[PATCH] D71374: Improve support of GNU mempcpy

[serge via Phabricator via cfe-commits]

serge-sans-paille added a comment.

In D71374#1783032 <https://reviews.llvm.org/D71374#1783032>, @Jim wrote:

> I am curious what is difference of code generation after applying your changes?


Before, when compiling

  #define _GNU_SOURCE
  #include <string.h>
  
  void* foo(void* to, void* from, unsigned n) {
    return mempcpy(mempcpy(to, from, n), from, n);
  }

We get (clang -O3)

  define i8* @foo(i8*, i8*, i32) #0 {
    %4 = alloca i8*, align 8
    %5 = alloca i8*, align 8
    %6 = alloca i32, align 4
    store i8* %0, i8** %4, align 8
    store i8* %1, i8** %5, align 8
    store i32 %2, i32* %6, align 4
    %7 = load i8*, i8** %4, align 8
    %8 = load i8*, i8** %5, align 8
    %9 = load i32, i32* %6, align 4
    %10 = zext i32 %9 to i64
    %11 = call i8* @mempcpy(i8* %7, i8* %8, i64 %10) #2
    %12 = load i8*, i8** %5, align 8
    %13 = load i32, i32* %6, align 4
    %14 = zext i32 %13 to i64
    %15 = call i8* @mempcpy(i8* %11, i8* %12, i64 %14) #2
    ret i8* %15
  }

And we now get

  define dso_local i8* @foo(i8* %to, i8* nocapture readonly %from, i32 %n) local_unnamed_addr #0 {
  entry:
    %conv = zext i32 %n to i64
    tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 %to, i8* align 1 %from, i64 %conv, i1 false)
    %0 = getelementptr i8, i8* %to, i64 %conv
    tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 %0, i8* align 1 %from, i64 %conv, i1 false)
    %1 = getelementptr i8, i8* %0, i64 %conv
    ret i8* %1
  }

Which looks much better to me, esp. as it unlocks memcpy-specific optimisations


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D71374/new/

https://reviews.llvm.org/D71374