[PATCH][X86] Add tablegen patterns to improve the selection of LZCNT and TZCNT instructions in a few cases.
Andrea Di Biagio
andrea.dibiagio at gmail.com
Fri May 16 08:38:54 PDT 2014
Hi,
This patch adds new ISel patters to improve the selection of TZCNT and LZCNT.
Instructions TZCNT (requires BMI1) and LZCNT (requires LZCNT), always
provides the operand size as output if the input operand is zero. We
can take advantage of this knowledge during instruction selection
stage in order to simplify a few corner case.
Scenario:
* we want to count the leading/trailing zeros of a value V;
* then we check if V is equal to zero; in case (if V is zero), we
propagate a constant equal to the sizeof (in bits) of V.
Instead of selecting a
tzcnt/lzcnt(V) + compare V with zero + conditional move
we only select
tzcnt/lzcnt.
The patch adds some test cases to verify that the new patterns do
their job and we don't introduce redundant compare+conditional move
instructions.
Please let me know if ok to submit.
Thanks,
Andrea Di Biagio
-------------- next part --------------
Index: lib/Target/X86/X86InstrInfo.td
===================================================================
--- lib/Target/X86/X86InstrInfo.td (revision 208994)
+++ lib/Target/X86/X86InstrInfo.td (working copy)
@@ -2002,6 +2002,46 @@
(implicit EFLAGS)]>, XS;
}
+let Predicates = [HasLZCNT] in {
+ def : Pat<(X86cmov (ctlz GR16:$src), (i16 16), (X86_COND_E),
+ (X86cmp GR16:$src, (i16 0))),
+ (LZCNT16rr GR16:$src)>;
+ def : Pat<(X86cmov (ctlz GR32:$src), (i32 32), (X86_COND_E),
+ (X86cmp GR32:$src, (i32 0))),
+ (LZCNT32rr GR32:$src)>;
+ def : Pat<(X86cmov (ctlz GR64:$src), (i64 64), (X86_COND_E),
+ (X86cmp GR64:$src, (i64 0))),
+ (LZCNT64rr GR64:$src)>;
+ def : Pat<(X86cmov (i16 16), (ctlz GR16:$src), (X86_COND_E),
+ (X86cmp GR16:$src, (i16 0))),
+ (LZCNT16rr GR16:$src)>;
+ def : Pat<(X86cmov (i32 32), (ctlz GR32:$src), (X86_COND_E),
+ (X86cmp GR32:$src, (i32 0))),
+ (LZCNT32rr GR32:$src)>;
+ def : Pat<(X86cmov (i64 64), (ctlz GR64:$src), (X86_COND_E),
+ (X86cmp GR64:$src, (i64 0))),
+ (LZCNT64rr GR64:$src)>;
+
+ def : Pat<(X86cmov (ctlz (loadi16 addr:$src)), (i16 16), (X86_COND_E),
+ (X86cmp (loadi16 addr:$src), (i16 0))),
+ (LZCNT16rm addr:$src)>;
+ def : Pat<(X86cmov (ctlz (loadi32 addr:$src)), (i32 32), (X86_COND_E),
+ (X86cmp (loadi32 addr:$src), (i32 0))),
+ (LZCNT32rm addr:$src)>;
+ def : Pat<(X86cmov (ctlz (loadi64 addr:$src)), (i64 64), (X86_COND_E),
+ (X86cmp (loadi64 addr:$src), (i64 0))),
+ (LZCNT64rm addr:$src)>;
+ def : Pat<(X86cmov (i16 16), (ctlz (loadi16 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi16 addr:$src), (i16 0))),
+ (LZCNT16rm addr:$src)>;
+ def : Pat<(X86cmov (i32 32), (ctlz (loadi32 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi32 addr:$src), (i32 0))),
