Fix `atan2` stdlib derivative + add tests. (#3218)

* Fix atan2 stdlib derivative. Add tests for atan2

* Create dstdlib-atan2.slang.expected.txt

* Update tests

3 files changed, 99 insertions, 3 deletions

diff --git a/source/slang/diff.meta.slang b/source/slang/diff.meta.slang
index 495b6b989..75c57018c 100644
--- a/source/slang/diff.meta.slang
+++ b/source/slang/diff.meta.slang
@@ -1259,7 +1259,7 @@ __generic<T : __BuiltinFloatingPointType>
 DifferentialPair<T> __d_atan2(DifferentialPair<T> dpy, DifferentialPair<T> dpx)
 {
     T.Differential dx = __mul_p_d(-dpy.p / (dpx.p * dpx.p + dpy.p * dpy.p), dpx.d);
-    T.Differential dy = __mul_p_d(-dpx.p / (dpx.p * dpx.p + dpy.p * dpy.p), dpy.d);
+    T.Differential dy = __mul_p_d(dpx.p / (dpx.p * dpx.p + dpy.p * dpy.p), dpy.d);
     return DifferentialPair<T>(
         atan2(dpy.p, dpx.p),
         T.dadd(dx, dy));
@@ -1271,8 +1271,8 @@ __generic<T : __BuiltinFloatingPointType>
 [BackwardDerivativeOf(atan2)]
 void __d_atan2(inout DifferentialPair<T> dpy, inout DifferentialPair<T> dpx, T.Differential dOut)
 {
-    dpx = diffPair(dpx.p, __mul_p_d(-dpy.p / (dpx.p * dpx.p + dpy.p * dpy.p), dpx.d));
-    dpy = diffPair(dpy.p, __mul_p_d(-dpx.p / (dpx.p * dpx.p + dpy.p * dpy.p), dpy.d));
+    dpx = diffPair(dpx.p, __mul_p_d(-dpy.p / (dpx.p * dpx.p + dpy.p * dpy.p), dOut));
+    dpy = diffPair(dpy.p, __mul_p_d(dpx.p / (dpx.p * dpx.p + dpy.p * dpy.p), dOut));
 }
 
 VECTOR_MATRIX_BINARY_DIFF_IMPL(atan2)
diff --git a/tests/autodiff-dstdlib/dstdlib-atan2.slang b/tests/autodiff-dstdlib/dstdlib-atan2.slang
new file mode 100644
index 000000000..c15222336
--- /dev/null
+++ b/tests/autodiff-dstdlib/dstdlib-atan2.slang
@@ -0,0 +1,85 @@
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+typedef DifferentialPair<float> dpfloat;
+typedef DifferentialPair<float2> dpfloat2;
+typedef DifferentialPair<float3> dpfloat3;
+
+[BackwardDifferentiable]
+float diffAtan2(float x, float y)
+{
+    return atan2(x, y);
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID: SV_DispatchThreadID)
+{
+    float eps = 1e-4;
+    {
+        dpfloat dpx = dpfloat(5.0, 1.0);
+        dpfloat dpy = dpfloat(3.0, 0.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[0] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(5.0, 0.0);
+        dpfloat dpy = dpfloat(3.0, 1.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[1] = res.d;
+    }
+
+    // Test the other 3 quadrants
+    {
+        dpfloat dpx = dpfloat(-5.0, 1.0);
+        dpfloat dpy = dpfloat(3.0, 0.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[2] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(-5.0, 0.0);
+        dpfloat dpy = dpfloat(3.0, 1.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[3] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(-5.0, 1.0);
+        dpfloat dpy = dpfloat(-3.0, 0.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[4] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(-5.0, 0.0);
+        dpfloat dpy = dpfloat(-3.0, 1.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[5] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(5.0, 1.0);
+        dpfloat dpy = dpfloat(-3.0, 0.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[6] = res.d;
+    }
+
+    {
+        dpfloat dpx = dpfloat(5.0, 0.0);
+        dpfloat dpy = dpfloat(-3.0, 1.0);
+        dpfloat res = __fwd_diff(diffAtan2)(dpx, dpy);
+        outputBuffer[7] = res.d;
+    }
+
+    {
+        dpfloat dpx = diffPair(5.0);
+        dpfloat dpy = diffPair(3.0);
+        __bwd_diff(diffAtan2)(dpx, dpy, 1.0);
+        outputBuffer[8] = dpx.d; // Should be equal to outputBuffer[0]
+        outputBuffer[9] = dpy.d; // Should be equal to outputBuffer[2]
+    }
+}
diff --git a/tests/autodiff-dstdlib/dstdlib-atan2.slang.expected.txt b/tests/autodiff-dstdlib/dstdlib-atan2.slang.expected.txt
new file mode 100644
index 000000000..5adffdaef
--- /dev/null
+++ b/tests/autodiff-dstdlib/dstdlib-atan2.slang.expected.txt
@@ -0,0 +1,11 @@
+type: float
+0.088235
+-0.147059
+0.088235
+0.147059
+-0.088235
+0.147059
+-0.088235
+-0.147059
+0.088235
+-0.147059