[llvm-bugs] [Bug 33247] New: Enable extractelement to alias with vector spills

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=33247

            Bug ID: 33247
           Summary: Enable extractelement to alias with vector spills
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: andrea.dibiagio at gmail.com, davide at freebsd.org,
                    filcab at gmail.com, llvm-bugs at lists.llvm.org,
                    spatel+llvm at rotateright.com

define i32 @popcnt_i128(<4 x i32> %a0) {
  %1  = tail call <2 x i64> asm sideeffect "nop",
"=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15},~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
  %2  = extractelement <4 x i32> %a0, i32 0
  %3  = extractelement <4 x i32> %a0, i32 1
  %4  = extractelement <4 x i32> %a0, i32 2
  %5  = extractelement <4 x i32> %a0, i32 3
  %6  = call i32 @llvm.ctpop.i32(i32 %2)
  %7  = call i32 @llvm.ctpop.i32(i32 %3)
  %8  = call i32 @llvm.ctpop.i32(i32 %4)
  %9  = call i32 @llvm.ctpop.i32(i32 %5)
  %10 = add i32 %6, %7
  %11 = add i32 %8, %9
  %12 = add i32 %10, %11
  ret i32 %12
}
declare i32 @llvm.ctpop.i32(i32) nounwind readnone

llc -mtriple=x86_64-unknown -mcpu=btver2

popcnt_i128:
        pushq   %rbp
        pushq   %r15
        pushq   %r14
        pushq   %r13
        pushq   %r12
        pushq   %rbx
        vmovd   %xmm0, -4(%rsp)         # 4-byte Folded Spill
        vpextrd $1, %xmm0, -8(%rsp)     # 4-byte Folded Spill
        vpextrd $2, %xmm0, -12(%rsp)    # 4-byte Folded Spill
        vpextrd $3, %xmm0, -16(%rsp)    # 4-byte Folded Spill
        #APP
        nop
        #NO_APP
        popcntl -4(%rsp), %ecx          # 4-byte Folded Reload
        popcntl -8(%rsp), %edx          # 4-byte Folded Reload
        popcntl -12(%rsp), %esi         # 4-byte Folded Reload
        popcntl -16(%rsp), %eax         # 4-byte Folded Reload
        addl    %ecx, %edx
        addl    %esi, %eax
        addl    %edx, %eax
        popq    %rbx
        popq    %r12
        popq    %r13
        popq    %r14
        popq    %r15
        popq    %rbp
        retq

Each extractelement is being spilled to stack separately, requiring 4 scalar
stores. Ideally the vector would be spilled whole and the reloads done as
scalars:

popcnt_i128_IDEAL:
        pushq   %rbp
        pushq   %r15
        pushq   %r14
        pushq   %r13
        pushq   %r12
        pushq   %rbx
        vmovdqa %xmm0, -16(%rsp)        # 16-byte Folded Spill
        #APP
        nop
        #NO_APP
        popcntl -4(%rsp), %ecx          # 4-byte Folded Reload
        popcntl -8(%rsp), %edx          # 4-byte Folded Reload
        popcntl -12(%rsp), %esi         # 4-byte Folded Reload
        popcntl -16(%rsp), %eax         # 4-byte Folded Reload
        addl    %ecx, %edx
        addl    %esi, %eax
        addl    %edx, %eax
        popq    %rbx
        popq    %r12
        popq    %r13
        popq    %r14
        popq    %r15
        popq    %rbp
        retq

-- 
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/20170531/9ee013eb/attachment.html>