<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - Folding a comparison to use pre-extended value results increases instructions"
href="https://bugs.llvm.org/show_bug.cgi?id=50055">50055</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Folding a comparison to use pre-extended value results increases instructions
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>llvm-dev@redking.me.uk
</td>
</tr>
<tr>
<th>CC</th>
<td>lebedev.ri@gmail.com, llvm-bugs@lists.llvm.org, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre><a href="https://wide.godbolt.org/z/3ob3M4qfe">https://wide.godbolt.org/z/3ob3M4qfe</a>
#include <x86intrin.h>
__m256i foo(const __m128i *src, __m256i *dst) {
__m256i ext = _mm256_cvtepu8_epi32( _mm_loadl_epi64( src ) );
*dst = _mm256_cmpgt_epi32( ext, _mm256_setzero_si256() );
return ext;
}
By performing the comparison on the pre-extended source value we end up with an
additional extension, and we lose the folded load.
define <4 x i64> @foo(<2 x i64>* %0, <4 x i64>* %1) {
%3 = getelementptr <2 x i64>, <2 x i64>* %0, i64 0, i64 0
%4 = load i64, i64* %3, align 1
%5 = insertelement <2 x i64> poison, i64 %4, i32 0
%6 = bitcast <2 x i64> %5 to <16 x i8>
%7 = shufflevector <16 x i8> %6, <16 x i8> undef, <8 x i32> <i32 0, i32 1,
i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%8 = zext <8 x i8> %7 to <8 x i32>
%9 = bitcast <8 x i32> %8 to <4 x i64>
%10 = icmp ne <8 x i8> %7, zeroinitializer
%11 = sext <8 x i1> %10 to <8 x i32>
%12 = bitcast <4 x i64>* %1 to <8 x i32>*
store <8 x i32> %11, <8 x i32>* %12, align 32
ret <4 x i64> %9
}
clang -O3 -march=haswell
foo:
vmovq (%rdi), %xmm1
vpmovzxbd %xmm1, %ymm0
vpxor %xmm2, %xmm2, %xmm2
vpcmpeqb %xmm2, %xmm1, %xmm1
vpcmpeqd %xmm2, %xmm2, %xmm2
vpxor %xmm2, %xmm1, %xmm1
vpmovsxbd %xmm1, %ymm1
vmovdqa %ymm1, (%rsi)
retq
gcc -O3 -march=haswell
foo:
vpmovzxbd (%rdi), %ymm0
vpxor %xmm1, %xmm1, %xmm1
vpcmpgtd %ymm1, %ymm0, %ymm1
vmovdqa %ymm1, (%rsi)
ret</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>