mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 19:23:23 +01:00
0d99339102
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
162 lines
7.4 KiB
LLVM
162 lines
7.4 KiB
LLVM
; RUN: llc -march=mips -relocation-model=static < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=O32 --check-prefix=O32BE %s
|
|
; RUN: llc -march=mipsel -relocation-model=static < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=O32 --check-prefix=O32LE %s
|
|
|
|
; RUN-TODO: llc -march=mips64 -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=O32 %s
|
|
; RUN-TODO: llc -march=mips64el -relocation-model=static -target-abi o32 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=O32 %s
|
|
|
|
; RUN: llc -march=mips64 -relocation-model=static -target-abi n32 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=N32 --check-prefix=NEW --check-prefix=NEWBE %s
|
|
; RUN: llc -march=mips64el -relocation-model=static -target-abi n32 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM32 --check-prefix=N32 --check-prefix=NEW --check-prefix=NEWLE %s
|
|
|
|
; RUN: llc -march=mips64 -relocation-model=static -target-abi n64 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM64 --check-prefix=N64 --check-prefix=NEW --check-prefix=NEWBE %s
|
|
; RUN: llc -march=mips64el -relocation-model=static -target-abi n64 < %s | FileCheck --check-prefix=ALL --check-prefix=SYM64 --check-prefix=N64 --check-prefix=NEW --check-prefix=NEWLE %s
|
|
|
|
; Test the effect of varargs on floating point types in the non-variable part
|
|
; of the argument list as specified by section 2 of the MIPSpro N32 Handbook.
|
|
;
|
|
; N32/N64 are almost identical in this area so many of their checks have been
|
|
; combined into the 'NEW' prefix (the N stands for New).
|
|
;
|
|
; On O32, varargs prevents all FPU argument register usage. This contradicts
|
|
; the N32 handbook, but agrees with the SYSV ABI and GCC's behaviour.
|
|
|
|
@floats = global [11 x float] zeroinitializer
|
|
@doubles = global [11 x double] zeroinitializer
|
|
|
|
define void @double_args(double %a, ...)
|
|
nounwind {
|
|
entry:
|
|
%0 = getelementptr [11 x double], [11 x double]* @doubles, i32 0, i32 1
|
|
store volatile double %a, double* %0
|
|
|
|
%ap = alloca i8*
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
call void @llvm.va_start(i8* %ap2)
|
|
%b = va_arg i8** %ap, double
|
|
%1 = getelementptr [11 x double], [11 x double]* @doubles, i32 0, i32 2
|
|
store volatile double %b, double* %1
|
|
call void @llvm.va_end(i8* %ap2)
|
|
ret void
|
|
}
|
|
|
|
; ALL-LABEL: double_args:
|
|
; We won't test the way the global address is calculated in this test. This is
|
|
; just to get the register number for the other checks.
|
|
; SYM32-DAG: addiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(doubles)
|
|
; SYM64-DAG: ld [[R2:\$[0-9]]], %got_disp(doubles)(
|
|
|
|
; O32 forbids using floating point registers for the non-variable portion.
|
|
; N32/N64 allow it.
|
|
; O32BE-DAG: mtc1 $5, [[FTMP1:\$f[0-9]*[02468]+]]
|
|
; O32BE-DAG: mtc1 $4, [[FTMP2:\$f[0-9]*[13579]+]]
|
|
; O32LE-DAG: mtc1 $4, [[FTMP1:\$f[0-9]*[02468]+]]
|
|
; O32LE-DAG: mtc1 $5, [[FTMP2:\$f[0-9]*[13579]+]]
|
|
; O32-DAG: sdc1 [[FTMP1]], 8([[R2]])
|
|
; NEW-DAG: sdc1 $f12, 8([[R2]])
|
|
|
|
; The varargs portion is dumped to stack
|
|
; O32-DAG: sw $6, 16($sp)
|
|
; O32-DAG: sw $7, 20($sp)
|
|
; NEW-DAG: sd $5, 8($sp)
|
|
; NEW-DAG: sd $6, 16($sp)
|
|
; NEW-DAG: sd $7, 24($sp)
|
|
; NEW-DAG: sd $8, 32($sp)
|
|
; NEW-DAG: sd $9, 40($sp)
|
|
; NEW-DAG: sd $10, 48($sp)
|
|
; NEW-DAG: sd $11, 56($sp)
|
|
|
|
; Get the varargs pointer
|
|
; O32 has 4 bytes padding, 4 bytes for the varargs pointer, and 8 bytes reserved
|
|
; for arguments 1 and 2.
