6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-15 04:30:12 +00:00
Code Issues Projects Releases Wiki Activity
3248223b66
llvm-6502/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll

66 lines
3.2 KiB
LLVM
Raw Normal View History

Fix broken x86_64 tests which specify non-64-bit cpu's. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134756 91177308-0d34-0410-b5e6-96231b3b80d8
2011-07-08 22:29:33 +00:00
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core2 | FileCheck %s
In the pre-RA scheduler, maintain cmp+br proximity. This is done by pushing physical register definitions close to their use, which happens to handle flag definitions if they're not glued to the branch. This seems to be generally a good thing though, so I didn't need to add a target hook yet. The primary motivation is to generate code closer to what people expect and rule out missed opportunity from enabling macro-op fusion. As a side benefit, we get several 2-5% gains on x86 benchmarks. There is one regression: SingleSource/Benchmarks/Shootout/lists slows down be -10%. But this is an independent scheduler bug that will be tracked separately. See rdar://problem/9283108. Incidentally, pre-RA scheduling is only half the solution. Fixing the later passes is tracked by: <rdar://problem/8932804> [pre-RA-sched] on x86, attempt to schedule CMP/TEST adjacent with condition jump Fixes: <rdar://problem/9262453> Scheduler unnecessary break of cmp/jump fusion git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129508 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-14 05:15:06 +00:00
; Reduced from JavaScriptCore
%"class.JSC::CodeLocationCall" = type { [8 x i8] }
%"class.JSC::JSGlobalData" = type { [4 x i8] }
%"class.JSC::FunctionPtr" = type { i8* }
%"class.JSC::Structure" = type { [4 x i8] }
%"class.JSC::UString" = type { i8* }
%"class.JSC::JSString" = type { [16 x i8], i32, %"class.JSC::UString", i32 }
declare hidden fastcc void @_ZN3JSCL23returnToThrowTrampolineEPNS_12JSGlobalDataENS_16ReturnAddressPtrERS2_(%"class.JSC::JSGlobalData"* nocapture, i8*, %"class.JSC::FunctionPtr"* nocapture) nounwind noinline ssp
; Avoid hoisting the test above loads or copies
; CHECK: %entry
For something like uint32_t hi(uint64_t res) { uint_32t hi = res >> 32; return !hi; } llvm IR looks like this: define i32 @hi(i64 %res) nounwind uwtable ssp { entry: %lnot = icmp ult i64 %res, 4294967296 %lnot.ext = zext i1 %lnot to i32 ret i32 %lnot.ext } The optimizer has optimize away the right shift and truncate but the resulting constant is too large to fit in the 32-bit immediate field. The resulting x86 code is worse as a result: movabsq $4294967296, %rax ## imm = 0x100000000 cmpq %rax, %rdi sbbl %eax, %eax andl $1, %eax This patch teaches the x86 lowering code to handle ult against a large immediate with trailing zeros. It will issue a right shift and a truncate followed by a comparison against a shifted immediate. shrq $32, %rdi testl %edi, %edi sete %al movzbl %al, %eax It also handles a ugt comparison against a large immediate with trailing bits set. i.e. X > 0x0ffffffff -> (X >> 32) >= 1 rdar://11866926 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160312 91177308-0d34-0410-b5e6-96231b3b80d8
2012-07-16 19:35:43 +00:00
; CHECK: test
In the pre-RA scheduler, maintain cmp+br proximity. This is done by pushing physical register definitions close to their use, which happens to handle flag definitions if they're not glued to the branch. This seems to be generally a good thing though, so I didn't need to add a target hook yet. The primary motivation is to generate code closer to what people expect and rule out missed opportunity from enabling macro-op fusion. As a side benefit, we get several 2-5% gains on x86 benchmarks. There is one regression: SingleSource/Benchmarks/Shootout/lists slows down be -10%. But this is an independent scheduler bug that will be tracked separately. See rdar://problem/9283108. Incidentally, pre-RA scheduling is only half the solution. Fixing the later passes is tracked by: <rdar://problem/8932804> [pre-RA-sched] on x86, attempt to schedule CMP/TEST adjacent with condition jump Fixes: <rdar://problem/9262453> Scheduler unnecessary break of cmp/jump fusion git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129508 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-14 05:15:06 +00:00
; CHECK-NOT: mov
