maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/SystemZ/Large/branch-range-01.py

106 lines
3.0 KiB
Python
Raw Normal View History

[SystemZ] Add long branch pass Before this change, the SystemZ backend would use BRCL for all branches and only consider shortening them to BRC when generating an object file. E.g. a branch on equal would use the JGE alias of BRCL in assembly output, but might be shortened to the JE alias of BRC in ELF output. This was a useful first step, but it had two problems: (1) The z assembler isn't traditionally supposed to perform branch shortening or branch relaxation. We followed this rule by not relaxing branches in assembler input, but that meant that generating assembly code and then assembling it would not produce the same result as going directly to object code; the former would give long branches everywhere, whereas the latter would use short branches where possible. (2) Other useful branches, like COMPARE AND BRANCH, do not have long forms. We would need to do something else before supporting them. (Although COMPARE AND BRANCH does not change the condition codes, the plan is to model COMPARE AND BRANCH as a CC-clobbering instruction during codegen, so that we can safely lower it to a separate compare and long branch where necessary. This is not a valid transformation for the assembler proper to make.) This patch therefore moves branch relaxation to a pre-emit pass. For now, calls are still shortened from BRASL to BRAS by the assembler, although this too is not really the traditional behaviour. The first test takes about 1.5s to run, and there are likely to be more tests in this vein once further branch types are added. The feeling on IRC was that 1.5s is a bit much for a single test, so I've restricted it to SystemZ hosts for now. The patch exposes (and fixes) some typos in the main CodeGen/SystemZ tests. A later patch will remove the {{g}}s from that directory. llvm-svn: 182274
2013-05-20 22:23:08 +08:00
# Test normal conditional branches in cases where the sheer number of
# instructions causes some branches to be out of range.
# RUN: python %s | llc -mtriple=s390x-linux-gnu | FileCheck %s
# Construct:
#
# before0:
# conditional branch to after0
# ...
# beforeN:
# conditional branch to after0
# main:
# 0xffd8 bytes, from MVIY instructions
# conditional branch to main
# after0:
# ...
# conditional branch to main
# afterN:
#
# Each conditional branch sequence occupies 8 bytes if it uses a short branch
# and 10 if it uses a long one. The ones before "main:" have to take the branch
# length into account -- which is 4 bytes for short branches -- so the final
# (0x28 - 4) / 8 == 4 blocks can use short branches. The ones after "main:"
# do not, so the first 0x28 / 8 == 5 can use short branches. However,
# the conservative algorithm we use makes one branch unnecessarily long
# on each side.
#
# CHECK: c %r4, 0(%r3)
# CHECK: jge [[LABEL:\.L[^ ]*]]
# CHECK: c %r4, 4(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 8(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 12(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 16(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 20(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 24(%r3)
# CHECK: j{{g?}}e [[LABEL]]
# CHECK: c %r4, 28(%r3)
# CHECK: je [[LABEL]]
# CHECK: c %r4, 32(%r3)
# CHECK: je [[LABEL]]
# CHECK: c %r4, 36(%r3)
# CHECK: je [[LABEL]]
# ...main goes here...
# CHECK: c %r4, 100(%r3)
# CHECK: je [[LABEL:\.L[^ ]*]]
# CHECK: c %r4, 104(%r3)
# CHECK: je [[LABEL]]
# CHECK: c %r4, 108(%r3)
# CHECK: je [[LABEL]]
# CHECK: c %r4, 112(%r3)
# CHECK: je [[LABEL]]
# CHECK: c %r4, 116(%r3)
# CHECK: j{{g?}}e [[LABEL]]
# CHECK: c %r4, 120(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 124(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 128(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 132(%r3)
# CHECK: jge [[LABEL]]
# CHECK: c %r4, 136(%r3)
# CHECK: jge [[LABEL]]
branch_blocks = 10
main_size = 0xffd8
print 'define void @f1(i8 *%base, i32 *%stop, i32 %limit) {'
print 'entry:'
print ' br label %before0'
print ''
for i in xrange(branch_blocks):
next = 'before%d' % (i + 1) if i + 1 < branch_blocks else 'main'
print 'before%d:' % i
print ' %%bstop%d = getelementptr i32 *%%stop, i64 %d' % (i, i)
Add TargetLowering::prepareVolatileOrAtomicLoad One unusual feature of the z architecture is that the result of a previous load can be reused indefinitely for subsequent loads, even if a cache-coherent store to that location is performed by another CPU. A special serializing instruction must be used if you want to force a load to be reattempted. Since volatile loads are not supposed to be omitted in this way, we should insert a serializing instruction before each such load. The same goes for atomic loads. The patch implements this at the IR->DAG boundary, in a similar way to atomic fences. It is a no-op for targets other than SystemZ. llvm-svn: 196905
2013-12-10 18:36:34 +08:00
print ' %%bcur%d = load i32 *%%bstop%d' % (i, i)
