2013-10-31 20:14:17 +08:00
|
|
|
; Test 16-bit conditional stores that are presented as selects. The volatile
|
|
|
|
; tests require z10, which use a branch instead of a LOCR.
|
2013-06-27 17:27:40 +08:00
|
|
|
;
|
2013-10-31 20:14:17 +08:00
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s
|
2013-06-27 17:27:40 +08:00
|
|
|
|
|
|
|
declare void @foo(i16 *)
|
|
|
|
|
|
|
|
; Test the simple case, with the loaded value first.
|
|
|
|
define void @f1(i16 *%ptr, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f1:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...and with the loaded value second
|
|
|
|
define void @f2(i16 *%ptr, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f2:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %alt, i16 %orig
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Test cases where the value is explicitly sign-extended to 32 bits, with the
|
|
|
|
; loaded value first.
|
|
|
|
define void @f3(i16 *%ptr, i32 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f3:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = sext i16 %orig to i32
|
|
|
|
%res = select i1 %cond, i32 %ext, i32 %alt
|
|
|
|
%trunc = trunc i32 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...and with the loaded value second
|
|
|
|
define void @f4(i16 *%ptr, i32 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f4:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = sext i16 %orig to i32
|
|
|
|
%res = select i1 %cond, i32 %alt, i32 %ext
|
|
|
|
%trunc = trunc i32 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Test cases where the value is explicitly zero-extended to 32 bits, with the
|
|
|
|
; loaded value first.
|
|
|
|
define void @f5(i16 *%ptr, i32 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f5:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = zext i16 %orig to i32
|
|
|
|
%res = select i1 %cond, i32 %ext, i32 %alt
|
|
|
|
%trunc = trunc i32 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...and with the loaded value second
|
|
|
|
define void @f6(i16 *%ptr, i32 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f6:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = zext i16 %orig to i32
|
|
|
|
%res = select i1 %cond, i32 %alt, i32 %ext
|
|
|
|
%trunc = trunc i32 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Test cases where the value is explicitly sign-extended to 64 bits, with the
|
|
|
|
; loaded value first.
|
|
|
|
define void @f7(i16 *%ptr, i64 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f7:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = sext i16 %orig to i64
|
|
|
|
%res = select i1 %cond, i64 %ext, i64 %alt
|
|
|
|
%trunc = trunc i64 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...and with the loaded value second
|
|
|
|
define void @f8(i16 *%ptr, i64 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f8:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = sext i16 %orig to i64
|
|
|
|
%res = select i1 %cond, i64 %alt, i64 %ext
|
|
|
|
%trunc = trunc i64 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Test cases where the value is explicitly zero-extended to 64 bits, with the
|
|
|
|
; loaded value first.
|
|
|
|
define void @f9(i16 *%ptr, i64 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f9:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = zext i16 %orig to i64
|
|
|
|
%res = select i1 %cond, i64 %ext, i64 %alt
|
|
|
|
%trunc = trunc i64 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...and with the loaded value second
|
|
|
|
define void @f10(i16 *%ptr, i64 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f10:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%ext = zext i16 %orig to i64
|
|
|
|
%res = select i1 %cond, i64 %alt, i64 %ext
|
|
|
|
%trunc = trunc i64 %res to i16
|
|
|
|
store i16 %trunc, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the high end of the aligned STH range.
|
|
|
|
define void @f11(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f11:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sth %r3, 4094(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 2047
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the next halfword up, which should use STHY instead of STH.
|
|
|
|
define void @f12(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f12:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sthy %r3, 4096(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 2048
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the high end of the aligned STHY range.
|
|
|
|
define void @f13(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f13:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sthy %r3, 524286(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 262143
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the next halfword up, which needs separate address logic.
|
|
|
|
; Other sequences besides this one would be OK.
|
|
|
|
define void @f14(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f14:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: agfi %r2, 524288
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 262144
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the low end of the STHY range.
|
|
|
|
define void @f15(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f15:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sthy %r3, -524288(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 -262144
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the next halfword down, which needs separate address logic.
|
|
|
|
; Other sequences besides this one would be OK.
