[llvm-commits] [llvm] r160312 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll test/CodeGen/X86/cmp.ll
Evan Cheng
evan.cheng at apple.com
Mon Jul 16 12:35:43 PDT 2012
Author: evancheng
Date: Mon Jul 16 14:35:43 2012
New Revision: 160312
URL: http://llvm.org/viewvc/llvm-project?rev=160312&view=rev
Log:
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
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll
llvm/trunk/test/CodeGen/X86/cmp.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=160312&r1=160311&r2=160312&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Mon Jul 16 14:35:43 2012
@@ -3059,6 +3059,50 @@
RHS = DAG.getConstant(0, RHS.getValueType());
return X86::COND_LE;
}
+ if (SetCCOpcode == ISD::SETULT || SetCCOpcode == ISD::SETUGE) {
+ unsigned TrailZeros = RHSC->getAPIntValue().countTrailingZeros();
+ if (TrailZeros >= 32) {
+ // The constant doesn't fit in cmp immediate field. Right shift LHS by
+ // the # of trailing zeros and truncate it to 32-bit. Then compare
+ // against shifted RHS.
+ assert(LHS.getValueType() == MVT::i64 && "Expecting a 64-bit cmp!");
+ DebugLoc dl = LHS.getDebugLoc();
+ LHS = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
+ DAG.getNode(ISD::SRL, dl, MVT::i64, LHS,
+ DAG.getConstant(TrailZeros, MVT::i8)));
+ uint64_t C = RHSC->getZExtValue() >> TrailZeros;
+
+ if (SetCCOpcode == ISD::SETULT) {
+ // X < 0x300000000 -> (X >> 32) < 3
+ // X < 0x100000000 -> (X >> 32) == 0
+ // X < 0x200000000 -> (X >> 33) == 0
+ if (C == 1) {
+ RHS = DAG.getConstant(0, MVT::i32);
+ return X86::COND_E;
+ }
+ RHS = DAG.getConstant(C, MVT::i32);
+ return X86::COND_B;
+ } else /* SetCCOpcode == ISD::SETUGE */ {
+ // X >= 0x100000000 -> (X >> 32) >= 1
+ RHS = DAG.getConstant(C, MVT::i32);
+ return X86::COND_AE;
+ }
+ }
+ }
+ if (SetCCOpcode == ISD::SETUGT) {
+ unsigned TrailOnes = RHSC->getAPIntValue().countTrailingOnes();
+ if (TrailOnes >= 32 && !RHSC->isAllOnesValue()) {
+ assert(LHS.getValueType() == MVT::i64 && "Expecting a 64-bit cmp!");
+ DebugLoc dl = LHS.getDebugLoc();
+ LHS = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
+ DAG.getNode(ISD::SRL, dl, MVT::i64, LHS,
+ DAG.getConstant(TrailOnes, MVT::i8)));
+ uint64_t C = (RHSC->getZExtValue()+1) >> TrailOnes;
+ // X > 0x0ffffffff -> (X >> 32) >= 1
+ RHS = DAG.getConstant(C, MVT::i32);
+ return X86::COND_AE;
+ }
+ }
}
switch (SetCCOpcode) {
Modified: llvm/trunk/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll?rev=160312&r1=160311&r2=160312&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll (original)
+++ llvm/trunk/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll Mon Jul 16 14:35:43 2012
@@ -12,9 +12,9 @@
; Avoid hoisting the test above loads or copies
; CHECK: %entry
-; CHECK: cmpq
+; CHECK: test
; CHECK-NOT: mov
-; CHECK: jb
+; CHECK: je
define i32 @cti_op_eq(i8** nocapture %args) nounwind ssp {
entry:
%0 = load i8** null, align 8
Modified: llvm/trunk/test/CodeGen/X86/cmp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/cmp.ll?rev=160312&r1=160311&r2=160312&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/cmp.ll (original)
+++ llvm/trunk/test/CodeGen/X86/cmp.ll Mon Jul 16 14:35:43 2012
@@ -90,3 +90,51 @@
; CHECK: encoding: [0x48,0x83,0x7c,0x24,0xf8,0x00]
}
+; rdar://11866926
+define i32 @test7(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test7:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: testl %edi, %edi
+; CHECK: sete
+ %lnot = icmp ult i64 %res, 4294967296
+ %lnot.ext = zext i1 %lnot to i32
+ ret i32 %lnot.ext
+}
+
+define i32 @test8(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test8:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: cmpl $3, %edi
+ %lnot = icmp ult i64 %res, 12884901888
+ %lnot.ext = zext i1 %lnot to i32
+ ret i32 %lnot.ext
+}
+
+define i32 @test9(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test9:
+; CHECK-NOT: movabsq
+; CHECK: shrq $33, %rdi
+; CHECK: testl %edi, %edi
+; CHECK: sete
+ %lnot = icmp ult i64 %res, 8589934592
+ %lnot.ext = zext i1 %lnot to i32
+ ret i32 %lnot.ext
+}
+
+define i32 @test10(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test10:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: cmpl $1, %edi
+; CHECK: setae
+ %lnot = icmp uge i64 %res, 4294967296
+ %lnot.ext = zext i1 %lnot to i32
+ ret i32 %lnot.ext
+}
+
More information about the llvm-commits
mailing list