2014-01-26 16:12:34 +08:00
|
|
|
; RUN: llc < %s -march=sparcv9 -mattr=+popc -disable-sparc-delay-filler -disable-sparc-leaf-proc | FileCheck %s
|
|
|
|
|
; RUN: llc < %s -march=sparcv9 -mattr=+popc | FileCheck %s -check-prefix=OPT
|
2013-04-02 12:09:02 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: ret2:
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; CHECK: mov %i1, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; OPT-LABEL: ret2:
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; OPT: mov %o1, %o0
|
2013-04-02 12:09:02 +08:00
|
|
|
define i64 @ret2(i64 %a, i64 %b) {
|
|
|
|
|
ret i64 %b
|
|
|
|
|
}
|
2013-04-02 12:09:12 +08:00
|
|
|
|
|
|
|
|
; CHECK: shl_imm
|
|
|
|
|
; CHECK: sllx %i0, 7, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; OPT-LABEL: shl_imm:
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
2013-06-03 05:48:17 +08:00
|
|
|
; OPT: sllx %o0, 7, %o0
|
2013-04-02 12:09:12 +08:00
|
|
|
define i64 @shl_imm(i64 %a) {
|
|
|
|
|
%x = shl i64 %a, 7
|
|
|
|
|
ret i64 %x
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: sra_reg
|
|
|
|
|
; CHECK: srax %i0, %i1, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; OPT-LABEL: sra_reg:
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
2013-06-03 05:48:17 +08:00
|
|
|
; OPT: srax %o0, %o1, %o0
|
2013-04-02 12:09:12 +08:00
|
|
|
define i64 @sra_reg(i64 %a, i64 %b) {
|
|
|
|
|
%x = ashr i64 %a, %b
|
|
|
|
|
ret i64 %x
|
|
|
|
|
}
|
2013-04-02 12:09:17 +08:00
|
|
|
|
|
|
|
|
; Immediate materialization. Many of these patterns could actually be merged
|
|
|
|
|
; into the restore instruction:
|
|
|
|
|
;
|
|
|
|
|
; restore %g0, %g0, %o0
|
|
|
|
|
;
|
|
|
|
|
; CHECK: ret_imm0
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; CHECK: mov 0, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
|
|
|
|
; OPT: ret_imm0
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; OPT: mov 0, %o0
|
2013-04-02 12:09:17 +08:00
|
|
|
define i64 @ret_imm0() {
|
|
|
|
|
ret i64 0
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: ret_simm13
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; CHECK: mov -4096, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
|
|
|
|
; OPT: ret_simm13
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
TableGen: fix operand counting for aliases
TableGen has a fairly dubious heuristic to decide whether an alias should be
printed: does the alias have lest operands than the real instruction. This is
bad enough (particularly with no way to override it), but it should at least be
calculated consistently for both strings.
This patch implements that logic: first get the *correct* string for the
variant, in the same way as the Matcher, without guessing; then count the
number of whitespace chars.
There are basically 4 changes this brings about after the previous
commits; all of these appear to be good, so I have changed the tests:
+ ARM64: we print "neg X, Y" instead of "sub X, xzr, Y".
+ ARM64: we skip implicit "uxtx" and "uxtw" modifiers.
+ Sparc: we print "mov A, B" instead of "or %g0, A, B".