For something like uint32_t hi(uint64_t res) { uint_32t hi = res >> 32; return !hi; } llvm IR looks like this: define i32 @hi(i64 %res) nounwind uwtable ssp { entry: %lnot = icmp ult i64 %res, 4294967296 %lnot.ext = zext i1 %lnot to i32 ret i32 %lnot.ext } The optimizer has optimize away the right shift and truncate but the resulting constant is too large to fit in the 32-bit immediate field. The resulting x86 code is worse as a result: movabsq $4294967296, %rax ## imm = 0x100000000 cmpq %rax, %rdi sbbl %eax, %eax andl $1, %eax This patch teaches the x86 lowering code to handle ult against a large immediate with trailing zeros. It will issue a right shift and a truncate followed by a comparison against a shifted immediate. shrq $32, %rdi testl %edi, %edi sete %al movzbl %al, %eax It also handles a ugt comparison against a large immediate with trailing bits set. i.e. X > 0x0ffffffff -> (X >> 32) >= 1 rdar://11866926 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160312 91177308-0d34-0410-b5e6-96231b3b80d8
2012-07-16 19:35:43 +00:00
; CHECK: je
In the pre-RA scheduler, maintain cmp+br proximity. This is done by pushing physical register definitions close to their use, which happens to handle flag definitions if they're not glued to the branch. This seems to be generally a good thing though, so I didn't need to add a target hook yet. The primary motivation is to generate code closer to what people expect and rule out missed opportunity from enabling macro-op fusion. As a side benefit, we get several 2-5% gains on x86 benchmarks. There is one regression: SingleSource/Benchmarks/Shootout/lists slows down be -10%. But this is an independent scheduler bug that will be tracked separately. See rdar://problem/9283108. Incidentally, pre-RA scheduling is only half the solution. Fixing the later passes is tracked by: <rdar://problem/8932804> [pre-RA-sched] on x86, attempt to schedule CMP/TEST adjacent with condition jump Fixes: <rdar://problem/9262453> Scheduler unnecessary break of cmp/jump fusion git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129508 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-14 05:15:06 +00:00
define i32 @cti_op_eq(i8** nocapture %args) nounwind ssp {
entry:
%0 = load i8** null, align 8
%tmp13 = bitcast i8* %0 to %"class.JSC::CodeLocationCall"*
%tobool.i.i.i = icmp ugt i8* undef, inttoptr (i64 281474976710655 to i8*)
%or.cond.i = and i1 %tobool.i.i.i, undef
br i1 %or.cond.i, label %if.then.i, label %if.end.i
if.then.i: ; preds = %entry
br i1 undef, label %if.then.i.i.i, label %_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit
if.then.i.i.i: ; preds = %if.then.i
%conv.i.i.i.i = trunc i64 undef to i32
br label %_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit
if.end.i: ; preds = %entry
br i1 undef, label %land.rhs.i121.i, label %_ZNK3JSC7JSValue8isStringEv.exit122.i
land.rhs.i121.i: ; preds = %if.end.i
%tmp.i.i117.i = load %"class.JSC::Structure"** undef, align 8
br label %_ZNK3JSC7JSValue8isStringEv.exit122.i
_ZNK3JSC7JSValue8isStringEv.exit122.i: ; preds = %land.rhs.i121.i, %if.end.i
%brmerge.i = or i1 undef, false
%or.cond = or i1 false, %brmerge.i
br i1 %or.cond, label %_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit, label %if.then.i92.i
if.then.i92.i: ; preds = %_ZNK3JSC7JSValue8isStringEv.exit122.i
tail call void @_ZNK3JSC8JSString11resolveRopeEPNS_9ExecStateE(%"class.JSC::JSString"* undef, %"class.JSC::CodeLocationCall"* %tmp13) nounwind
unreachable
_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit: ; preds = %_ZNK3JSC7JSValue8isStringEv.exit122.i, %if.then.i.i.i, %if.then.i
%1 = load i8** undef, align 8
br i1 undef, label %do.end39, label %do.body27
do.body27: ; preds = %_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit
%tmp30 = bitcast i8* %1 to %"class.JSC::JSGlobalData"*
%2 = getelementptr inbounds i8** %args, i64 -1
%3 = bitcast i8** %2 to %"class.JSC::FunctionPtr"*
tail call fastcc void @_ZN3JSCL23returnToThrowTrampolineEPNS_12JSGlobalDataENS_16ReturnAddressPtrERS2_(%"class.JSC::JSGlobalData"* %tmp30, i8* undef, %"class.JSC::FunctionPtr"* %3)
unreachable
do.end39: ; preds = %_ZN3JSC7JSValue19equalSlowCaseInlineEPNS_9ExecStateES0_S0_.exit
ret i32 undef
}
declare void @_ZNK3JSC8JSString11resolveRopeEPNS_9ExecStateE(%"class.JSC::JSString"*, %"class.JSC::CodeLocationCall"*)
Reference in New Issue Copy Permalink