Skip to content

Commit

Permalink
Work around clang bug 36144 by replacing anonymous label numbers.
Browse files Browse the repository at this point in the history
  • Loading branch information
sneves committed Feb 13, 2020
1 parent 9856372 commit 2450ce1
Show file tree
Hide file tree
Showing 6 changed files with 108 additions and 108 deletions.
28 changes: 14 additions & 14 deletions blake3_avx2_x86-64_unix.S
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ blake3_hash_many_avx2:
or eax, ebx
xor edx, edx
.p2align 5
1:
9:
movzx ebx, byte ptr [rbp+0x48]
or ebx, eax
add rdx, 64
Expand Down Expand Up @@ -1231,7 +1231,7 @@ blake3_hash_many_avx2:
vpxor ymm6, ymm6, ymm14
vpxor ymm7, ymm7, ymm15
movzx eax, byte ptr [rbp+0x38]
jne 1b
jne 9b
mov rbx, qword ptr [rbp+0x50]
vunpcklps ymm8, ymm0, ymm1
vunpcklps ymm9, ymm2, ymm3
Expand Down Expand Up @@ -1374,7 +1374,7 @@ blake3_hash_many_avx2:
vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
vmovdqa ymm10, ymm2
mov al, 7
1:
9:
vpaddd ymm0, ymm0, ymm4
vpaddd ymm8, ymm8, ymm12
vmovdqa ymmword ptr [rsp+0x40], ymm4
Expand Down Expand Up @@ -1470,7 +1470,7 @@ blake3_hash_many_avx2:
vpshufd ymm2, ymm2, 0x93
vpshufd ymm10, ymm10, 0x93
dec al
je 1f
je 9f
vmovdqa ymm4, ymmword ptr [rsp+0x40]
vmovdqa ymm5, ymmword ptr [rsp+0x80]
vshufps ymm12, ymm4, ymm5, 214
Expand Down Expand Up @@ -1503,8 +1503,8 @@ blake3_hash_many_avx2:
vmovdqa ymm14, ymm5
vmovdqa ymm5, ymmword ptr [rsp+0x40]
vmovdqa ymm6, ymmword ptr [rsp+0x80]
jmp 1b
1:
jmp 9b
9:
vpxor ymm0, ymm0, ymm2
vpxor ymm1, ymm1, ymm3
vpxor ymm8, ymm8, ymm10
Expand Down Expand Up @@ -1577,7 +1577,7 @@ blake3_hash_many_avx2:
vpshufd ymm6, ymm6, 0x93
vpshufd ymm7, ymm7, 0x93
mov al, 7
1:
9:
vpaddd ymm0, ymm0, ymm4
vpaddd ymm0, ymm0, ymm1
vpxor ymm3, ymm3, ymm0
Expand Down Expand Up @@ -1621,7 +1621,7 @@ blake3_hash_many_avx2:
vpshufd ymm3, ymm3, 0x4E
vpshufd ymm2, ymm2, 0x93
dec al
jz 1f
jz 9f
vshufps ymm8, ymm4, ymm5, 214
vpshufd ymm9, ymm4, 0x0F
vpshufd ymm4, ymm8, 0x39
Expand All @@ -1635,8 +1635,8 @@ blake3_hash_many_avx2:
vpshufd ymm7, ymm6, 0x1E
vmovdqa ymm5, ymm9
vmovdqa ymm6, ymm8
jmp 1b
1:
jmp 9b
9:
vpxor ymm0, ymm0, ymm2
vpxor ymm1, ymm1, ymm3
mov eax, r13d
Expand Down Expand Up @@ -1693,7 +1693,7 @@ blake3_hash_many_avx2:
vpshufd xmm6, xmm6, 0x93