+ (LZCNT32rm addr:$src)>;
+ def : Pat<(X86cmov (i64 64), (ctlz (loadi64 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi64 addr:$src), (i64 0))),
+ (LZCNT64rm addr:$src)>;
+}
+
//===----------------------------------------------------------------------===//
// BMI Instructions
//
@@ -2078,6 +2118,47 @@
(BLSI64rr GR64:$src)>;
}
+let Predicates = [HasBMI] in {
+ def : Pat<(X86cmov (cttz GR16:$src), (i16 16), (X86_COND_E),
+ (X86cmp GR16:$src, (i16 0))),
+ (TZCNT16rr GR16:$src)>;
+ def : Pat<(X86cmov (cttz GR32:$src), (i32 32), (X86_COND_E),
+ (X86cmp GR32:$src, (i32 0))),
+ (TZCNT32rr GR32:$src)>;
+ def : Pat<(X86cmov (cttz GR64:$src), (i64 64), (X86_COND_E),
+ (X86cmp GR64:$src, (i64 0))),
+ (TZCNT64rr GR64:$src)>;
+ def : Pat<(X86cmov (i16 16), (cttz GR16:$src), (X86_COND_E),
+ (X86cmp GR16:$src, (i16 0))),
+ (TZCNT16rr GR16:$src)>;
+ def : Pat<(X86cmov (i32 32), (cttz GR32:$src), (X86_COND_E),
+ (X86cmp GR32:$src, (i32 0))),
+ (TZCNT32rr GR32:$src)>;
+ def : Pat<(X86cmov (i64 64), (cttz GR64:$src), (X86_COND_E),
+ (X86cmp GR64:$src, (i64 0))),
+ (TZCNT64rr GR64:$src)>;
+
+ def : Pat<(X86cmov (cttz (loadi16 addr:$src)), (i16 16), (X86_COND_E),
+ (X86cmp (loadi16 addr:$src), (i16 0))),
+ (TZCNT16rm addr:$src)>;
+ def : Pat<(X86cmov (cttz (loadi32 addr:$src)), (i32 32), (X86_COND_E),
+ (X86cmp (loadi32 addr:$src), (i32 0))),
+ (TZCNT32rm addr:$src)>;
+ def : Pat<(X86cmov (cttz (loadi64 addr:$src)), (i64 64), (X86_COND_E),
+ (X86cmp (loadi64 addr:$src), (i64 0))),
+ (TZCNT64rm addr:$src)>;
+ def : Pat<(X86cmov (i16 16), (cttz (loadi16 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi16 addr:$src), (i16 0))),
+ (TZCNT16rm addr:$src)>;
+ def : Pat<(X86cmov (i32 32), (cttz (loadi32 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi32 addr:$src), (i32 0))),
+ (TZCNT32rm addr:$src)>;
+ def : Pat<(X86cmov (i64 64), (cttz (loadi64 addr:$src)), (X86_COND_E),
+ (X86cmp (loadi64 addr:$src), (i64 0))),
+ (TZCNT64rm addr:$src)>;
+}
+
+
multiclass bmi_bextr_bzhi<bits<8> opc, string mnemonic, RegisterClass RC,
X86MemOperand x86memop, Intrinsic Int,
PatFrag ld_frag> {
Index: test/CodeGen/X86/lzcnt-tzcnt.ll
===================================================================
--- test/CodeGen/X86/lzcnt-tzcnt.ll (revision 0)
+++ test/CodeGen/X86/lzcnt-tzcnt.ll (working copy)
@@ -0,0 +1,448 @@
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+bmi,+lzcnt | FileCheck %s
+
+; LZCNT and TZCNT will always produce the operand size when the input operand
+; is zero. This test is to verify that we efficiently select LZCNT/TZCNT
+; based on the fact that the 'icmp+select' sequence is always redundant
+; in every function defined below.
