[PATCH] x86/Intel: correct AVX512F scatter insn element sizes
H.J. Lu
hjl.tools@gmail.com
Wed Jul 20 16:48:39 GMT 2022
On Wed, Jul 20, 2022 at 12:57 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> I clearly screwed up in 6ff00b5e12e7 ("x86/Intel: correct permitted
> operand sizes for AVX512 scatter/gather") giving all AVX512F scatter
> insns Dword element size. Update testcases (also their gather parts),
> utilizing that there previously were two identical lines each (for no
> apparent reason).
> ---
> Perhaps another candidate to also go on the 2.39 branch.
OK for master and 2.39 branch.
Thanks.
> --- a/gas/testsuite/gas/i386/avx512f.s
> +++ b/gas/testsuite/gas/i386/avx512f.s
> @@ -11109,22 +11109,22 @@ _start:
> vfnmsub231ss xmm6{k7}, xmm5, DWORD PTR [edx-516] # AVX512F
>
> vgatherdpd zmm6{k1}, [ebp+ymm7*8-123] # AVX512F
> - vgatherdpd zmm6{k1}, [ebp+ymm7*8-123] # AVX512F
> + vgatherdpd zmm6{k1}, qword ptr [ebp+ymm7*8-123] # AVX512F
> vgatherdpd zmm6{k1}, [eax+ymm7+256] # AVX512F
> vgatherdpd zmm6{k1}, [ecx+ymm7*4+1024] # AVX512F
>
> vgatherdps zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vgatherdps zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vgatherdps zmm6{k1}, dword ptr [ebp+zmm7*8-123] # AVX512F
> vgatherdps zmm6{k1}, [eax+zmm7+256] # AVX512F
> vgatherdps zmm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> vgatherqpd zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vgatherqpd zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vgatherqpd zmm6{k1}, qword ptr [ebp+zmm7*8-123] # AVX512F
> vgatherqpd zmm6{k1}, [eax+zmm7+256] # AVX512F
> vgatherqpd zmm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> vgatherqps ymm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vgatherqps ymm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vgatherqps ymm6{k1}, dword ptr [ebp+zmm7*8-123] # AVX512F
> vgatherqps ymm6{k1}, [eax+zmm7+256] # AVX512F
> vgatherqps ymm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> @@ -12401,22 +12401,22 @@ _start:
> vpexpandq zmm6{k7}{z}, zmm5 # AVX512F
>
> vpgatherdd zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vpgatherdd zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vpgatherdd zmm6{k1}, dword ptr [ebp+zmm7*8-123] # AVX512F
> vpgatherdd zmm6{k1}, [eax+zmm7+256] # AVX512F
> vpgatherdd zmm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> vpgatherdq zmm6{k1}, [ebp+ymm7*8-123] # AVX512F
> - vpgatherdq zmm6{k1}, [ebp+ymm7*8-123] # AVX512F
> + vpgatherdq zmm6{k1}, qword ptr [ebp+ymm7*8-123] # AVX512F
> vpgatherdq zmm6{k1}, [eax+ymm7+256] # AVX512F
> vpgatherdq zmm6{k1}, [ecx+ymm7*4+1024] # AVX512F
>
> vpgatherqd ymm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vpgatherqd ymm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vpgatherqd ymm6{k1}, dword ptr [ebp+zmm7*8-123] # AVX512F
> vpgatherqd ymm6{k1}, [eax+zmm7+256] # AVX512F
> vpgatherqd ymm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> vpgatherqq zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> - vpgatherqq zmm6{k1}, [ebp+zmm7*8-123] # AVX512F
> + vpgatherqq zmm6{k1}, qword ptr [ebp+zmm7*8-123] # AVX512F
> vpgatherqq zmm6{k1}, [eax+zmm7+256] # AVX512F
> vpgatherqq zmm6{k1}, [ecx+zmm7*4+1024] # AVX512F
>
> @@ -12706,22 +12706,22 @@ _start:
> vporq zmm6, zmm5, qword bcst [edx-1032] # AVX512F
>
> vpscatterdd [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> - vpscatterdd [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> + vpscatterdd dword ptr [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> vpscatterdd [eax+zmm7+256]{k1}, zmm6 # AVX512F
> vpscatterdd [ecx+zmm7*4+1024]{k1}, zmm6 # AVX512F
>
> vpscatterdq [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> - vpscatterdq [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> + vpscatterdq qword ptr [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> vpscatterdq [eax+ymm7+256]{k1}, zmm6 # AVX512F
> vpscatterdq [ecx+ymm7*4+1024]{k1}, zmm6 # AVX512F
>
> vpscatterqd [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> - vpscatterqd [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> + vpscatterqd dword ptr [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> vpscatterqd [eax+zmm7+256]{k1}, ymm6 # AVX512F
> vpscatterqd [ecx+zmm7*4+1024]{k1}, ymm6 # AVX512F
>
> vpscatterqq [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> - vpscatterqq [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> + vpscatterqq qword ptr [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> vpscatterqq [eax+zmm7+256]{k1}, zmm6 # AVX512F