vpshufd xmm7, xmm7, 0x93
mov al, 7
1:
9:
vpaddd xmm0, xmm0, xmm4
vpaddd xmm0, xmm0, xmm1
vpxor xmm3, xmm3, xmm0
Expand Down Expand Up @@ -1737,7 +1737,7 @@ blake3_hash_many_avx2:
vpshufd xmm3, xmm3, 0x4E
vpshufd xmm2, xmm2, 0x93
dec al
jz 1f
jz 9f
vshufps xmm8, xmm4, xmm5, 214
vpshufd xmm9, xmm4, 0x0F
vpshufd xmm4, xmm8, 0x39
Expand All @@ -1751,8 +1751,8 @@ blake3_hash_many_avx2:
vpshufd xmm7, xmm6, 0x1E
vmovdqa xmm5, xmm9
vmovdqa xmm6, xmm8
jmp 1b
1:
jmp 9b
9:
vpxor xmm0, xmm0, xmm2
vpxor xmm1, xmm1, xmm3
mov eax, r13d
Expand Down
28 changes: 14 additions & 14 deletions blake3_avx2_x86-64_windows_gnu.S
Original file line number Diff line number Diff line change
Expand Up @@ -77,7 +77,7 @@ blake3_hash_many_avx2:
or eax, ebx
xor edx, edx
.p2align 5
1:
9:
movzx ebx, byte ptr [rbp+0x88]
or ebx, eax
add rdx, 64
Expand Down Expand Up @@ -1245,7 +1245,7 @@ blake3_hash_many_avx2:
vpxor ymm6, ymm6, ymm14
vpxor ymm7, ymm7, ymm15
movzx eax, byte ptr [rbp+0x78]
jne 1b
jne 9b
mov rbx, qword ptr [rbp+0x90]
vunpcklps ymm8, ymm0, ymm1
vunpcklps ymm9, ymm2, ymm3
Expand Down Expand Up @@ -1396,7 +1396,7 @@ blake3_hash_many_avx2:
vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
vmovdqa ymm10, ymm2
mov al, 7
1:
9:
vpaddd ymm0, ymm0, ymm4
vpaddd ymm8, ymm8, ymm12
vmovdqa ymmword ptr [rsp+0x40], ymm4
Expand Down Expand Up @@ -1492,7 +1492,7 @@ blake3_hash_many_avx2:
vpshufd ymm2, ymm2, 0x93
vpshufd ymm10, ymm10, 0x93
dec al
je 1f
je 9f
vmovdqa ymm4, ymmword ptr [rsp+0x40]
vmovdqa ymm5, ymmword ptr [rsp+0x80]
vshufps ymm12, ymm4, ymm5, 214
Expand Down Expand Up @@ -1525,8 +1525,8 @@ blake3_hash_many_avx2:
vmovdqa ymm14, ymm5
vmovdqa ymm5, ymmword ptr [rsp+0x40]
vmovdqa ymm6, ymmword ptr [rsp+0x80]
jmp 1b
1:
jmp 9b
9:
vpxor ymm0, ymm0, ymm2
vpxor ymm1, ymm1, ymm3
vpxor ymm8, ymm8, ymm10
Expand Down Expand Up @@ -1599,7 +1599,7 @@ blake3_hash_many_avx2:
vpshufd ymm6, ymm6, 0x93
vpshufd ymm7, ymm7, 0x93
mov al, 7
1:
9:
vpaddd ymm0, ymm0, ymm4
vpaddd ymm0, ymm0, ymm1
vpxor ymm3, ymm3, ymm0
Expand Down Expand Up @@ -1643,7 +1643,7 @@ blake3_hash_many_avx2:
vpshufd ymm3, ymm3, 0x4E
vpshufd ymm2, ymm2, 0x93
dec al
jz 1f
jz 9f
vshufps ymm8, ymm4, ymm5, 214
vpshufd ymm9, ymm4, 0x0F
vpshufd ymm4, ymm8, 0x39
Expand All @@ -1657,8 +1657,8 @@ blake3_hash_many_avx2:
