https://llvm.org/bugs/show_bug.cgi?id=30986
Bug ID: 30986
Summary: [X86][SSE] Failure to split vector loads for
scalarized operations
Product: libraries
Version: trunk
Hardware: PC
OS: Windows NT
Status: NEW
Severity: normal
Priority: P
Component: Backend: X86
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected], [email protected],
[email protected], [email protected]
Classification: Unclassified
As discussed on https://reviews.llvm.org/D26521.
Scalarized cases fail to split any vector loads, resulting in a lot of extra
(potentially very slow) vector -> gpr traffic:
clang -S -O3 -march=btver2
#include <x86intrin.h>
__m128i popcnt1(__m128i *in) {
return (__m128i) {
__builtin_popcountll(in[0][0]),
__builtin_popcountll(in[0][1]) };
}
popcnt1(long long __vector(2)*):
vmovdqu (%rdi), %xmm0
vmovq %xmm0, %rax
vpextrq $1, %xmm0, %rcx
popcntq %rax, %rax
popcntq %rcx, %rcx
vmovq %rcx, %xmm0
vmovq %rax, %xmm1
vpunpcklqdq %xmm0, %xmm1, %xmm0 # xmm0 = xmm1[0],xmm0[0]
retq
It would be better as:
popcnt1(long long __vector(2)*):
popcntq (%rdi), %rax
popcntq 8(%rdi), %rcx
vmovq %rcx, %xmm0
vmovq %rax, %xmm1
vpunpcklqdq %xmm0, %xmm1, %xmm0 # xmm0 = xmm1[0],xmm0[0]
retq
Something similar happens when the source vector is spilled - it is first
restored to a vector register and then transferred to gprs. In some cases after
the results are transferred back to the vector register and then spilled again
- we could have spilled the scalar values in vector order directly...
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
[email protected]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
