2014-12-16 20:15:45 +00:00
; RUN: llc < %s -verify-machineinstrs -march=x86-64 -asm-verbose=false -mtriple=x86_64-unknown-linux-gnu -mcpu=nehalem -post-RA-scheduler=true -schedmodel=false | FileCheck %s
2009-09-25 23:58:45 +00:00
; Currently, floating-point selects are lowered to CFG triangles.
; This means that one side of the select is always unconditionally
; evaluated, however with MachineSink we can sink the other side so
; that it's conditionally evaluated.
2013-07-14 06:24:09 +00:00
; CHECK-LABEL: foo:
2009-10-18 19:57:27 +00:00
; CHECK-NEXT: testb $1, %dil
2012-04-16 13:49:17 +00:00
; CHECK-NEXT: jne
2009-10-18 19:57:27 +00:00
; CHECK-NEXT: divsd
2012-04-16 13:49:17 +00:00
; CHECK-NEXT: movaps
Enable machine sinking critical edge splitting. e.g.
define double @foo(double %x, double %y, i1 %c) nounwind {
%a = fdiv double %x, 3.2
%z = select i1 %c, double %a, double %y
ret double %z
}
Was:
_foo:
divsd LCPI0_0(%rip), %xmm0
testb $1, %dil
jne LBB0_2
movaps %xmm1, %xmm0
LBB0_2:
ret
Now:
_foo:
testb $1, %dil
je LBB0_2
divsd LCPI0_0(%rip), %xmm0
ret
LBB0_2:
movaps %xmm1, %xmm0
ret
This avoids the divsd when early exit is taken.
rdar://8454886
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114372 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-20 22:52:00 +00:00
; CHECK-NEXT: ret
; CHECK: divsd
2009-09-25 23:58:45 +00:00
define double @foo ( double %x , double %y , i1 %c ) nounwind {
%a = fdiv double %x , 3.2
%b = fdiv double %y , 3.3
%z = select i1 %c , double %a , double %b
ret double %z
}
Enable machine sinking critical edge splitting. e.g.
define double @foo(double %x, double %y, i1 %c) nounwind {
%a = fdiv double %x, 3.2
%z = select i1 %c, double %a, double %y
ret double %z
}
Was:
_foo:
divsd LCPI0_0(%rip), %xmm0
testb $1, %dil
jne LBB0_2
movaps %xmm1, %xmm0
LBB0_2:
ret
Now:
_foo:
testb $1, %dil
je LBB0_2
divsd LCPI0_0(%rip), %xmm0
ret
LBB0_2:
movaps %xmm1, %xmm0
ret
This avoids the divsd when early exit is taken.
rdar://8454886
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114372 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-20 22:52:00 +00:00
; Make sure the critical edge is broken so the divsd is sunken below
; the conditional branch.
; rdar://8454886
2013-07-14 06:24:09 +00:00
; CHECK-LABEL: split:
Enable machine sinking critical edge splitting. e.g.
define double @foo(double %x, double %y, i1 %c) nounwind {
%a = fdiv double %x, 3.2
%z = select i1 %c, double %a, double %y
ret double %z
}
Was:
_foo:
divsd LCPI0_0(%rip), %xmm0
testb $1, %dil
jne LBB0_2
movaps %xmm1, %xmm0
LBB0_2:
ret
Now:
_foo:
testb $1, %dil
je LBB0_2
divsd LCPI0_0(%rip), %xmm0
ret
LBB0_2:
movaps %xmm1, %xmm0
ret
This avoids the divsd when early exit is taken.
rdar://8454886
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114372 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-20 22:52:00 +00:00
; CHECK-NEXT: testb $1, %dil
2013-10-15 23:33:07 +00:00
; CHECK-NEXT: je
2012-04-16 13:49:17 +00:00
; CHECK: divsd
2013-10-15 23:33:07 +00:00
; CHECK: movaps
; CHECK: ret
Enable machine sinking critical edge splitting. e.g.
define double @foo(double %x, double %y, i1 %c) nounwind {
%a = fdiv double %x, 3.2
%z = select i1 %c, double %a, double %y
ret double %z
}
Was:
_foo:
divsd LCPI0_0(%rip), %xmm0
testb $1, %dil
jne LBB0_2
movaps %xmm1, %xmm0
LBB0_2:
ret
Now:
_foo:
testb $1, %dil
je LBB0_2
divsd LCPI0_0(%rip), %xmm0
ret
LBB0_2:
movaps %xmm1, %xmm0
ret
This avoids the divsd when early exit is taken.
rdar://8454886
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114372 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-20 22:52:00 +00:00
define double @split ( double %x , double %y , i1 %c ) nounwind {
%a = fdiv double %x , 3.2
%z = select i1 %c , double %a , double %y
ret double %z
}
2009-09-25 23:58:45 +00:00
; Hoist floating-point constant-pool loads out of loops.