|
|
; N32/N64 has 8 bytes for the varargs pointer, and no reserved area.
|
|
; O32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 16
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
|
|
; Increment the pointer then get the varargs arg
|
|
; LLVM will rebind the load to the stack pointer instead of the varargs pointer
|
|
; during lowering. This is fine and doesn't change the behaviour.
|
|
; O32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR]], [[VAPTR]], 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
; O32-DAG: ldc1 [[FTMP1:\$f[0-9]+]], 16($sp)
|
|
; NEW-DAG: ldc1 [[FTMP1:\$f[0-9]+]], 8($sp)
|
|
; ALL-DAG: sdc1 [[FTMP1]], 16([[R2]])
|
|
|
|
define void @float_args(float %a, ...) nounwind {
|
|
entry:
|
|
%0 = getelementptr [11 x float], [11 x float]* @floats, i32 0, i32 1
|
|
store volatile float %a, float* %0
|
|
|
|
%ap = alloca i8*
|
|
%ap2 = bitcast i8** %ap to i8*
|
|
call void @llvm.va_start(i8* %ap2)
|
|
%b = va_arg i8** %ap, float
|
|
%1 = getelementptr [11 x float], [11 x float]* @floats, i32 0, i32 2
|
|
store volatile float %b, float* %1
|
|
call void @llvm.va_end(i8* %ap2)
|
|
ret void
|
|
}
|
|
|
|
; ALL-LABEL: float_args:
|
|
; We won't test the way the global address is calculated in this test. This is
|
|
; just to get the register number for the other checks.
|
|
; SYM32-DAG: addiu [[R2:\$[0-9]+]], ${{[0-9]+}}, %lo(floats)
|
|
; SYM64-DAG: ld [[R2:\$[0-9]]], %got_disp(floats)(
|
|
|
|
; The first four arguments are the same in O32/N32/N64.
|
|
; The non-variable portion should be unaffected.
|
|
; O32-DAG: sw $4, 4([[R2]])
|
|
; NEW-DAG: swc1 $f12, 4([[R2]])
|
|
|
|
; The varargs portion is dumped to stack
|
|
; O32-DAG: sw $5, 12($sp)
|
|
; O32-DAG: sw $6, 16($sp)
|
|
; O32-DAG: sw $7, 20($sp)
|
|
; NEW-DAG: sd $5, 8($sp)
|
|
; NEW-DAG: sd $6, 16($sp)
|
|
; NEW-DAG: sd $7, 24($sp)
|
|
; NEW-DAG: sd $8, 32($sp)
|
|
; NEW-DAG: sd $9, 40($sp)
|
|
; NEW-DAG: sd $10, 48($sp)
|
|
; NEW-DAG: sd $11, 56($sp)
|
|
|
|
; Get the varargs pointer
|
|
; O32 has 4 bytes padding, 4 bytes for the varargs pointer, and should have 8
|
|
; bytes reserved for arguments 1 and 2 (the first float arg) but as discussed in
|
|
; arguments-float.ll, GCC doesn't agree with MD00305 and treats floats as 4
|
|
; bytes so we only have 12 bytes total.
|
|
; N32/N64 has 8 bytes for the varargs pointer, and no reserved area.
|
|
; O32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 12
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR:\$[0-9]+]], $sp, 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
|
|
; Increment the pointer then get the varargs arg
|
|
; LLVM will rebind the load to the stack pointer instead of the varargs pointer
|
|
; during lowering. This is fine and doesn't change the behaviour.
|
|
; Also, in big-endian mode the offset must be increased by 4 to retrieve the
|
|
; correct half of the argument slot.
|
|
;
|
|
; O32-DAG: addiu [[VAPTR]], [[VAPTR]], 4
|
|
; O32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N32-DAG: addiu [[VAPTR]], [[VAPTR]], 8
|
|
; N32-DAG: sw [[VAPTR]], 4($sp)
|
|
; N64-DAG: daddiu [[VAPTR]], [[VAPTR]], 8
|
|
; N64-DAG: sd [[VAPTR]], 0($sp)
|
|
; O32-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 12($sp)
|
|
; NEWLE-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 8($sp)
|
|
; NEWBE-DAG: lwc1 [[FTMP1:\$f[0-9]+]], 12($sp)
|
|
; ALL-DAG: swc1 [[FTMP1]], 8([[R2]])
|
|
|
|
declare void @llvm.va_start(i8*)
|
|
declare void @llvm.va_copy(i8*, i8*)
|
|
declare void @llvm.va_end(i8*)
|