[llvm-bugs] [Bug 27100] New: Optimize memset for AVX2

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=27100

            Bug ID: 27100
           Summary: Optimize memset for AVX2
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: hjl.tools at gmail.com
                CC: david.l.kreitzer at intel.com, llvm-bugs at lists.llvm.org,
                    zia.ansari at intel.com
    Classification: Unclassified

For

[hjl at gnu-6 memcpy-3]$ cat c.i
extern char *src, *dst;

void
foo3 (int x)
{
  __builtin_memset (dst, x, 64);
}

void
foo4 (int x)
{
  __builtin_memset (dst, x, 32);
}

void
foo5 (int x)
{
  __builtin_memset (dst, x, 16);
}
[hjl at gnu-6 memcpy-3]$ 

clang trunk -O2 -mavx2 generates

[hjl at gnu-6 memcpy-3]$ cat c.s
    .text
    .file    "c.i"
    .globl    foo3
    .p2align    4, 0x90
    .type    foo3, at function
foo3:                                   # @foo3
    .cfi_startproc
# BB#0:
    movl    dst(%rip), %eax
    movzbl    %dil, %ecx
    movabsq    $72340172838076673, %rdx # imm = 0x101010101010101
    imulq    %rcx, %rdx
    movq    %rdx, 56(%eax)
    movq    %rdx, 48(%eax)
    movq    %rdx, 40(%eax)
    movq    %rdx, 32(%eax)
    movq    %rdx, 24(%eax)
    movq    %rdx, 16(%eax)
    movq    %rdx, 8(%eax)
    movq    %rdx, (%eax)
    retq
.Lfunc_end0:
    .size    foo3, .Lfunc_end0-foo3
    .cfi_endproc

    .globl    foo4
    .p2align    4, 0x90
    .type    foo4, at function
foo4:                                   # @foo4
    .cfi_startproc
# BB#0:
    movl    dst(%rip), %eax
    movzbl    %dil, %ecx
    movabsq    $72340172838076673, %rdx # imm = 0x101010101010101
    imulq    %rcx, %rdx
    movq    %rdx, 24(%eax)
    movq    %rdx, 16(%eax)
    movq    %rdx, 8(%eax)
    movq    %rdx, (%eax)
    retq
.Lfunc_end1:
    .size    foo4, .Lfunc_end1-foo4
    .cfi_endproc

    .globl    foo5
    .p2align    4, 0x90
    .type    foo5, at function
foo5:                                   # @foo5
    .cfi_startproc
# BB#0:
    movl    dst(%rip), %eax
    movzbl    %dil, %ecx
    movabsq    $72340172838076673, %rdx # imm = 0x101010101010101
    imulq    %rcx, %rdx
    movq    %rdx, 8(%eax)
    movq    %rdx, (%eax)
    retq
.Lfunc_end2:
    .size    foo5, .Lfunc_end2-foo5
    .cfi_endproc

Will
    .text
    .p2align 4,,15
    .globl    foo3
    .type    foo3, @function
foo3:
.LFB0:
    .cfi_startproc
    movq    dst(%rip), %rax
    vmovd    %edi, %xmm0
    pushq    %rbp
    .cfi_def_cfa_offset 16
    .cfi_offset 6, -16
    vpbroadcastb    %xmm0, %ymm0
    vmovdqu    %ymm0, (%rax)
    movq    %rsp, %rbp
    .cfi_def_cfa_register 6
    vmovdqu    %ymm0, 32(%rax)
    vzeroupper
    popq    %rbp
    .cfi_def_cfa 7, 8
    ret
    .cfi_endproc
.LFE0:
    .size    foo3, .-foo3
    .p2align 4,,15
    .globl    foo4
    .type    foo4, @function
foo4:
.LFB1:
    .cfi_startproc
    movq    dst(%rip), %rax
    vmovd    %edi, %xmm0
    pushq    %rbp
    .cfi_def_cfa_offset 16
    .cfi_offset 6, -16
    vpbroadcastb    %xmm0, %ymm0
    vmovdqu    %ymm0, (%rax)
    movq    %rsp, %rbp
    .cfi_def_cfa_register 6
    vzeroupper
    popq    %rbp
    .cfi_def_cfa 7, 8
    ret
    .cfi_endproc
.LFE1:
    .size    foo4, .-foo4
    .p2align 4,,15
    .globl    foo5
    .type    foo5, @function
foo5:
.LFB2:
    .cfi_startproc
    movq    dst(%rip), %rax
    vmovd    %edi, %xmm0
    vpbroadcastb    %xmm0, %xmm0
    vmovdqu    %xmm0, (%rax)
    ret
    .cfi_endproc
.LFE2:
    .size    foo5, .-foo5


be better?

-- 
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/20160328/cc74ce36/attachment.html>