Skip to content

Commit 1d1f7ee

Browse files
austin880625goldsteinn
austin880625
authored and
goldsteinn
committed
[InstCombine] Add regression test cases for bitreverse optimization; NFC
Fold the following case on IR InstCombine pass. This patch includes the new test cases for this optimization ``` bitreverse(logic_op(x, bitreverse(y))) -> logic_op(bitreverse(x), y) bitreverse(logic_op(bitreverse(x), y)) -> logic_op(x, bitreverse(y)) bitreverse(logic_op(bitreverse(x), bitreverse(y))) -> logic_op(x, y) with multi-use ``` Reviewed By: goldstein.w.n Differential Revision: https://reviews.llvm.org/D151245
1 parent 171dfc5 commit 1d1f7ee

File tree

1 file changed

+150
-0
lines changed

1 file changed

+150
-0
lines changed

‎llvm/test/Transforms/InstCombine/bitreverse.ll

Lines changed: 150 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3,9 +3,13 @@
33

44
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
55

6+
declare i16 @llvm.bitreverse.i16(i16)
67
declare i32 @llvm.bitreverse.i32(i32)
8+
declare i64 @llvm.bitreverse.i64(i64)
79
declare <2 x i8> @llvm.bitreverse.v2i8(<2 x i8>)
10+
declare <2 x i32> @llvm.bitreverse.v2i32(<2 x i32>)
811
declare void @use_i32(i32)
12+
declare void @use_i64(i64)
913

