[llvm-bugs] [Bug 30316] New: [ppc] slow instructions caused by unnecessary st/ld forwarding
via llvm-bugs
llvm-bugs at lists.llvm.org
Wed Sep 7 14:31:44 PDT 2016
https://llvm.org/bugs/show_bug.cgi?id=30316
Bug ID: 30316
Summary: [ppc] slow instructions caused by unnecessary st/ld
forwarding
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Backend: PowerPC
Assignee: unassignedbugs at nondot.org
Reporter: carrot at google.com
CC: llvm-bugs at lists.llvm.org
Classification: Unclassified
The source code is
class C{
public:
C();
bool Next() {
++current;
return !Done();
}
bool Done() const { return current >= num; }
int val() const { return vals[current];}
int* vals;
int current;
int num;
};
int bar() {
C c;
int s = 0;
while (c.Next())
s |= c.val();
return s;
}
While compiled with options
--target=powerpc64le-grtev4-linux-gnu -m64 -O2 -mvsx -mcpu=power8
The while loop is translated to:
.LBB0_2: # %while.body
# =>This Inner Loop Header: Depth=1
extsw 7, 3
addi 3, 4, 1 // ++current
sldi 11, 7, 2
lwzx 7, 6, 11
stw 3, 112(31) // store current
ori 2, 2, 0
ld 4, 112(31) // load both current and num
or 5, 7, 5
rldicl 12, 4, 32, 32 // extract num
cmpw 0, 4, 12
blt 0, .LBB0_2
The interesting part is the code from c.Next(), after the variable current
stored to memory, it is immediately loaded together with another var num in one
load instruction. It triggers the slow store forwarding, or even slower than
store forwarding, since only partial value of the load is in store queue,
another part of the load is in cache.
Actually the value current is already in register 3, it doesn't need to be
loaded again.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160907/fe526103/attachment.html>
More information about the llvm-bugs
mailing list