|
|
|
|
define void @f16(i16 *%base, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f16:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: agfi %r2, -524290
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
[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
llvm-svn: 230786
2015-02-28 03:29:02 +08:00
|
|
|
%ptr = getelementptr i16, i16 *%base, i64 -262145
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check that STHY allows an index.
|
|
|
|
define void @f17(i64 %base, i64 %index, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f17:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r2
|
|
|
|
; CHECK: sthy %r4, 4096(%r3,%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: br %r14
|
|
|
|
%add1 = add i64 %base, %index
|
|
|
|
%add2 = add i64 %add1, 4096
|
|
|
|
%ptr = inttoptr i64 %add2 to i16 *
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check that volatile loads are not matched.
|
|
|
|
define void @f18(i16 *%ptr, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f18:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: lh {{%r[0-5]}}, 0(%r2)
|
|
|
|
; CHECK: {{jl|jnl}} [[LABEL:[^ ]*]]
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: sth {{%r[0-5]}}, 0(%r2)
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load volatile i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...likewise stores. In this case we should have a conditional load into %r3.
|
|
|
|
define void @f19(i16 *%ptr, i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f19:
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: lh %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: sth %r3, 0(%r2)
|
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store volatile i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check that atomic loads are not matched. The transformation is OK for
|
|
|
|
; the "unordered" case tested here, but since we don't try to handle atomic
|
|
|
|
; operations at all in this context, it seems better to assert that than
|
|
|
|
; to restrict the test to a stronger ordering.
|
|
|
|
define void @f20(i16 *%ptr, i16 %alt, i32 %limit) {
|
|
|
|
; FIXME: should use a normal load instead of CS.
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f20:
|
2013-12-10 18:49:34 +08:00
|
|
|
; CHECK: lh {{%r[0-9]+}}, 0(%r2)
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: {{jl|jnl}} [[LABEL:[^ ]*]]
|
|
|
|
; CHECK: [[LABEL]]:
|
2013-12-10 18:49:34 +08:00
|
|
|
; CHECK: sth {{%r[0-9]+}}, 0(%r2)
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load atomic i16 , i16 *%ptr unordered, align 2
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; ...likewise stores.
|
|
|
|
define void @f21(i16 *%ptr, i16 %alt, i32 %limit) {
|
|
|
|
; FIXME: should use a normal store instead of CS.
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f21:
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jhe [[LABEL:[^ ]*]]
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: lh %r3, 0(%r2)
|
|
|
|
; CHECK: [[LABEL]]:
|
2013-12-10 18:49:34 +08:00
|
|
|
; CHECK: sth %r3, 0(%r2)
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: br %r14
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store atomic i16 %res, i16 *%ptr unordered, align 2
|
|
|
|
ret void
|
|
|
|
}
|
|
|
|
|
|
|
|
; Try a frame index base.
|
|
|
|
define void @f22(i16 %alt, i32 %limit) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f22:
|
2013-06-27 17:27:40 +08:00
|
|
|
; CHECK: brasl %r14, foo@PLT
|
|
|
|
; CHECK-NOT: %r15
|
|
|
|
; CHECK: jl [[LABEL:[^ ]*]]
|
|
|
|
; CHECK-NOT: %r15
|
|
|
|
; CHECK: sth {{%r[0-9]+}}, {{[0-9]+}}(%r15)
|
|
|
|
; CHECK: [[LABEL]]:
|
|
|
|
; CHECK: brasl %r14, foo@PLT
|
|
|
|
; CHECK: br %r14
|
|
|
|
%ptr = alloca i16
|
|
|
|
call void @foo(i16 *%ptr)
|
2013-09-18 17:56:40 +08:00
|
|
|
%cond = icmp ult i32 %limit, 420
|
2015-02-28 05:17:42 +08:00
|
|
|
%orig = load i16 , i16 *%ptr
|
2013-06-27 17:27:40 +08:00
|
|
|
%res = select i1 %cond, i16 %orig, i16 %alt
|
|
|
|
store i16 %res, i16 *%ptr
|
|
|
|
call void @foo(i16 *%ptr)
|
|
|
|
ret void
|
|
|
|
}
|