+ Sparc: we print "fcmpX A, B" instead of "fcmpX %fcc0, A, B"
llvm-svn: 208969
2014-05-16 17:42:04 +08:00
|
|
|
; OPT: mov -4096, %o0
|
2013-04-02 12:09:17 +08:00
|
|
|
define i64 @ret_simm13() {
|
|
|
|
|
ret i64 -4096
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: ret_sethi
|
|
|
|
|
; CHECK: sethi 4, %i0
|
|
|
|
|
; CHECK-NOT: or
|
|
|
|
|
; CHECK: restore
|
2013-06-03 05:48:17 +08:00
|
|
|
|
|
|
|
|
; OPT: ret_sethi
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
2013-06-03 05:48:17 +08:00
|
|
|
; OPT: sethi 4, %o0
|
2013-04-02 12:09:17 +08:00
|
|
|
define i64 @ret_sethi() {
|
|
|
|
|
ret i64 4096
|
|
|
|
|
}
|
|
|
|
|
|
2013-06-03 05:48:17 +08:00
|
|
|
; CHECK: ret_sethi_or
|
2013-04-02 12:09:17 +08:00
|
|
|
; CHECK: sethi 4, [[R:%[goli][0-7]]]
|
|
|
|
|
; CHECK: or [[R]], 1, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
|
|
|
|
; OPT: ret_sethi_or
|
|
|
|
|
; OPT: sethi 4, [[R:%[go][0-7]]]
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
2013-06-03 05:48:17 +08:00
|
|
|
; OPT: or [[R]], 1, %o0
|
|
|
|
|
|
2013-04-02 12:09:17 +08:00
|
|
|
define i64 @ret_sethi_or() {
|
|
|
|
|
ret i64 4097
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: ret_nimm33
|
|
|
|
|
; CHECK: sethi 4, [[R:%[goli][0-7]]]
|
|
|
|
|
; CHECK: xor [[R]], -4, %i0
|
2013-06-03 05:48:17 +08:00
|
|
|
|
|
|
|
|
; OPT: ret_nimm33
|
|
|
|
|
; OPT: sethi 4, [[R:%[go][0-7]]]
|
2014-01-10 10:55:27 +08:00
|
|
|
; OPT: retl
|
2013-06-03 05:48:17 +08:00
|
|
|
; OPT: xor [[R]], -4, %o0
|
|
|
|
|
|
2013-04-02 12:09:17 +08:00
|
|
|
define i64 @ret_nimm33() {
|
|
|
|
|
ret i64 -4100
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: ret_bigimm
|
|
|
|
|
; CHECK: sethi
|
|
|
|
|
; CHECK: sethi
|
|
|
|
|
define i64 @ret_bigimm() {
|
|
|
|
|
ret i64 6800754272627607872
|
|
|
|
|
}
|
2013-04-02 12:09:23 +08:00
|
|
|
|
2013-04-22 05:18:03 +08:00
|
|
|
; CHECK: ret_bigimm2
|
|
|
|
|
; CHECK: sethi 1048576
|
|
|
|
|
define i64 @ret_bigimm2() {
|
|
|
|
|
ret i64 4611686018427387904 ; 0x4000000000000000
|
|
|
|
|
}
|
|
|
|
|
|
2013-04-02 12:09:23 +08:00
|
|
|
; CHECK: reg_reg_alu
|
|
|
|
|
; CHECK: add %i0, %i1, [[R0:%[goli][0-7]]]
|
|
|
|
|
; CHECK: sub [[R0]], %i2, [[R1:%[goli][0-7]]]
|
|
|
|
|
; CHECK: andn [[R1]], %i0, %i0
|
|
|
|
|
define i64 @reg_reg_alu(i64 %x, i64 %y, i64 %z) {
|
|
|
|
|
%a = add i64 %x, %y
|
|
|
|
|
%b = sub i64 %a, %z
|
|
|
|
|
%c = xor i64 %x, -1
|
|
|
|
|
%d = and i64 %b, %c
|
|
