zeroshade commented on code in PR #1348:
URL: https://github.com/apache/arrow-go/pull/1348#discussion_r4199537145
##########
arrow/compute/internal/kernels/_lib/get_take_indices_avx2_amd64.s:
##########
@@ -0,0 +1,136 @@
+ .intel_syntax noprefix
+ .file "get_take_indices_avx2_amd64.cc"
+ .text
+ .globl get_take_indices_uint32_avx2
+ .p2align 4
+ .type get_take_indices_uint32_avx2,@function
+get_take_indices_uint32_avx2: # @get_take_indices_uint32_avx2
+# %bb.0:
+ push r14
+ push rbx
+ test rcx, rcx
+ jle .LBB0_10
+# %bb.1:
+ dec rcx
+ je .LBB0_2
+# %bb.11:
+ vpxor xmm0, xmm0, xmm0
+ vmovdqu xmm1, xmmword ptr [rdx + 384]
+ xor eax, eax
+ vmovdqu xmm2, xmmword ptr [rdx + 352]
+ vmovdqu xmm3, xmmword ptr [rdx + 368]
+ xor r9d, r9d
+ jmp .LBB0_12
+ .p2align 4
+.LBB0_14: # in Loop: Header=BB0_12 Depth=1
+ vpor xmm4, xmm0, xmm2
+ vpor xmm5, xmm0, xmm3
+ vmovdqu xmmword ptr [rsi + 4*rax], xmm4
+ vmovdqu xmmword ptr [rsi + 4*rax + 16], xmm5
+ add rax, 8
+.LBB0_19: # in Loop: Header=BB0_12 Depth=1
+ vpaddd xmm0, xmm0, xmm1
+ inc r9
+ cmp rcx, r9
+ je .LBB0_3
+.LBB0_12: # =>This Inner Loop Header: Depth=1
+ movzx r10d, byte ptr [rdi + r9]
+ test r10d, r10d
+ je .LBB0_19
+# %bb.13: # in Loop: Header=BB0_12 Depth=1
+ cmp r10d, 255
+ je .LBB0_14
+# %bb.15: # in Loop: Header=BB0_12 Depth=1
+ mov r11d, r10d
+ and r11d, 15
+ mov r14d, r10d
+ shr r14d, 4
+ movzx ebx, byte ptr [rdx + r11 + 336]
+ movzx r11d, byte ptr [rdx + r14 + 336]
+ test rbx, rbx
+ je .LBB0_17
+# %bb.16: # in Loop: Header=BB0_12 Depth=1
+ mov r14d, r10d
Review Comment:
Regenerating from the `.cc` puts `ebp` here (`mov ebp, r10d; shl bpl, 4`),
and `vpaddd` instead of `vpor` at L26/27/56. This file looks hand-edited.
##########
arrow/compute/internal/kernels/Makefile:
##########
@@ -76,6 +76,9 @@ _lib/scalar_comparison_sse4_amd64.s: _lib/scalar_comparison.cc
_lib/filter_uint32_avx2_amd64.s: _lib/filter_uint32.cc
$(CXX) -std=c++17 -S $(C_FLAGS) $(ASM_FLAGS_AVX2) $^ -o $@ ;
$(PERL_FIXUP_ROTATE) $@
+_lib/get_take_indices_avx2_amd64.s: _lib/get_take_indices_avx2_amd64.cc
Review Comment:
This recipe doesn't reproduce the committed `_lib` output. Please make sure
`make` round-trips to the checked-in files.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