[SystemZ] Add long branch pass Before this change, the SystemZ backend would use BRCL for all branches and only consider shortening them to BRC when generating an object file. E.g. a branch on equal would use the JGE alias of BRCL in assembly output, but might be shortened to the JE alias of BRC in ELF output. This was a useful first step, but it had two problems: (1) The z assembler isn't traditionally supposed to perform branch shortening or branch relaxation. We followed this rule by not relaxing branches in assembler input, but that meant that generating assembly code and then assembling it would not produce the same result as going directly to object code; the former would give long branches everywhere, whereas the latter would use short branches where possible. (2) Other useful branches, like COMPARE AND BRANCH, do not have long forms. We would need to do something else before supporting them. (Although COMPARE AND BRANCH does not change the condition codes, the plan is to model COMPARE AND BRANCH as a CC-clobbering instruction during codegen, so that we can safely lower it to a separate compare and long branch where necessary. This is not a valid transformation for the assembler proper to make.) This patch therefore moves branch relaxation to a pre-emit pass. For now, calls are still shortened from BRASL to BRAS by the assembler, although this too is not really the traditional behaviour. The first test takes about 1.5s to run, and there are likely to be more tests in this vein once further branch types are added. The feeling on IRC was that 1.5s is a bit much for a single test, so I've restricted it to SystemZ hosts for now. The patch exposes (and fixes) some typos in the main CodeGen/SystemZ tests. A later patch will remove the {{g}}s from that directory. llvm-svn: 182274
2013-05-20 22:23:08 +08:00
print ' %%btest%d = icmp eq i32 %%limit, %%bcur%d' % (i, i)
print ' br i1 %%btest%d, label %%after0, label %%%s' % (i, next)
print ''
print '%s:' % next
a, b = 1, 1
for i in xrange(0, main_size, 6):
a, b = b, a + b
offset = 4096 + b % 500000
value = a % 256
print ' %%ptr%d = getelementptr i8 *%%base, i64 %d' % (i, offset)
print ' store volatile i8 %d, i8 *%%ptr%d' % (value, i)
for i in xrange(branch_blocks):
print ' %%astop%d = getelementptr i32 *%%stop, i64 %d' % (i, i + 25)
Add TargetLowering::prepareVolatileOrAtomicLoad One unusual feature of the z architecture is that the result of a previous load can be reused indefinitely for subsequent loads, even if a cache-coherent store to that location is performed by another CPU. A special serializing instruction must be used if you want to force a load to be reattempted. Since volatile loads are not supposed to be omitted in this way, we should insert a serializing instruction before each such load. The same goes for atomic loads. The patch implements this at the IR->DAG boundary, in a similar way to atomic fences. It is a no-op for targets other than SystemZ. llvm-svn: 196905
2013-12-10 18:36:34 +08:00
print ' %%acur%d = load i32 *%%astop%d' % (i, i)
[SystemZ] Add long branch pass Before this change, the SystemZ backend would use BRCL for all branches and only consider shortening them to BRC when generating an object file. E.g. a branch on equal would use the JGE alias of BRCL in assembly output, but might be shortened to the JE alias of BRC in ELF output. This was a useful first step, but it had two problems: (1) The z assembler isn't traditionally supposed to perform branch shortening or branch relaxation. We followed this rule by not relaxing branches in assembler input, but that meant that generating assembly code and then assembling it would not produce the same result as going directly to object code; the former would give long branches everywhere, whereas the latter would use short branches where possible. (2) Other useful branches, like COMPARE AND BRANCH, do not have long forms. We would need to do something else before supporting them. (Although COMPARE AND BRANCH does not change the condition codes, the plan is to model COMPARE AND BRANCH as a CC-clobbering instruction during codegen, so that we can safely lower it to a separate compare and long branch where necessary. This is not a valid transformation for the assembler proper to make.) This patch therefore moves branch relaxation to a pre-emit pass. For now, calls are still shortened from BRASL to BRAS by the assembler, although this too is not really the traditional behaviour. The first test takes about 1.5s to run, and there are likely to be more tests in this vein once further branch types are added. The feeling on IRC was that 1.5s is a bit much for a single test, so I've restricted it to SystemZ hosts for now. The patch exposes (and fixes) some typos in the main CodeGen/SystemZ tests. A later patch will remove the {{g}}s from that directory. llvm-svn: 182274
2013-05-20 22:23:08 +08:00
print ' %%atest%d = icmp eq i32 %%limit, %%acur%d' % (i, i)
print ' br i1 %%atest%d, label %%main, label %%after%d' % (i, i)
print ''
print 'after%d:' % i
print ' ret void'
print '}'