Adding a differentiable standard library (#2465)

4 files changed, 169 insertions, 0 deletions

diff --git a/tests/autodiff/custom-intrinsic.slang b/tests/autodiff/custom-intrinsic.slang
new file mode 100644
index 000000000..8ce354edc
--- /dev/null
+++ b/tests/autodiff/custom-intrinsic.slang
@@ -0,0 +1,119 @@
+//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], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+typedef __DifferentialPair<float> dpfloat;
+
+typealias IDFloat = IFloat & IDifferentiable;
+
+namespace myintrinsiclib
+{
+    __generic<T : IDFloat>
+    __target_intrinsic(hlsl)
+    __target_intrinsic(glsl)
+    __target_intrinsic(cuda, "$P_exp($0)")
+    __target_intrinsic(cpp, "$P_exp($0)")
+    __target_intrinsic(spirv_direct, "12 resultType resultId glsl450 27 _0")
+    [__custom_jvp(d_exp<T>)]
+    T exp(T x);
+
+    __generic<T : IDFloat>
+    __DifferentialPair<T> d_exp(__DifferentialPair<T> dpx)
+    {
+        return __DifferentialPair<T>(
+            exp(dpx.p()),
+            T.dmul(exp(dpx.p()), dpx.d()));
+    }
+
+    
+    // Sine
+    __generic<T : IDFloat>
+    __target_intrinsic(hlsl)
+    __target_intrinsic(glsl)
+    __target_intrinsic(cuda, "$P_sin($0)")
+    __target_intrinsic(cpp, "$P_sin($0)")
+    __target_intrinsic(spirv_direct, "12 resultType resultId glsl450 13 _0")
+    [__custom_jvp(d_sin<T>)]
+    T sin(T x);
+
+    __generic<T : IDFloat>
+    __DifferentialPair<T> d_sin(__DifferentialPair<T> dpx)
+    {
+        return __DifferentialPair<T>(
+            sin(dpx.p()),
+            T.dmul(cos(dpx.p()), dpx.d()));
+    }
+
+    // Cosine
+    __generic<T : IDFloat>
+    __target_intrinsic(hlsl)
+    __target_intrinsic(glsl)
+    __target_intrinsic(cuda, "$P_cos($0)")
+    __target_intrinsic(cpp, "$P_cos($0)")
+    __target_intrinsic(spirv_direct, "12 resultType resultId glsl450 14 _0")
+    [__custom_jvp(d_cos<T>)]
+    T cos(T x);
+
+    __generic<T : IDFloat>
+    __DifferentialPair<T> d_cos(__DifferentialPair<T> dpx)
+    {
+        return __DifferentialPair<T>(
+            cos(dpx.p()),
+            T.dmul(-sin(dpx.p()), dpx.d()));
+    }
+
+    // Sine and cosine
+    __generic<T : IDFloat>
+    __target_intrinsic(hlsl)
+    __target_intrinsic(cuda, "$P_sincos($0, $1, $2)")
+    [__custom_jvp(d_sincos<T>)]
+    void sincos(T x, out T s, out T c)
+    {
+        s = sin(x);
+        c = cos(x);
+    }
+
+    __generic<T : IDFloat>
+    void d_sincos(__DifferentialPair<T> x, out __DifferentialPair<T> s, out __DifferentialPair<T> c)
+    {
+        T _s;
+        T _c;
+        sincos(x.p(), _s, _c);
+
+        s = __DifferentialPair<T>(_s, T.dmul(_c, x.d()));
+        c = __DifferentialPair<T>(_c, T.dmul(-_s, x.d()));
+    }
+};
+
+__differentiate_jvp float f(float x)
+{
+    return myintrinsiclib.exp(x);
+}
+
+__differentiate_jvp float g(float x)
+{
+    float s;
+    float t;
+    myintrinsiclib.sincos(x, s, t);
+
+    return s + t;
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID: SV_DispatchThreadID)
+{
+    {
+        dpfloat dpa = dpfloat(2.0, 1.0);
+
+        outputBuffer[0] = f(dpa.p());        // Expect: 7.389056
+        outputBuffer[1] = __jvp(f)(dpa).d(); // Expect: 7.389056
+
+        // g() needs additional handling of  IRMakeDifferentialPair(PtrType). This needs to 
+        // generate a new var, load from the individual vars and store into the pair var.
+
+        //outputBuffer[2] = g(dpa.p());        // Expect: 1.381773
+        //outputBuffer[3] = __jvp(g)(dpa).d(); // Expect: -0.301168
+    }
+}
\ No newline at end of file
diff --git a/tests/autodiff/custom-intrinsic.slang.expected.txt b/tests/autodiff/custom-intrinsic.slang.expected.txt
new file mode 100644
index 000000000..ce22a5b95
--- /dev/null
+++ b/tests/autodiff/custom-intrinsic.slang.expected.txt
@@ -0,0 +1,6 @@
+type: float
+7.389056
+7.389056
+0.0
+0.0
+0.0
\ No newline at end of file
diff --git a/tests/autodiff/dstdlib.slang b/tests/autodiff/dstdlib.slang
new file mode 100644
index 000000000..6c7ecffbe
--- /dev/null
+++ b/tests/autodiff/dstdlib.slang
@@ -0,0 +1,37 @@
+//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], stride=4):out,name=outputBuffer
+RWStructuredBuffer<float> outputBuffer;
+
+typedef __DifferentialPair<float> dpfloat;
+
+__differentiate_jvp float f(float x)
+{
+    return dstd.exp(x);
+}
+
+__differentiate_jvp float g(float x)
+{
+    return dstd.sin(x);
+}
+
+__differentiate_jvp float h(float x)
+{
+    return dstd.cos(x);
+}
+
+[numthreads(1, 1, 1)]
+void computeMain(uint3 dispatchThreadID: SV_DispatchThreadID)
+{
+    {
+        dpfloat dpa = dpfloat(2.0, 1.0);
+
+        outputBuffer[0] = f(dpa.p());        // Expect: 7.389056
+        outputBuffer[1] = __jvp(f)(dpa).d(); // Expect: 7.389056
+        outputBuffer[2] = g(dpa.p());        // Expect: 0.909297 
+        outputBuffer[3] = __jvp(g)(dpa).d(); // Expect: -0.416146
+        outputBuffer[4] = h(dpa.p());        // Expect: -0.416146
+        outputBuffer[5] = __jvp(h)(dpa).d(); // Expect: -0.909297 
+    }
+}
\ No newline at end of file
diff --git a/tests/autodiff/dstdlib.slang.expected.txt b/tests/autodiff/dstdlib.slang.expected.txt
new file mode 100644
index 000000000..82053b379
--- /dev/null
+++ b/tests/autodiff/dstdlib.slang.expected.txt
@@ -0,0 +1,7 @@
+type: float
+7.389056
+7.389056
+0.909297
+-0.416147
+-0.416147
+-0.909297
\ No newline at end of file