> vpscatterqq [ecx+zmm7*4+1024]{k1}, zmm6 # AVX512F
>
> @@ -13162,22 +13162,22 @@ _start:
> vrsqrt14ss xmm6{k7}, xmm5, DWORD PTR [edx-516] # AVX512F
>
> vscatterdpd [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> - vscatterdpd [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> + vscatterdpd qword ptr [ebp+ymm7*8-123]{k1}, zmm6 # AVX512F
> vscatterdpd [eax+ymm7+256]{k1}, zmm6 # AVX512F
> vscatterdpd [ecx+ymm7*4+1024]{k1}, zmm6 # AVX512F
>
> vscatterdps [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> - vscatterdps [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> + vscatterdps dword ptr [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> vscatterdps [eax+zmm7+256]{k1}, zmm6 # AVX512F
> vscatterdps [ecx+zmm7*4+1024]{k1}, zmm6 # AVX512F
>
> vscatterqpd [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> - vscatterqpd [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> + vscatterqpd qword ptr [ebp+zmm7*8-123]{k1}, zmm6 # AVX512F
> vscatterqpd [eax+zmm7+256]{k1}, zmm6 # AVX512F
> vscatterqpd [ecx+zmm7*4+1024]{k1}, zmm6 # AVX512F
>
> vscatterqps [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> - vscatterqps [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> + vscatterqps dword ptr [ebp+zmm7*8-123]{k1}, ymm6 # AVX512F
> vscatterqps [eax+zmm7+256]{k1}, ymm6 # AVX512F
> vscatterqps [ecx+zmm7*4+1024]{k1}, ymm6 # AVX512F
>
> --- a/gas/testsuite/gas/i386/x86-64-avx512f.s
> +++ b/gas/testsuite/gas/i386/x86-64-avx512f.s
> @@ -11618,23 +11618,23 @@ _start:
> vfnmsub231ss xmm30{k7}, xmm29, DWORD PTR [rdx-516] # AVX512F
>
> vgatherdpd zmm30{k1}, [r14+ymm31*8-123] # AVX512F
> - vgatherdpd zmm30{k1}, [r14+ymm31*8-123] # AVX512F
> + vgatherdpd zmm30{k1}, qword ptr [r14+ymm31*8-123] # AVX512F
> vgatherdpd zmm30{k1}, [r9+ymm31+256] # AVX512F
> vgatherdpd zmm30{k1}, [rcx+ymm31*4+1024] # AVX512F
>
> vgatherdps zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vgatherdps zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vgatherdps zmm30{k1}, dword ptr [r14+zmm31*8-123] # AVX512F
> vgatherdps zmm30{k1}, [r9+zmm31+256] # AVX512F
> vgatherdps zmm30{k1}, [rcx+zmm31*4+1024] # AVX512F
>
> vgatherqpd zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vgatherqpd zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vgatherqpd zmm30{k1}, qword ptr [r14+zmm31*8-123] # AVX512F
> vgatherqpd zmm30{k1}, [r9+zmm31+256] # AVX512F
> vgatherqpd zmm30{k1}, [rcx+zmm31*4+1024] # AVX512F
> vgatherqpd zmm3{k1}, [r14+zmm19*8+123] # AVX512F
>
> vgatherqps ymm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vgatherqps ymm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vgatherqps ymm30{k1}, dword ptr [r14+zmm31*8-123] # AVX512F
> vgatherqps ymm30{k1}, [r9+zmm31+256] # AVX512F
> vgatherqps ymm30{k1}, [rcx+zmm31*4+1024] # AVX512F
>
> @@ -13021,22 +13021,22 @@ _start:
> vpexpandq zmm30{k7}{z}, zmm29 # AVX512F
>
> vpgatherdd zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vpgatherdd zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vpgatherdd zmm30{k1}, dword ptr [r14+zmm31*8-123] # AVX512F
> vpgatherdd zmm30{k1}, [r9+zmm31+256] # AVX512F
> vpgatherdd zmm30{k1}, [rcx+zmm31*4+1024] # AVX512F
>
> vpgatherdq zmm30{k1}, [r14+ymm31*8-123] # AVX512F
> - vpgatherdq zmm30{k1}, [r14+ymm31*8-123] # AVX512F
> + vpgatherdq zmm30{k1}, qword ptr [r14+ymm31*8-123] # AVX512F
> vpgatherdq zmm30{k1}, [r9+ymm31+256] # AVX512F
> vpgatherdq zmm30{k1}, [rcx+ymm31*4+1024] # AVX512F
>
> vpgatherqd ymm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vpgatherqd ymm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vpgatherqd ymm30{k1}, dword ptr [r14+zmm31*8-123] # AVX512F
> vpgatherqd ymm30{k1}, [r9+zmm31+256] # AVX512F
> vpgatherqd ymm30{k1}, [rcx+zmm31*4+1024] # AVX512F
>
> vpgatherqq zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> - vpgatherqq zmm30{k1}, [r14+zmm31*8-123] # AVX512F
> + vpgatherqq zmm30{k1}, qword ptr [r14+zmm31*8-123] # AVX512F
> vpgatherqq zmm30{k1}, [r9+zmm31+256] # AVX512F
> vpgatherqq zmm30{k1}, [rcx+zmm31*4+1024] # AVX512F
>
> @@ -13326,22 +13326,22 @@ _start:
> vporq zmm30, zmm29, qword bcst [rdx-1032] # AVX512F
>
> vpscatterdd [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> - vpscatterdd [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> + vpscatterdd dword ptr [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> vpscatterdd [r9+zmm31+256]{k1}, zmm30 # AVX512F
