From 24c9611a8296395dfad23c6ae136fab585dff735 Mon Sep 17 00:00:00 2001
From: Sai Praveen Bangaru <31557731+saipraveenb25@users.noreply.github.com>
Date: Tue, 18 Mar 2025 03:28:59 -0700
Subject: Fix `IRVar` hoisting when its already in the right block. (#6626)

Co-authored-by: Ellie Hermaszewska <ellieh@nvidia.com>
---
 source/slang/slang-ir-util.cpp                   |  4 +-
 tests/autodiff/force-inline-differentiable.slang | 47 ++++++++++++++++++++++++
 2 files changed, 50 insertions(+), 1 deletion(-)
 create mode 100644 tests/autodiff/force-inline-differentiable.slang

diff --git a/source/slang/slang-ir-util.cpp b/source/slang/slang-ir-util.cpp
index 4aef7e7ba..58bb7aaf2 100644
--- a/source/slang/slang-ir-util.cpp
+++ b/source/slang/slang-ir-util.cpp
@@ -2162,7 +2162,9 @@ void legalizeDefUse(IRGlobalValueWithCode* func)
             {
                 // If inst is an var, this is easy, we just move it to the
                 // common dominator.
-                var->insertBefore(commonDominator->getTerminator());
+                if (var->getParent() != commonDominator)
+                    var->insertBefore(commonDominator->getTerminator());
+
                 if (shouldInitializeVar)
                 {
                     IRBuilder builder(func);
diff --git a/tests/autodiff/force-inline-differentiable.slang b/tests/autodiff/force-inline-differentiable.slang
new file mode 100644
index 000000000..5d22a525f
--- /dev/null
+++ b/tests/autodiff/force-inline-differentiable.slang
@@ -0,0 +1,47 @@
+//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHECK): -slang -compute -shaderobj -output-using-type
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+// A general test for differentiable, force-inlined functions.
+//
+// This test was added to expose a bug in legalizeDefUse() where an IRVar can be be hoisted to 
+// after its uses (when it is already in the same block, but gets reinserted), and cause 
+// a validation bug.
+// 
+
+__generic<T : __BuiltinFloatingPointType, let R : int, let C : int>
+[ForceInline, PreferRecompute, Differentiable]
+vector<T, R> column(matrix<T, R, C> m, int i) {
+    vector<T, R> result;
+    [ForceUnroll]
+    for (int j = 0; j < R; j++) {
+        result[j] = m[j][i];
+    }
+    return result;
+}
+
+struct Data {
+    float4x4 m;
+    float foo() {
+        return length(column(m, 0).xyz);
+    }
+};
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 id : SV_DispatchThreadID)
+{
+    Data data;
+    // Initialize matrix with some values
+    data.m = float4x4(
+        3.0, 0.0, 0.0, 0.0,
+        0.0, 5.0, 0.0, 0.0,
+        0.0, 0.0, 7.0, 0.0,
+        0.0, 0.0, 0.0, 11.0);
+    
+    // Calculate and store result
+    outputBuffer[0] = data.foo();
+    
+    // CHECK: type: float
+    // CHECK: 3.0
+}
\ No newline at end of file
-- 
cgit v1.2.3