|
|
|
ret i64 %d
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: reg_imm_alu
|
|
|
|
|
; CHECK: add %i0, -5, [[R0:%[goli][0-7]]]
|
|
|
|
|
; CHECK: xor [[R0]], 2, %i0
|
|
|
|
|
define i64 @reg_imm_alu(i64 %x, i64 %y, i64 %z) {
|
|
|
|
|
%a = add i64 %x, -5
|
|
|
|
|
%b = xor i64 %a, 2
|
|
|
|
|
ret i64 %b
|
|
|
|
|
}
|
2013-04-02 12:09:28 +08:00
|
|
|
|
|
|
|
|
; CHECK: loads
|
|
|
|
|
; CHECK: ldx [%i0]
|
|
|
|
|
; CHECK: stx %
|
|
|
|
|
; CHECK: ld [%i1]
|
|
|
|
|
; CHECK: st %
|
|
|
|
|
; CHECK: ldsw [%i2]
|
|
|
|
|
; CHECK: stx %
|
|
|
|
|
; CHECK: ldsh [%i3]
|
|
|
|
|
; CHECK: sth %
|
|
|
|
|
define i64 @loads(i64* %p, i32* %q, i32* %r, i16* %s) {
|
2015-02-28 05:17:42 +08:00
|
|
|
%a = load i64, i64* %p
|
2013-04-02 12:09:28 +08:00
|
|
|
%ai = add i64 1, %a
|
|
|
|
|
store i64 %ai, i64* %p
|
2015-02-28 05:17:42 +08:00
|
|
|
%b = load i32, i32* %q
|
2013-04-02 12:09:28 +08:00
|
|
|
%b2 = zext i32 %b to i64
|
|
|
|
|
%bi = trunc i64 %ai to i32
|
|
|
|
|
store i32 %bi, i32* %q
|
2015-02-28 05:17:42 +08:00
|
|
|
%c = load i32, i32* %r
|
2013-04-02 12:09:28 +08:00
|
|
|
%c2 = sext i32 %c to i64
|
|
|
|
|
store i64 %ai, i64* %p
|
2015-02-28 05:17:42 +08:00
|
|
|
%d = load i16, i16* %s
|
2013-04-02 12:09:28 +08:00
|
|
|
%d2 = sext i16 %d to i64
|
|
|
|
|
%di = trunc i64 %ai to i16
|
|
|
|
|
store i16 %di, i16* %s
|
|
|
|
|
|
|
|
|
|
%x1 = add i64 %a, %b2
|
|
|
|
|
%x2 = add i64 %c2, %d2
|
|
|
|
|
%x3 = add i64 %x1, %x2
|
|
|
|
|
ret i64 %x3
|
|
|
|
|
}
|
|
|
|
|
|
2013-06-08 06:55:05 +08:00
|
|
|
; CHECK: load_bool
|
|
|
|
|
; CHECK: ldub [%i0], %i0
|
|
|
|
|
define i64 @load_bool(i1* %p) {
|
2015-02-28 05:17:42 +08:00
|
|
|
%a = load i1, i1* %p
|
2013-06-08 06:55:05 +08:00
|
|
|
%b = zext i1 %a to i64
|
|
|
|
|
ret i64 %b
|
|
|
|
|
}
|
|
|
|
|
|
2013-04-02 12:09:28 +08:00
|
|
|
; CHECK: stores
|
|
|
|
|
; CHECK: ldx [%i0+8], [[R:%[goli][0-7]]]
|
|
|
|
|
; CHECK: stx [[R]], [%i0+16]
|
|
|
|
|
; CHECK: st [[R]], [%i1+-8]
|
|
|
|
|
; CHECK: sth [[R]], [%i2+40]
|
|
|
|
|
; CHECK: stb [[R]], [%i3+-20]
|
|
|
|
|
define void @stores(i64* %p, i32* %q, i16* %r, i8* %s) {
|
[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
|
|
|
%p1 = getelementptr i64, i64* %p, i64 1
|
|
|
|
|
%p2 = getelementptr i64, i64* %p, i64 2
|
2015-02-28 05:17:42 +08:00
|
|
|
%pv = load i64, i64* %p1
|
2013-04-02 12:09:28 +08:00
|
|
|
store i64 %pv, i64* %p2
|
|
|
|
|
|
[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
