<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 - Missed vector code optimizations - res += X vs res = res + X"
href="https://bugs.llvm.org/show_bug.cgi?id=47439">47439</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Missed vector code optimizations - res += X vs res = res + X
</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>Linux
</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>david.bolvansky@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>typedef float Value;
struct Vec
{
Vec(Value x=0, Value y=0, Value z=0, Value t=0) :
X(x),Y(y),Z(z),T(t){}
Vec & operator+=(const Vec & a) {
X += a.X;
Y += a.Y;
Z += a.Z;
T += a.T;
return *this;
}
Value X;
Value Y;
Value Z;
Value T;
};
inline Vec
operator+(Vec const & a, Vec const & b) {
return
Vec(a.X+b.X,a.Y+b.Y,a.Z+b.Z,a.T+b.T);
}
inline Vec
operator*(Vec const & a, Value s) {
return Vec(a.X*s,a.Y*s,a.Z*s,a.T*s);
}
inline Vec
operator*(Value s, Vec const & a) {
return a*s;
}
void foo(Vec & res, Value s, Vec const & v1, Vec
const & v2) {
res += s*(v1+v2);
}
void bar(Vec & res, Value s, Vec const & v1, Vec
const & v2) {
res = res + s*(v1+v2);
}
Clang -Ofast:
foo(Vec&, float, Vec const&, Vec const&): # @foo(Vec&, float, Vec const&, Vec
const&)
movups xmm1, xmmword ptr [rsi]
movups xmm2, xmmword ptr [rdx]
addps xmm2, xmm1
shufps xmm0, xmm0, 0 # xmm0 = xmm0[0,0,0,0]
mulps xmm0, xmm2
movups xmm1, xmmword ptr [rdi]
addps xmm1, xmm0
movups xmmword ptr [rdi], xmm1
ret
bar(Vec&, float, Vec const&, Vec const&): # @bar(Vec&, float, Vec const&, Vec
const&)
movsd xmm1, qword ptr [rsi] # xmm1 = mem[0],zero
movsd xmm2, qword ptr [rsi + 8] # xmm2 = mem[0],zero
movsd xmm3, qword ptr [rdx] # xmm3 = mem[0],zero
addps xmm3, xmm1
movsd xmm1, qword ptr [rdx + 8] # xmm1 = mem[0],zero
addps xmm1, xmm2
shufps xmm0, xmm0, 0 # xmm0 = xmm0[0,0,0,0]
mulps xmm3, xmm0
mulps xmm0, xmm1
movsd xmm1, qword ptr [rdi] # xmm1 = mem[0],zero
addps xmm1, xmm3
movsd xmm2, qword ptr [rdi + 8] # xmm2 = mem[0],zero
addps xmm2, xmm0
movlps qword ptr [rdi], xmm1
movlps qword ptr [rdi + 8], xmm2
ret
GCC -Ofast:
foo(Vec&, float, Vec const&, Vec const&):
movups xmm1, XMMWORD PTR [rsi]
movups xmm2, XMMWORD PTR [rdx]
shufps xmm0, xmm0, 0
movups xmm3, XMMWORD PTR [rdi]
addps xmm1, xmm2
mulps xmm1, xmm0
addps xmm1, xmm3
movups XMMWORD PTR [rdi], xmm1
ret
bar(Vec&, float, Vec const&, Vec const&):
movups xmm1, XMMWORD PTR [rsi]
movups xmm2, XMMWORD PTR [rdx]
shufps xmm0, xmm0, 0
movups xmm3, XMMWORD PTR [rdi]
addps xmm1, xmm2
mulps xmm1, xmm0
addps xmm1, xmm3
movups XMMWORD PTR [rdi], xmm1
ret
Function bar should be optimized better and match function foo.
<a href="https://godbolt.org/z/6q3h16">https://godbolt.org/z/6q3h16</a></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>