+
+
+define i16 @test1_ctlz(i16 %v) {
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 %v, 0
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test1_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test2_ctlz(i32 %v) {
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 %v, 0
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test2_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test3_ctlz(i64 %v) {
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 %v, 0
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test3_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test4_ctlz(i16 %v) {
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test4_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test5_ctlz(i32 %v) {
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test5_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test6_ctlz(i64 %v) {
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test6_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test7_ctlz(i16 %v) {
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 %cnt, i16 16
+ ret i16 %cond
+}
+; CHECK-LABEL: test7_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test8_ctlz(i32 %v) {
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 %cnt, i32 32
+ ret i32 %cond
+}
+; CHECK-LABEL: test8_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test9_ctlz(i64 %v) {
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 %cnt, i64 64
+ ret i64 %cond
+}
+; CHECK-LABEL: test9_ctlz
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test10_ctlz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 %v, 0
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test10_ctlz
+; CHECK-NOT: movw
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test11_ctlz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 %v, 0
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test11_ctlz
+; CHECK-NOT: movd
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test12_ctlz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 %v, 0
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test12_ctlz
+; CHECK-NOT: movq
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test13_ctlz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test13_ctlz
+; CHECK-NOT: movw
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test14_ctlz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test14_ctlz
+; CHECK-NOT: movd
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test15_ctlz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test15_ctlz
+; CHECK-NOT: movq
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test16_ctlz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.ctlz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 %cnt, i16 16
+ ret i16 %cond
+}
+; CHECK-LABEL: test16_ctlz
+; CHECK-NOT: movw
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test17_ctlz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.ctlz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 %cnt, i32 32
+ ret i32 %cond
+}
+; CHECK-LABEL: test17_ctlz
+; CHECK-NOT: movd
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test18_ctlz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.ctlz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 %cnt, i64 64
+ ret i64 %cond
+}
+; CHECK-LABEL: test18_ctlz
+; CHECK-NOT: movq
+; CHECK: lzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test1_cttz(i16 %v) {
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 %v, 0
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test1_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test2_cttz(i32 %v) {
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 %v, 0
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test2_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test3_cttz(i64 %v) {
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 %v, 0
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test3_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test4_cttz(i16 %v) {
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test4_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test5_cttz(i32 %v) {
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test5_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test6_cttz(i64 %v) {
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test6_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test7_cttz(i16 %v) {
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 %cnt, i16 16
+ ret i16 %cond
+}
+; CHECK-LABEL: test7_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test8_cttz(i32 %v) {
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 %cnt, i32 32
+ ret i32 %cond
+}
+; CHECK-LABEL: test8_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test9_cttz(i64 %v) {
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 %cnt, i64 64
+ ret i64 %cond
+}
+; CHECK-LABEL: test9_cttz
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test10_cttz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 %v, 0
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test10_cttz
+; CHECK-NOT: movw
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test11_cttz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 %v, 0
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test11_cttz
+; CHECK-NOT: movd
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test12_cttz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 %v, 0
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test12_cttz
+; CHECK-NOT: movq
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test13_cttz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 16, i16 %cnt
+ ret i16 %cond
+}
+; CHECK-LABEL: test13_cttz
+; CHECK-NOT: movw
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test14_cttz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 32, i32 %cnt
+ ret i32 %cond
+}
+; CHECK-LABEL: test14_cttz
+; CHECK-NOT: movd
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test15_cttz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 64, i64 %cnt
+ ret i64 %cond
+}
+; CHECK-LABEL: test15_cttz
+; CHECK-NOT: movq
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i16 @test16_cttz(i16* %ptr) {
+ %v = load i16* %ptr
+ %cnt = tail call i16 @llvm.cttz.i16(i16 %v, i1 true)
+ %tobool = icmp eq i16 0, %v
+ %cond = select i1 %tobool, i16 %cnt, i16 16
+ ret i16 %cond
+}
+; CHECK-LABEL: test16_cttz
+; CHECK-NOT: movw
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i32 @test17_cttz(i32* %ptr) {
+ %v = load i32* %ptr
+ %cnt = tail call i32 @llvm.cttz.i32(i32 %v, i1 true)
+ %tobool = icmp eq i32 0, %v
+ %cond = select i1 %tobool, i32 %cnt, i32 32
+ ret i32 %cond
+}
+; CHECK-LABEL: test17_cttz
+; CHECK-NOT: movd
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+define i64 @test18_cttz(i64* %ptr) {
+ %v = load i64* %ptr
+ %cnt = tail call i64 @llvm.cttz.i64(i64 %v, i1 true)
+ %tobool = icmp eq i64 0, %v
+ %cond = select i1 %tobool, i64 %cnt, i64 64
+ ret i64 %cond
+}
+; CHECK-LABEL: test18_cttz
+; CHECK-NOT: movq
+; CHECK: tzcnt
+; CHECK-NEXT: ret
+
+
+declare i64 @llvm.cttz.i64(i64, i1)
+declare i32 @llvm.cttz.i32(i32, i1)
+declare i16 @llvm.cttz.i16(i16, i1)
+declare i64 @llvm.ctlz.i64(i64, i1)
+declare i32 @llvm.ctlz.i32(i32, i1)
+declare i16 @llvm.ctlz.i16(i16, i1)
+
+
More information about the llvm-commits
mailing list