https://llvm.org/bugs/show_bug.cgi?id=15077
Simon Pilgrim <llvm-...@redking.me.uk> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |FIXED
--- Comment #4 from Simon Pilgrim <llvm-...@redking.me.uk> ---
Resolving this. With rL284939 we have accurate vector shift costs for SSE4.1+
which means that we can correctly vectorize the inner loop as 4 * <4 x i32>:
LBB0_8:
vpmovsxbd -12(%ebx), %xmm5
vmovdqu -48(%edi), %xmm3
vmovdqu -32(%edi), %xmm4
vmovdqu -16(%edi), %xmm2
vmovdqu (%edi), %xmm1
vpsrldq $12, %xmm5, %xmm6 # xmm6 =
xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
vpsrlq $32, %xmm5, %xmm7
vpsrld %xmm6, %xmm3, %xmm6
vpsrld %xmm7, %xmm3, %xmm7
vpblendw $240, %xmm6, %xmm7, %xmm6 # xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
vpunpckhdq %xmm0, %xmm5, %xmm7 # xmm7 = xmm5[2],xmm0[2],xmm5[3],xmm0[3]
vpmovzxdq %xmm5, %xmm5 # xmm5 = xmm5[0],zero,xmm5[1],zero
vpsrld %xmm7, %xmm3, %xmm7
vpsrld %xmm5, %xmm3, %xmm3
vpmovsxbd -8(%ebx), %xmm5
vpblendw $240, %xmm7, %xmm3, %xmm3 # xmm3 = xmm3[0,1,2,3],xmm7[4,5,6,7]
vpblendw $204, %xmm6, %xmm3, %xmm3 # xmm3 =
xmm3[0,1],xmm6[2,3],xmm3[4,5],xmm6[6,7]
vpsrldq $12, %xmm5, %xmm6 # xmm6 =
xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
vpsrlq $32, %xmm5, %xmm7
vpsrld %xmm6, %xmm4, %xmm6
vpsrld %xmm7, %xmm4, %xmm7
vpblendw $240, %xmm6, %xmm7, %xmm6 # xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
vpunpckhdq %xmm0, %xmm5, %xmm7 # xmm7 = xmm5[2],xmm0[2],xmm5[3],xmm0[3]
vpmovzxdq %xmm5, %xmm5 # xmm5 = xmm5[0],zero,xmm5[1],zero
vpsrld %xmm7, %xmm4, %xmm7
vpsrld %xmm5, %xmm4, %xmm4
vpmovsxbd -4(%ebx), %xmm5
vpblendw $240, %xmm7, %xmm4, %xmm4 # xmm4 = xmm4[0,1,2,3],xmm7[4,5,6,7]
vpblendw $204, %xmm6, %xmm4, %xmm4 # xmm4 =
xmm4[0,1],xmm6[2,3],xmm4[4,5],xmm6[6,7]
vpsrldq $12, %xmm5, %xmm6 # xmm6 =
xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
vpsrlq $32, %xmm5, %xmm7
vpsrld %xmm6, %xmm2, %xmm6
vpsrld %xmm7, %xmm2, %xmm7
vpblendw $240, %xmm6, %xmm7, %xmm6 # xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
vpunpckhdq %xmm0, %xmm5, %xmm7 # xmm7 = xmm5[2],xmm0[2],xmm5[3],xmm0[3]
vpmovzxdq %xmm5, %xmm5 # xmm5 = xmm5[0],zero,xmm5[1],zero
vpsrld %xmm7, %xmm2, %xmm7
vpsrld %xmm5, %xmm2, %xmm2
vpmovsxbd (%ebx), %xmm5
vmovdqu %xmm3, -48(%edi)
vmovdqu %xmm4, -32(%edi)
addl $16, %ebx
vpblendw $240, %xmm7, %xmm2, %xmm2 # xmm2 = xmm2[0,1,2,3],xmm7[4,5,6,7]
vpblendw $204, %xmm6, %xmm2, %xmm2 # xmm2 =
xmm2[0,1],xmm6[2,3],xmm2[4,5],xmm6[6,7]
vmovdqu %xmm2, -16(%edi)
vpsrldq $12, %xmm5, %xmm6 # xmm6 =
xmm5[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
vpsrlq $32, %xmm5, %xmm7
vpsrld %xmm6, %xmm1, %xmm6
vpsrld %xmm7, %xmm1, %xmm7
vpblendw $240, %xmm6, %xmm7, %xmm6 # xmm6 = xmm7[0,1,2,3],xmm6[4,5,6,7]
vpunpckhdq %xmm0, %xmm5, %xmm7 # xmm7 = xmm5[2],xmm0[2],xmm5[3],xmm0[3]
vpmovzxdq %xmm5, %xmm5 # xmm5 = xmm5[0],zero,xmm5[1],zero
vpsrld %xmm7, %xmm1, %xmm7
vpsrld %xmm5, %xmm1, %xmm1
vpblendw $240, %xmm7, %xmm1, %xmm1 # xmm1 = xmm1[0,1,2,3],xmm7[4,5,6,7]
vpblendw $204, %xmm6, %xmm1, %xmm1 # xmm1 =
xmm1[0,1],xmm6[2,3],xmm1[4,5],xmm6[6,7]
vmovdqu %xmm1, (%edi)
addl $64, %edi
addl $-16, %ebp
jne LBB0_8
movl %esi, %edi
cmpl %eax, %esi
je LBB0_10
--
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