|
|
|
%q2 = getelementptr i32, i32* %q, i32 -2
|
2013-04-02 12:09:28 +08:00
|
|
|
%qv = trunc i64 %pv to i32
|
|
|
|
|
store i32 %qv, i32* %q2
|
|
|
|
|
|
[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
|
|
|
%r2 = getelementptr i16, i16* %r, i16 20
|
2013-04-02 12:09:28 +08:00
|
|
|
%rv = trunc i64 %pv to i16
|
|
|
|
|
store i16 %rv, i16* %r2
|
|
|
|
|
|
[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
|
|
|
%s2 = getelementptr i8, i8* %s, i8 -20
|
2013-04-02 12:09:28 +08:00
|
|
|
%sv = trunc i64 %pv to i8
|
|
|
|
|
store i8 %sv, i8* %s2
|
|
|
|
|
|
|
|
|
|
ret void
|
|
|
|
|
}
|
2013-04-14 13:48:50 +08:00
|
|
|
|
|
|
|
|
; CHECK: promote_shifts
|
|
|
|
|
; CHECK: ldub [%i0], [[R:%[goli][0-7]]]
|
|
|
|
|
; CHECK: sll [[R]], [[R]], %i0
|
|
|
|
|
define i8 @promote_shifts(i8* %p) {
|
2015-02-28 05:17:42 +08:00
|
|
|
%L24 = load i8, i8* %p
|
|
|
|
|
%L32 = load i8, i8* %p
|
2013-04-14 13:48:50 +08:00
|
|
|
%B36 = shl i8 %L24, %L32
|
|
|
|
|
ret i8 %B36
|
|
|
|
|
}
|
2013-04-16 10:57:02 +08:00
|
|
|
|
|
|
|
|
; CHECK: multiply
|
|
|
|
|
; CHECK: mulx %i0, %i1, %i0
|
|
|
|
|
define i64 @multiply(i64 %a, i64 %b) {
|
|
|
|
|
%r = mul i64 %a, %b
|
|
|
|
|
ret i64 %r
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: signed_divide
|
|
|
|
|
; CHECK: sdivx %i0, %i1, %i0
|
|
|
|
|
define i64 @signed_divide(i64 %a, i64 %b) {
|
|
|
|
|
%r = sdiv i64 %a, %b
|
|
|
|
|
ret i64 %r
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: unsigned_divide
|
|
|
|
|
; CHECK: udivx %i0, %i1, %i0
|
|
|
|
|
define i64 @unsigned_divide(i64 %a, i64 %b) {
|
|
|
|
|
%r = udiv i64 %a, %b
|
|
|
|
|
ret i64 %r
|
|
|
|
|
}
|
2013-05-20 03:14:24 +08:00
|
|
|
|
|
|
|
|
define void @access_fi() {
|
|
|
|
|
entry:
|
|
|
|
|
%b = alloca [32 x i8], align 1
|
[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
|
|
|
%arraydecay = getelementptr inbounds [32 x i8], [32 x i8]* %b, i64 0, i64 0
|
2013-05-20 03:14:24 +08:00
|
|
|
call void @g(i8* %arraydecay) #2
|
|
|
|
|
ret void
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare void @g(i8*)
|
2013-05-20 08:28:36 +08:00
|
|
|
|
|
|
|
|
; CHECK: expand_setcc
|
2013-06-07 08:03:36 +08:00
|
|
|
; CHECK: cmp %i0, 1
|
2013-05-20 08:28:36 +08:00
|
|
|
; CHECK: movl %xcc, 1,
|
|
|
|
|
define i32 @expand_setcc(i64 %a) {
|
|
|
|
|
%cond = icmp sle i64 %a, 0
|
|
|
|
|
%cast2 = zext i1 %cond to i32
|
|
|
|
|
%RV = sub i32 1, %cast2
|
|
|
|
|
ret i32 %RV
|
|
