outputBuffer[0] = dp.d.x;
+}
diff --git a/tests/autodiff/nested-loop-unroll.slang.expected.txt b/tests/autodiff/nested-loop-unroll.slang.expected.txt
new file mode 100644
index 000000000..c34bf7c0d
--- /dev/null
+++ b/tests/autodiff/nested-loop-unroll.slang.expected.txt
@@ -0,0 +1,5 @@
+type: float
+60.000000
+0.000000
+0.000000
+0.000000