//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -shaderobj -output-using-type
//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -shaderobj -output-using-type
//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -shaderobj -output-using-type

//TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0], stride=4):out,name=outputBuffer
RWStructuredBuffer<float> outputBuffer;

typedef DifferentialPair<float> dpfloat;

[BackwardDifferentiable]
float f(float x)
{
    return exp(x);
}

[BackwardDifferentiable]
float g(float x)
{
    return sin(x);
}

[BackwardDifferentiable]
float h(float x)
{
    return 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] = __fwd_diff(f)(dpa).d; // Expect: 7.389056
        outputBuffer[2] = g(dpa.p);        // Expect: 0.909297
        outputBuffer[3] = __fwd_diff(g)(dpa).d; // Expect: -0.416146
        outputBuffer[4] = h(dpa.p);        // Expect: -0.416146
        outputBuffer[5] = __fwd_diff(h)(dpa).d; // Expect: -0.909297
    }

    {
        dpfloat dpa = dpfloat(2.0, 0.0);
        __bwd_diff(f)(dpa, 1.0);
        outputBuffer[6] = dpa.d; // Expect: 7.389056
        __bwd_diff(g)(dpa, 1.0);
        outputBuffer[7] = dpa.d; // Expect: -0.416146
        __bwd_diff(h)(dpa, 1.0);
        outputBuffer[8] = dpa.d; // Expect: -0.909297
    }
}