> vpscatterdd [rcx+zmm31*4+1024]{k1}, zmm30 # AVX512F
>
> vpscatterdq [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> - vpscatterdq [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> + vpscatterdq qword ptr [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> vpscatterdq [r9+ymm31+256]{k1}, zmm30 # AVX512F
> vpscatterdq [rcx+ymm31*4+1024]{k1}, zmm30 # AVX512F
>
> vpscatterqd [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> - vpscatterqd [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> + vpscatterqd dword ptr [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> vpscatterqd [r9+zmm31+256]{k1}, ymm30 # AVX512F
> vpscatterqd [rcx+zmm31*4+1024]{k1}, ymm30 # AVX512F
>
> vpscatterqq [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> - vpscatterqq [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> + vpscatterqq qword ptr [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> vpscatterqq [r9+zmm31+256]{k1}, zmm30 # AVX512F
> vpscatterqq [rcx+zmm31*4+1024]{k1}, zmm30 # AVX512F
>
> @@ -13782,22 +13782,22 @@ _start:
> vrsqrt14ss xmm30{k7}, xmm29, DWORD PTR [rdx-516] # AVX512F
>
> vscatterdpd [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> - vscatterdpd [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> + vscatterdpd qword ptr [r14+ymm31*8-123]{k1}, zmm30 # AVX512F
> vscatterdpd [r9+ymm31+256]{k1}, zmm30 # AVX512F
> vscatterdpd [rcx+ymm31*4+1024]{k1}, zmm30 # AVX512F
>
> vscatterdps [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> - vscatterdps [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> + vscatterdps dword ptr [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> vscatterdps [r9+zmm31+256]{k1}, zmm30 # AVX512F
> vscatterdps [rcx+zmm31*4+1024]{k1}, zmm30 # AVX512F
>
> vscatterqpd [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> - vscatterqpd [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> + vscatterqpd qword ptr [r14+zmm31*8-123]{k1}, zmm30 # AVX512F
> vscatterqpd [r9+zmm31+256]{k1}, zmm30 # AVX512F
> vscatterqpd [rcx+zmm31*4+1024]{k1}, zmm30 # AVX512F
>
> vscatterqps [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> - vscatterqps [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> + vscatterqps dword ptr [r14+zmm31*8-123]{k1}, ymm30 # AVX512F
> vscatterqps [r9+zmm31+256]{k1}, ymm30 # AVX512F
> vscatterqps [rcx+zmm31*4+1024]{k1}, ymm30 # AVX512F
>
> --- a/opcodes/i386-opc.tbl
> +++ b/opcodes/i386-opc.tbl
> @@ -2278,10 +2278,10 @@ vcompressps, 0x668A, None, CpuAVX512F, M
> vpcompressq, 0x668B, None, CpuAVX512F, Modrm|MaskingMorZ|Space0F38|VexW=2|Disp8MemShift=3|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegYMM|RegZMM, RegXMM|RegYMM|RegZMM|Unspecified|BaseIndex }
> vpcompressd, 0x668B, None, CpuAVX512F, Modrm|MaskingMorZ|Space0F38|VexW=1|Disp8MemShift=2|CheckRegSize|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegXMM|RegYMM|RegZMM, RegXMM|RegYMM|RegZMM|Unspecified|BaseIndex }
>
> -vpscatterdq, 0x66A0, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB256|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
> -vpscatterqq, 0x66A1, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
> -vscatterdpd, 0x66A2, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB256|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
> -vscatterqpd, 0x66A3, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
> +vpscatterdq, 0x66A0, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB256|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Qword|Unspecified|BaseIndex }
> +vpscatterqq, 0x66A1, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Qword|Unspecified|BaseIndex }
> +vscatterdpd, 0x66A2, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB256|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Qword|Unspecified|BaseIndex }
> +vscatterqpd, 0x66A3, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW1|Disp8MemShift=3|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Qword|Unspecified|BaseIndex }
>
> vpscatterdd, 0x66A0, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW0|Disp8MemShift=2|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
> vscatterdps, 0x66A2, None, CpuAVX512F, Modrm|EVex=1|Masking=2|NoDefMask|Space0F38|VexW0|Disp8MemShift=2|VecSIB512|No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, { RegZMM, Dword|Unspecified|BaseIndex }
--
H.J.
More information about the Binutils
mailing list