[llvm-bugs] [Bug 42716] New: Optimize lerp with two FMAs

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=42716

            Bug ID: 42716
           Summary: Optimize lerp with two FMAs
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: david.bolvansky at gmail.com
                CC: llvm-bugs at lists.llvm.org

Tip from: https://devblogs.nvidia.com/lerp-faster-cuda/

float lerp(float a, float b, float c) {
    return (1-c)*a +c*b;
}


float lerp2(float a, float b, float c)
{
    return a + c * (b - a);
}

float rf2(float a, float b, float c) {
    return std::fma(c, b, std::fma(-c, a, a));
}

Clang trunk -Ofast -std=c++17 -mfma


lerp(float, float, float):                             # @lerp(float, float,
float)
        vmovss  xmm3, dword ptr [rip + .LCPI0_0] # xmm3 = mem[0],zero,zero,zero
        vsubss  xmm3, xmm3, xmm2
        vmulss  xmm1, xmm2, xmm1
        vfmadd213ss     xmm0, xmm3, xmm1 # xmm0 = (xmm3 * xmm0) + xmm1
        ret
lerp2(float, float, float):                            # @lerp2(float, float,
float)
        vsubss  xmm1, xmm1, xmm0
        vfmadd231ss     xmm0, xmm2, xmm1 # xmm0 = (xmm2 * xmm1) + xmm0
        ret
rf2(float, float, float):                              # @rf2(float, float,
float)
        vfnmadd213ss    xmm0, xmm2, xmm0 # xmm0 = -(xmm2 * xmm0) + xmm0
        vfmadd231ss     xmm0, xmm2, xmm1 # xmm0 = (xmm2 * xmm1) + xmm0
        ret


InstCombine?

A) Missing fold: (1-c)*a +c*b -> a + c * (b - a) 
B) Fold a + c * (b - a) to llvm.fma(c, b, llvm.fma(-c, a, a)) if FMA enabled

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20190722/46e4f6c3/attachment.html>