2013-07-14 06:24:09 +00:00
; CHECK-LABEL: bar:
2009-09-25 23:58:45 +00:00
; CHECK: movsd
; CHECK: align
define void @bar ( double * nocapture %p , i64 %n ) nounwind {
entry:
%0 = icmp sgt i64 %n , 0
br i1 %0 , label %bb , label %return
bb:
%i.03 = phi i64 [ 0 , %entry ] , [ %3 , %bb ]
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
%scevgep = getelementptr double , double * %p , i64 %i.03
2015-02-27 21:17:42 +00:00
%1 = load double , double * %scevgep , align 8
2009-09-25 23:58:45 +00:00
%2 = fdiv double 3.200000e+00 , %1
store double %2 , double * %scevgep , align 8
%3 = add nsw i64 %i.03 , 1
%exitcond = icmp eq i64 %3 , %n
br i1 %exitcond , label %return , label %bb
return:
ret void
}
2009-10-28 01:13:53 +00:00
; Sink instructions with dead EFLAGS defs.
2010-06-03 07:54:20 +00:00
; FIXME: Unfail the zzz test if we can correctly mark pregs with the kill flag.
2013-10-15 23:33:07 +00:00
;
2010-06-03 07:54:20 +00:00
; See <rdar://problem/8030636>. This test isn't valid after we made machine
; sinking more conservative about sinking instructions that define a preg into a
; block when we don't know if the preg is killed within the current block.
2009-10-28 01:13:53 +00:00
2010-06-03 07:54:20 +00:00
; FIXMEHECK: zzz:
; FIXMEHECK: je
; FIXMEHECK-NEXT: orb
; define zeroext i8 @zzz(i8 zeroext %a, i8 zeroext %b) nounwind readnone {
; entry:
; %tmp = zext i8 %a to i32 ; <i32> [#uses=1]
; %tmp2 = icmp eq i8 %a, 0 ; <i1> [#uses=1]
; %tmp3 = or i8 %b, -128 ; <i8> [#uses=1]
; %tmp4 = and i8 %b, 127 ; <i8> [#uses=1]
; %b_addr.0 = select i1 %tmp2, i8 %tmp4, i8 %tmp3 ; <i8> [#uses=1]
; ret i8 %b_addr.0
; }
2009-10-28 03:21:57 +00:00
2011-10-10 23:18:02 +00:00
; Codegen should hoist and CSE these constants.
2013-07-14 06:24:09 +00:00
; CHECK-LABEL: vv:
2011-10-10 23:18:02 +00:00
; CHECK: LCPI3_0(%rip), %xmm0
; CHECK: LCPI3_1(%rip), %xmm1
; CHECK: LCPI3_2(%rip), %xmm2
; CHECK: align
; CHECK-NOT: LCPI
; CHECK: ret
@_minusZero.6007 = internal constant < 4 x float > < float -0.000000e+00 , float -0.000000e+00 , float -0.000000e+00 , float -0.000000e+00 > ; <<4 x float>*> [#uses=0]
@twoTo23.6008 = internal constant < 4 x float > < float 8.388608e+06 , float 8.388608e+06 , float 8.388608e+06 , float 8.388608e+06 > ; <<4 x float>*> [#uses=0]
define void @vv ( float * %y , float * %x , i32 * %n ) nounwind ssp {
entry:
br label %bb60
bb: ; preds = %bb60
2011-10-11 18:09:58 +00:00
%i.0 = phi i32 [ 0 , %bb60 ] ; <i32> [#uses=2]
2011-10-10 23:18:02 +00:00
%0 = bitcast float * %x_addr.0 to < 4 x float > * ; <<4 x float>*> [#uses=1]
2015-02-27 21:17:42 +00:00
%1 = load < 4 x float > , < 4 x float > * %0 , align 16 ; <<4 x float>> [#uses=4]
2011-10-10 23:18:02 +00:00
%tmp20 = bitcast < 4 x float > %1 to < 4 x i32 > ; <<4 x i32>> [#uses=1]
%tmp22 = and < 4 x i32 > %tmp20 , < i32 2147483647 , i32 2147483647 , i32 2147483647 , i32 2147483647 > ; <<4 x i32>> [#uses=1]
%tmp23 = bitcast < 4 x i32 > %tmp22 to < 4 x float > ; <<4 x float>> [#uses=1]
%tmp25 = bitcast < 4 x float > %1 to < 4 x i32 > ; <<4 x i32>> [#uses=1]
%tmp27 = and < 4 x i32 > %tmp25 , < i32 -2147483648 , i32 -2147483648 , i32 -2147483648 , i32 -2147483648 > ; <<4 x i32>> [#uses=2]
%tmp30 = call < 4 x float > @llvm.x86.sse.cmp.ps ( < 4 x float > %tmp23 , < 4 x float > < float 8.388608e+06 , float 8.388608e+06 , float 8.388608e+06 , float 8.388608e+06 > , i8 5 ) ; <<4 x float>> [#uses=1]
