Skip to content

Commit 3318a72

Browse files
authored
[InstCombine] Fold (ct{t,l}z Pow2) -> Log2(Pow2) (#122620)
- **[InstCombine] Add tests for folding `(ct{t,l}z Pow2)`; NFC** - **[InstCombine] Fold `(ct{t,l}z Pow2)` -> `Log2(Pow2)`** Do so we can find `Log2(Pow2)` for "free" with `takeLog2` https://alive2.llvm.org/ce/z/CL77fo
1 parent 09a8b7c commit 3318a72

File tree

2 files changed

+106
-0
lines changed

2 files changed

+106
-0
lines changed

llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -588,6 +588,19 @@ static Instruction *foldCttzCtlz(IntrinsicInst &II, InstCombinerImpl &IC) {
588588
}
589589
}
590590

591+
// cttz(Pow2) -> Log2(Pow2)
592+
// ctlz(Pow2) -> BitWidth - 1 - Log2(Pow2)
593+
if (auto *R = IC.tryGetLog2(Op0, match(Op1, m_One()))) {
594+
if (IsTZ)
595+
return IC.replaceInstUsesWith(II, R);
596+
BinaryOperator *BO = BinaryOperator::CreateSub(
597+
ConstantInt::get(R->getType(), R->getType()->getScalarSizeInBits() - 1),
598+
R);
599+
BO->setHasNoSignedWrap();
600+
BO->setHasNoUnsignedWrap();
601+
return BO;
602+
}
603+
591604
KnownBits Known = IC.computeKnownBits(Op0, 0, &II);
592605

593606
// Create a mask for bits above (ctlz) or below (cttz) the first known one.

llvm/test/Transforms/InstCombine/cttz.ll

Lines changed: 93 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -297,3 +297,96 @@ define i16 @cttz_assume(i16 %x) {
297297
%cttz = call i16 @llvm.cttz.i16(i16 %x, i1 false)
298298
ret i16 %cttz
299299
}
300+
301+
302+
declare void @use.i8(i8)
303+
define i8 @fold_ctz_log2(i8 %x) {
304+
; CHECK-LABEL: @fold_ctz_log2(
305+
; CHECK-NEXT: [[R:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 5)
306+
; CHECK-NEXT: ret i8 [[R]]
307+
;
308+
%p2 = shl i8 1, %x
309+
%v = call i8 @llvm.umin(i8 %p2, i8 32)
310+
%r = call i8 @llvm.cttz(i8 %v, i1 false)
311+
ret i8 %r
312+
}
313+
314+
define i9 @fold_ctz_log2_i9_okay(i9 %x) {
315+
; CHECK-LABEL: @fold_ctz_log2_i9_okay(
316+
; CHECK-NEXT: [[R:%.*]] = call i9 @llvm.umin.i9(i9 [[X:%.*]], i9 5)
317+
; CHECK-NEXT: ret i9 [[R]]
318+
;
319+
%p2 = shl i9 1, %x
320+
%v = call i9 @llvm.umin(i9 %p2, i9 32)
321+
%r = call i9 @llvm.cttz(i9 %v, i1 false)
322+
ret i9 %r
323+
}
324+
325+
define i8 @fold_ctz_log2_maybe_z(i8 %x, i8 %y, i1 %c) {
326+
; CHECK-LABEL: @fold_ctz_log2_maybe_z(
327+
; CHECK-NEXT: [[V:%.*]] = shl i8 2, [[V_V:%.*]]
328+
; CHECK-NEXT: [[P2_2:%.*]] = shl i8 4, [[Y:%.*]]
329+
; CHECK-NEXT: [[V1:%.*]] = select i1 [[C:%.*]], i8 [[V]], i8 [[P2_2]]
330+
; CHECK-NEXT: [[R:%.*]] = call range(i8 1, 9) i8 @llvm.cttz.i8(i8 [[V1]], i1 false)
331+
; CHECK-NEXT: ret i8 [[R]]
332+
;
333+
%p2 = shl i8 2, %x
334+
%p2_2 = shl i8 4, %y
335+
%v = select i1 %c, i8 %p2, i8 %p2_2
336+
%r = call i8 @llvm.cttz(i8 %v, i1 false)
337+
ret i8 %r
338+
}
339+
340+
define i8 @fold_ctz_log2_maybe_z_okay(i8 %x, i8 %y, i1 %c) {
341+
; CHECK-LABEL: @fold_ctz_log2_maybe_z_okay(
342+
; CHECK-NEXT: [[X:%.*]] = add i8 [[X1:%.*]], 1
343+
; CHECK-NEXT: [[Y:%.*]] = add i8 [[Y1:%.*]], 2
344+
; CHECK-NEXT: [[V_V:%.*]] = select i1 [[C:%.*]], i8 [[X]], i8 [[Y]]
345+
; CHECK-NEXT: ret i8 [[V_V]]
346+
;
347+
%p2 = shl i8 2, %x
348+
%p2_2 = shl i8 4, %y
349+
%v = select i1 %c, i8 %p2, i8 %p2_2
350+
%r = call i8 @llvm.cttz(i8 %v, i1 true)
351+
ret i8 %r
352+
}
353+
354+
define i8 @fold_clz_log2(i8 %x) {
355+
; CHECK-LABEL: @fold_clz_log2(
356+
; CHECK-NEXT: [[TMP1:%.*]] = call i8 @llvm.umin.i8(i8 [[X:%.*]], i8 5)
357+
; CHECK-NEXT: [[R:%.*]] = xor i8 [[TMP1]], 7
358+
; CHECK-NEXT: ret i8 [[R]]
359+
;
360+
%p2 = shl i8 1, %x
361+
%v = call i8 @llvm.umin(i8 %p2, i8 32)
362+
%r = call i8 @llvm.ctlz(i8 %v, i1 false)
363+
ret i8 %r
364+
}
365+
366+
define i8 @fold_clz_log2_multiuse_fail(i8 %x) {
367+
; CHECK-LABEL: @fold_clz_log2_multiuse_fail(
368+
; CHECK-NEXT: [[P2:%.*]] = shl nuw i8 2, [[X:%.*]]
369+
; CHECK-NEXT: [[V:%.*]] = call i8 @llvm.umin.i8(i8 [[P2]], i8 32)
370+
; CHECK-NEXT: call void @use.i8(i8 [[V]])
371+
; CHECK-NEXT: [[R:%.*]] = call range(i8 2, 9) i8 @llvm.ctlz.i8(i8 [[V]], i1 true)
372+
; CHECK-NEXT: ret i8 [[R]]
373+
;
374+
%p2 = shl nuw i8 2, %x
375+
%v = call i8 @llvm.umin(i8 %p2, i8 32)
376+
call void @use.i8(i8 %v)
377+
%r = call i8 @llvm.ctlz(i8 %v, i1 true)
378+
ret i8 %r
379+
}
380+
381+
382+
define i9 @fold_clz_log2_i9(i9 %x) {
383+
; CHECK-LABEL: @fold_clz_log2_i9(
384+
; CHECK-NEXT: [[TMP1:%.*]] = call i9 @llvm.umin.i9(i9 [[X:%.*]], i9 5)
385+
; CHECK-NEXT: [[R:%.*]] = sub nuw nsw i9 8, [[TMP1]]
386+
; CHECK-NEXT: ret i9 [[R]]
387+
;
388+
%p2 = shl i9 1, %x
389+
%v = call i9 @llvm.umin(i9 %p2, i9 32)
390+
%r = call i9 @llvm.ctlz(i9 %v, i1 true)
391+
ret i9 %r
392+
}

0 commit comments

Comments
 (0)