https://llvm.org/bugs/show_bug.cgi?id=26667
Bug ID: 26667
Summary: [Regression] Wrong codegen for AVX permute intrinsic
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Backend: X86
Assignee: unassignedb...@nondot.org
Reporter: yunzhong_...@playstation.sony.com
CC: llvm-bugs@lists.llvm.org, llvm-...@redking.me.uk
Classification: Unclassified
// test.c
typedef double __m256d __attribute__((__vector_size__(32)));
typedef double __v4df __attribute__ ((__vector_size__ (32)));
#define _mm256_permute2f128_pd(V1, V2, M) \
(__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)(V1), (__v4df)(V2), (M))
__m256d foo(__m256d Lhs, __m256d Rhs) {
return _mm256_permute2f128_pd(Lhs, Rhs, 50);
}
// end test.c
$ clang -O0 -S -mavx test.c -o -
Here is the assembly sequence produced by r260062:
andq $-32, %rsp
subq $96, %rsp
vmovapd %ymm0, 32(%rsp)
vmovapd %ymm1, (%rsp)
vmovapd 32(%rsp), %ymm0
# vperm2f128 $src3, $src2, $src1, $dst
# with mask=$50, $dst = $src2, so in this case, ymm0 <= ymm1
vperm2f128 $50, %ymm1, %ymm0, %ymm0 # ymm0 = ymm1[0,1,2,3]
And here is the assembly sequence produced by r260063 and later
andq $-32, %rsp
subq $128, %rsp
vmovapd %ymm0, 64(%rsp)
vmovapd %ymm1, 32(%rsp)
vmovapd 64(%rsp), %ymm0
vmovaps %ymm1, (%rsp) # 32-byte Spill
# and in this case, ymm0 is still ymm0
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
