https://bugs.llvm.org/show_bug.cgi?id=40869

            Bug ID: 40869
           Summary: [X86] Poor broadcast folding from ext/trunc loads
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedb...@nondot.org
          Reporter: llvm-...@redking.me.uk
                CC: craig.top...@gmail.com, llvm-bugs@lists.llvm.org,
                    llvm-...@redking.me.uk, spatel+l...@rotateright.com

e.g. (from vector-shuffle-512-v32.ll)

llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx 

define <32 x i16> @insert_dup_elt1_mem_v32i16_i32(i32* %ptr) #0 {
; KNL-LABEL: insert_dup_elt1_mem_v32i16_i32:
; KNL:       ## %bb.0:
; KNL-NEXT:    vpbroadcastw 2(%rdi), %ymm0
; KNL-NEXT:    vmovdqa %ymm0, %ymm1
; KNL-NEXT:    retq
;
; SKX-LABEL: insert_dup_elt1_mem_v32i16_i32:
; SKX:       ## %bb.0:
; SKX-NEXT:    movzwl 2(%rdi), %eax
; SKX-NEXT:    vpbroadcastw %eax, %zmm0
; SKX-NEXT:    retq
  %tmp = load i32, i32* %ptr, align 4
  %tmp1 = insertelement <4 x i32> zeroinitializer, i32 %tmp, i32 0
  %tmp2 = bitcast <4 x i32> %tmp1 to <8 x i16>
  %tmp3 = shufflevector <8 x i16> %tmp2, <8 x i16> undef, <32 x i32> <i32 1,
i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1,
i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1,
i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
  ret <32 x i16> %tmp3
}

Notice how the KNL (AVX2) version manages to fold but SKX (AVX512BWVL) ymm
broadcasts fail.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to