vpshufd ymm7, ymm6, 0x1E
vmovdqa ymm5, ymm9
vmovdqa ymm6, ymm8
jmp 1b
1:
jmp 9b
9:
vpxor ymm0, ymm0, ymm2
vpxor ymm1, ymm1, ymm3
mov eax, r13d
Expand Down Expand Up @@ -1715,7 +1715,7 @@ blake3_hash_many_avx2:
vpshufd xmm6, xmm6, 0x93
vpshufd xmm7, xmm7, 0x93
mov al, 7
1:
9:
vpaddd xmm0, xmm0, xmm4
vpaddd xmm0, xmm0, xmm1
vpxor xmm3, xmm3, xmm0
Expand Down Expand Up @@ -1759,7 +1759,7 @@ blake3_hash_many_avx2:
vpshufd xmm3, xmm3, 0x4E
vpshufd xmm2, xmm2, 0x93
dec al
jz 1f
jz 9f
vshufps xmm8, xmm4, xmm5, 214
vpshufd xmm9, xmm4, 0x0F
vpshufd xmm4, xmm8, 0x39
Expand All @@ -1773,8 +1773,8 @@ blake3_hash_many_avx2:
vpshufd xmm7, xmm6, 0x1E
vmovdqa xmm5, xmm9
vmovdqa xmm6, xmm8
jmp 1b
1:
jmp 9b
9:
vpxor xmm0, xmm0, xmm2
vpxor xmm1, xmm1, xmm3
mov eax, r13d
Expand Down
44 changes: 22 additions & 22 deletions blake3_avx512_x86-64_unix.S
Original file line number Diff line number Diff line change
Expand Up @@ -66,7 +66,7 @@ blake3_hash_many_avx512:
or eax, ebx
xor edx, edx
.p2align 5
1:
9:
movzx ebx, byte ptr [rbp+0x48]
or ebx, eax
add rdx, 64
Expand Down Expand Up @@ -1011,7 +1011,7 @@ blake3_hash_many_avx512:
vpxord zmm6, zmm6, zmm14
vpxord zmm7, zmm7, zmm15
movzx eax, byte ptr [rbp+0x38]
jne 1b
jne 9b
mov rbx, qword ptr [rbp+0x50]
vpunpckldq zmm16, zmm0, zmm1
vpunpckhdq zmm17, zmm0, zmm1
Expand Down Expand Up @@ -2089,7 +2089,7 @@ blake3_hash_many_avx512:
vpshufd zmm6, zmm6, 0x93
vpshufd zmm7, zmm7, 0x93
mov al, 7
1:
9:
vpaddd zmm0, zmm0, zmm4
vpaddd zmm0, zmm0, zmm1
vpxord zmm3, zmm3, zmm0
Expand Down Expand Up @@ -2125,7 +2125,7 @@ blake3_hash_many_avx512:
vpshufd zmm3, zmm3, 0x4E
vpshufd zmm2, zmm2, 0x93
dec al
jz 1f
jz 9f
vshufps zmm8, zmm4, zmm5, 214
vpshufd zmm9, zmm4, 0x0F
vpshufd zmm4, zmm8, 0x39
Expand All @@ -2139,8 +2139,8 @@ blake3_hash_many_avx512:
vpshufd zmm7, zmm6, 0x1E
vmovdqa32 zmm5, zmm9
vmovdqa32 zmm6, zmm8
jmp 1b
1:
jmp 9b
9:
vpxord zmm0, zmm0, zmm2
vpxord zmm1, zmm1, zmm3
mov eax, r13d
Expand Down Expand Up @@ -2206,7 +2206,7 @@ blake3_hash_many_avx512:
vpshufd ymm6, ymm6, 0x93
vpshufd ymm7, ymm7, 0x93
mov al, 7
1:
9:
vpaddd ymm0, ymm0, ymm4
vpaddd ymm0, ymm0, ymm1
vpxord ymm3, ymm3, ymm0
Expand Down Expand Up @@ -2242,7 +2242,7 @@ blake3_hash_many_avx512:
vpshufd ymm3, ymm3, 0x4E
vpshufd ymm2, ymm2, 0x93
dec al
jz 1f
jz 9f