|
|
|
}
|
2013-05-20 08:53:25 +08:00
|
|
|
|
|
|
|
|
; CHECK: spill_i64
|
|
|
|
|
; CHECK: stx
|
|
|
|
|
; CHECK: ldx
|
|
|
|
|
define i64 @spill_i64(i64 %x) {
|
|
|
|
|
call void asm sideeffect "", "~{i0},~{i1},~{i2},~{i3},~{i4},~{i5},~{o0},~{o1},~{o2},~{o3},~{o4},~{o5},~{o7},~{l0},~{l1},~{l2},~{l3},~{l4},~{l5},~{l6},~{l7},~{g1},~{g2},~{g3},~{g4},~{g5},~{g6},~{g7}"()
|
|
|
|
|
ret i64 %x
|
|
|
|
|
}
|
2013-05-20 09:01:43 +08:00
|
|
|
|
|
|
|
|
; CHECK: bitcast_i64_f64
|
|
|
|
|
; CHECK: std
|
|
|
|
|
; CHECK: ldx
|
|
|
|
|
define i64 @bitcast_i64_f64(double %x) {
|
|
|
|
|
%y = bitcast double %x to i64
|
|
|
|
|
ret i64 %y
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; CHECK: bitcast_f64_i64
|
|
|
|
|
; CHECK: stx
|
|
|
|
|
; CHECK: ldd
|
|
|
|
|
define double @bitcast_f64_i64(i64 %x) {
|
|
|
|
|
%y = bitcast i64 %x to double
|
|
|
|
|
ret double %y
|
|
|
|
|
}
|
2013-06-03 08:21:54 +08:00
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: store_zero:
|
2013-06-03 08:21:54 +08:00
|
|
|
; CHECK: stx %g0, [%i0]
|
|
|
|
|
; CHECK: stx %g0, [%i1+8]
|
|
|
|
|
|
2013-07-14 14:24:09 +08:00
|
|
|
; OPT-LABEL: store_zero:
|
2013-06-03 08:21:54 +08:00
|
|
|
; OPT: stx %g0, [%o0]
|
|
|
|
|
; OPT: stx %g0, [%o1+8]
|
|
|
|
|
define i64 @store_zero(i64* nocapture %a, i64* nocapture %b) {
|
|
|
|
|
entry:
|
|
|
|
|
store i64 0, i64* %a, align 8
|
[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
|
|
|
%0 = getelementptr inbounds i64, i64* %b, i32 1
|
2013-06-03 08:21:54 +08:00
|
|
|
store i64 0, i64* %0, align 8
|
|
|
|
|
ret i64 0
|
|
|
|
|
}
|
2013-11-03 13:59:07 +08:00
|
|
|
|
|
|
|
|
; CHECK-LABEL: bit_ops
|
|
|
|
|
; CHECK: popc
|
|
|
|
|
|
|
|
|
|
; OPT-LABEL: bit_ops
|
|
|
|
|
; OPT: popc
|
|
|
|
|
|
|
|
|
|
define i64 @bit_ops(i64 %arg) {
|
|
|
|
|
entry:
|
|
|
|
|
%0 = tail call i64 @llvm.ctpop.i64(i64 %arg)
|
|
|
|
|
%1 = tail call i64 @llvm.ctlz.i64(i64 %arg, i1 true)
|
|
|
|
|
%2 = tail call i64 @llvm.cttz.i64(i64 %arg, i1 true)
|
|
|
|
|
%3 = tail call i64 @llvm.bswap.i64(i64 %arg)
|
|
|
|
|
%4 = add i64 %0, %1
|
|
|
|
|
%5 = add i64 %2, %3
|
|
|
|
|
%6 = add i64 %4, %5
|
|
|
|
|
ret i64 %6
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare i64 @llvm.ctpop.i64(i64) nounwind readnone
|
|
|
|
|
declare i64 @llvm.ctlz.i64(i64, i1) nounwind readnone
|
|
|
|
|
declare i64 @llvm.cttz.i64(i64, i1) nounwind readnone
|
|
|
|
|
declare i64 @llvm.bswap.i64(i64) nounwind readnone
|