%tmp34 = bitcast < 4 x float > %tmp30 to < 4 x i32 > ; <<4 x i32>> [#uses=1]
%tmp36 = xor < 4 x i32 > %tmp34 , < i32 -1 , i32 -1 , i32 -1 , i32 -1 > ; <<4 x i32>> [#uses=1]
%tmp37 = and < 4 x i32 > %tmp36 , < i32 1258291200 , i32 1258291200 , i32 1258291200 , i32 1258291200 > ; <<4 x i32>> [#uses=1]
%tmp42 = or < 4 x i32 > %tmp37 , %tmp27 ; <<4 x i32>> [#uses=1]
%tmp43 = bitcast < 4 x i32 > %tmp42 to < 4 x float > ; <<4 x float>> [#uses=2]
%tmp45 = fadd < 4 x float > %1 , %tmp43 ; <<4 x float>> [#uses=1]
%tmp47 = fsub < 4 x float > %tmp45 , %tmp43 ; <<4 x float>> [#uses=2]
%tmp49 = call < 4 x float > @llvm.x86.sse.cmp.ps ( < 4 x float > %1 , < 4 x float > %tmp47 , i8 1 ) ; <<4 x float>> [#uses=1]
%2 = bitcast < 4 x float > %tmp49 to < 4 x i32 > ; <<4 x i32>> [#uses=1]
%3 = call < 4 x float > @llvm.x86.sse2.cvtdq2ps ( < 4 x i32 > %2 ) nounwind readnone ; <<4 x float>> [#uses=1]
%tmp53 = fadd < 4 x float > %tmp47 , %3 ; <<4 x float>> [#uses=1]
%tmp55 = bitcast < 4 x float > %tmp53 to < 4 x i32 > ; <<4 x i32>> [#uses=1]
%tmp57 = or < 4 x i32 > %tmp55 , %tmp27 ; <<4 x i32>> [#uses=1]
%tmp58 = bitcast < 4 x i32 > %tmp57 to < 4 x float > ; <<4 x float>> [#uses=1]
%4 = bitcast float * %y_addr.0 to < 4 x float > * ; <<4 x float>*> [#uses=1]
store < 4 x float > %tmp58 , < 4 x float > * %4 , align 16
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@230786 91177308-0d34-0410-b5e6-96231b3b80d8
2015-02-27 19:29:02 +00:00
%5 = getelementptr float , float * %x_addr.0 , i64 4 ; <float*> [#uses=1]
%6 = getelementptr float , float * %y_addr.0 , i64 4 ; <float*> [#uses=1]
2011-10-10 23:18:02 +00:00
%7 = add i32 %i.0 , 4 ; <i32> [#uses=1]
2015-02-27 21:17:42 +00:00
%8 = load i32 , i32 * %n , align 4 ; <i32> [#uses=1]
2011-10-11 18:09:58 +00:00
%9 = icmp sgt i32 %8 , %7 ; <i1> [#uses=1]
br i1 %9 , label %bb60 , label %return
2011-10-10 23:18:02 +00:00
bb60: ; preds = %bb, %entry
%x_addr.0 = phi float * [ %x , %entry ] , [ %5 , %bb ] ; <float*> [#uses=2]
%y_addr.0 = phi float * [ %y , %entry ] , [ %6 , %bb ] ; <float*> [#uses=2]
2011-10-11 18:09:58 +00:00
br label %bb
2011-10-10 23:18:02 +00:00
return: ; preds = %bb60
ret void
}
2009-10-28 03:21:57 +00:00
declare < 4 x float > @llvm.x86.sse.cmp.ps ( < 4 x float > , < 4 x float > , i8 ) nounwind readnone
declare < 4 x float > @llvm.x86.sse2.cvtdq2ps ( < 4 x i32 > ) nounwind readnone
2009-10-30 22:18:41 +00:00
; CodeGen should use the correct register class when extracting
; a load from a zero-extending load for hoisting.
2013-07-14 06:24:09 +00:00
; CHECK-LABEL: default_get_pch_validity:
2009-10-30 22:18:41 +00:00
; CHECK: movl cl_options_count(%rip), %ecx
@cl_options_count = external constant i32 ; <i32*> [#uses=2]
define void @default_get_pch_validity ( ) nounwind {
entry:
2015-02-27 21:17:42 +00:00
%tmp4 = load i32 , i32 * @cl_options_count , align 4 ; <i32> [#uses=1]
2009-10-30 22:18:41 +00:00
%tmp5 = icmp eq i32 %tmp4 , 0 ; <i1> [#uses=1]
br i1 %tmp5 , label %bb6 , label %bb2
bb2: ; preds = %bb2, %entry
%i.019 = phi i64 [ 0 , %entry ] , [ %tmp25 , %bb2 ] ; <i64> [#uses=1]
%tmp25 = add i64 %i.019 , 1 ; <i64> [#uses=2]
2015-02-27 21:17:42 +00:00
%tmp11 = load i32 , i32 * @cl_options_count , align 4 ; <i32> [#uses=1]
2009-10-30 22:18:41 +00:00
%tmp12 = zext i32 %tmp11 to i64 ; <i64> [#uses=1]
%tmp13 = icmp ugt i64 %tmp12 , %tmp25 ; <i1> [#uses=1]
br i1 %tmp13 , label %bb2 , label %bb6
bb6: ; preds = %bb2, %entry
ret void
}