vshufps ymm8, ymm4, ymm5, 214
vpshufd ymm9, ymm4, 0x0F
vpshufd ymm4, ymm8, 0x39
Expand All @@ -2256,8 +2256,8 @@ blake3_hash_many_avx512:
vpshufd ymm7, ymm6, 0x1E
vmovdqa ymm5, ymm9
vmovdqa ymm6, ymm8
jmp 1b
1:
jmp 9b
9:
vpxor ymm0, ymm0, ymm2
vpxor ymm1, ymm1, ymm3
mov eax, r13d
Expand Down Expand Up @@ -2309,7 +2309,7 @@ blake3_hash_many_avx512:
vpshufd xmm6, xmm6, 0x93
vpshufd xmm7, xmm7, 0x93
mov al, 7
1:
9:
vpaddd xmm0, xmm0, xmm4
vpaddd xmm0, xmm0, xmm1
vpxord xmm3, xmm3, xmm0
Expand Down Expand Up @@ -2345,7 +2345,7 @@ blake3_hash_many_avx512:
vpshufd xmm3, xmm3, 0x4E
vpshufd xmm2, xmm2, 0x93
dec al
jz 1f
jz 9f
vshufps xmm8, xmm4, xmm5, 214
vpshufd xmm9, xmm4, 0x0F
vpshufd xmm4, xmm8, 0x39
Expand All @@ -2359,8 +2359,8 @@ blake3_hash_many_avx512:
vpshufd xmm7, xmm6, 0x1E
vmovdqa xmm5, xmm9
vmovdqa xmm6, xmm8
jmp 1b
1:
jmp 9b
9:
vpxor xmm0, xmm0, xmm2
vpxor xmm1, xmm1, xmm3
mov eax, r13d
Expand Down Expand Up @@ -2393,7 +2393,7 @@ blake3_compress_in_place_avx512:
vpshufd xmm6, xmm6, 0x93
vpshufd xmm7, xmm7, 0x93
mov al, 7
1:
9:
vpaddd xmm0, xmm0, xmm4
vpaddd xmm0, xmm0, xmm1
vpxord xmm3, xmm3, xmm0
Expand Down Expand Up @@ -2429,7 +2429,7 @@ blake3_compress_in_place_avx512:
vpshufd xmm3, xmm3, 0x4E
vpshufd xmm2, xmm2, 0x93
dec al
jz 1f
jz 9f
vshufps xmm8, xmm4, xmm5, 214
vpshufd xmm9, xmm4, 0x0F
vpshufd xmm4, xmm8, 0x39
Expand All @@ -2443,8 +2443,8 @@ blake3_compress_in_place_avx512:
vpshufd xmm7, xmm6, 0x1E
vmovdqa xmm5, xmm9
vmovdqa xmm6, xmm8
jmp 1b
1:
jmp 9b
9:
vpxor xmm0, xmm0, xmm2
vpxor xmm1, xmm1, xmm3
vmovdqu xmmword ptr [rdi], xmm0
Expand Down Expand Up @@ -2475,7 +2475,7 @@ blake3_compress_xof_avx512:
vpshufd xmm6, xmm6, 0x93
vpshufd xmm7, xmm7, 0x93
mov al, 7
1:
9:
vpaddd xmm0, xmm0, xmm4
vpaddd xmm0, xmm0, xmm1
vpxord xmm3, xmm3, xmm0
Expand Down Expand Up @@ -2511,7 +2511,7 @@ blake3_compress_xof_avx512:
vpshufd xmm3, xmm3, 0x4E
vpshufd xmm2, xmm2, 0x93
dec al
jz 1f
jz 9f
vshufps xmm8, xmm4, xmm5, 214
vpshufd xmm9, xmm4, 0x0F
vpshufd xmm4, xmm8, 0x39
Expand All @@ -2525,8 +2525,8 @@ blake3_compress_xof_avx512:
vpshufd xmm7, xmm6, 0x1E
vmovdqa xmm5, xmm9
vmovdqa xmm6, xmm8
jmp 1b
1:
jmp 9b
9:
vpxor xmm0, xmm0, xmm2
vpxor xmm1, xmm1, xmm3
vpxor xmm2, xmm2, [rdi]
Expand Down
Loading

0 comments on commit 2450ce1

Please sign in to comment.