1014
;pairwise reverse
1115
;template <typename T>
@@ -368,3 +372,149 @@ define i64 @PR59897(i1 %X1_2) {
368372
%X0_3x2x5x0 = zext i32 %X0_3x2x4x0 to i64
369373
ret i64 %X0_3x2x5x0
370374
}
375+
376+
; Issue#62236
377+
; Fold: BITREVERSE( OP( BITREVERSE(x), y ) ) -> OP( x, BITREVERSE(y) )
378+
379+
define i16 @rev_xor_lhs_rev16(i16 %a, i16 %b) #0 {
380+
; CHECK-LABEL: @rev_xor_lhs_rev16(
381+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i16 @llvm.bitreverse.i16(i16 [[A:%.*]])
382+
; CHECK-NEXT: [[TMP2:%.*]] = xor i16 [[TMP1]], [[B:%.*]]
383+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i16 @llvm.bitreverse.i16(i16 [[TMP2]])
384+
; CHECK-NEXT: ret i16 [[TMP3]]
385+
;
386+
%1 = tail call i16 @llvm.bitreverse.i16(i16 %a)
387+
%2 = xor i16 %1, %b
388+
%3 = tail call i16 @llvm.bitreverse.i16(i16 %2)
389+
ret i16 %3
390+
}
391+
392+
define i32 @rev_and_rhs_rev32(i32 %a, i32 %b) #0 {
393+
; CHECK-LABEL: @rev_and_rhs_rev32(
394+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i32 @llvm.bitreverse.i32(i32 [[B:%.*]])
395+
; CHECK-NEXT: [[TMP2:%.*]] = and i32 [[TMP1]], [[A:%.*]]
396+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i32 @llvm.bitreverse.i32(i32 [[TMP2]])
397+
; CHECK-NEXT: ret i32 [[TMP3]]
398+
;
399+
%1 = tail call i32 @llvm.bitreverse.i32(i32 %b)
400+
%2 = and i32 %a, %1
401+
%3 = tail call i32 @llvm.bitreverse.i32(i32 %2)
402+
ret i32 %3
403+
}
404+
405+
define i32 @rev_or_rhs_rev32(i32 %a, i32 %b) #0 {
406+
; CHECK-LABEL: @rev_or_rhs_rev32(
407+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i32 @llvm.bitreverse.i32(i32 [[B:%.*]])
408+
; CHECK-NEXT: [[TMP2:%.*]] = or i32 [[TMP1]], [[A:%.*]]
409+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i32 @llvm.bitreverse.i32(i32 [[TMP2]])
410+
; CHECK-NEXT: ret i32 [[TMP3]]
411+
;
412+
%1 = tail call i32 @llvm.bitreverse.i32(i32 %b)
413+
%2 = or i32 %a, %1
414+
%3 = tail call i32 @llvm.bitreverse.i32(i32 %2)
415+
ret i32 %3
416+
}
417+
418+
define i64 @rev_or_rhs_rev64(i64 %a, i64 %b) #0 {
419+
; CHECK-LABEL: @rev_or_rhs_rev64(
420+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
421+
; CHECK-NEXT: [[TMP2:%.*]] = or i64 [[TMP1]], [[A:%.*]]
422+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP2]])
423+
; CHECK-NEXT: ret i64 [[TMP3]]
424+
;
425+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %b)
426+
%2 = or i64 %a, %1
427+
%3 = tail call i64 @llvm.bitreverse.i64(i64 %2)
428+
ret i64 %3
429+
}
430+
431+
define i64 @rev_xor_rhs_rev64(i64 %a, i64 %b) #0 {
432+
; CHECK-LABEL: @rev_xor_rhs_rev64(
433+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
434+
; CHECK-NEXT: [[TMP2:%.*]] = xor i64 [[TMP1]], [[A:%.*]]
435+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP2]])
436+
; CHECK-NEXT: ret i64 [[TMP3]]
437+
;
438+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %b)
439+
%2 = xor i64 %a, %1
440+
%3 = tail call i64 @llvm.bitreverse.i64(i64 %2)
441+
ret i64 %3
442+
}
443+
444+
define <2 x i32> @rev_xor_rhs_i32vec(<2 x i32> %a, <2 x i32> %b) #0 {
445+
; CHECK-LABEL: @rev_xor_rhs_i32vec(
446+
; CHECK-NEXT: [[TMP1:%.*]] = tail call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> [[B:%.*]])
447+
; CHECK-NEXT: [[TMP2:%.*]] = xor <2 x i32> [[TMP1]], [[A:%.*]]
448+
; CHECK-NEXT: [[TMP3:%.*]] = tail call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> [[TMP2]])
449+
; CHECK-NEXT: ret <2 x i32> [[TMP3]]
450+
;
451+
%1 = tail call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> %b)
452+
%2 = xor <2 x i32> %a, %1
453+
%3 = tail call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> %2)
454+
ret <2 x i32> %3
455+
}
456+
457+
define i64 @rev_and_rhs_rev64_multiuse1(i64 %a, i64 %b) #0 {
458+
; CHECK-LABEL: @rev_and_rhs_rev64_multiuse1(
459+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
460+
; CHECK-NEXT: [[TMP2:%.*]] = and i64 [[TMP1]], [[A:%.*]]
461+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP2]])
462+
; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[TMP2]], [[TMP3]]
463+
; CHECK-NEXT: ret i64 [[TMP4]]
464+
;
465+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %b)
466+
%2 = and i64 %a, %1
467+
%3 = tail call i64 @llvm.bitreverse.i64(i64 %2)
468+
%4 = mul i64 %2, %3 ;increase use of logical op
469+
ret i64 %4
470+
}
471+
472+
define i64 @rev_and_rhs_rev64_multiuse2(i64 %a, i64 %b) #0 {
473+
; CHECK-LABEL: @rev_and_rhs_rev64_multiuse2(
474+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
475+
; CHECK-NEXT: [[TMP2:%.*]] = and i64 [[TMP1]], [[A:%.*]]
476+
; CHECK-NEXT: [[TMP3:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP2]])
477+
; CHECK-NEXT: [[TMP4:%.*]] = mul i64 [[TMP1]], [[TMP3]]
478+
; CHECK-NEXT: ret i64 [[TMP4]]
479+
;
480+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %b)
481+
%2 = and i64 %a, %1
482+
%3 = tail call i64 @llvm.bitreverse.i64(i64 %2)
483+
%4 = mul i64 %1, %3 ;increase use of inner bitreverse
484+
ret i64 %4
485+
}
486+
487+
define i64 @rev_all_operand64(i64 %a, i64 %b) #0 {
488+
; CHECK-LABEL: @rev_all_operand64(
489+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[A:%.*]])
490+
; CHECK-NEXT: [[TMP2:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
491+
; CHECK-NEXT: [[TMP3:%.*]] = and i64 [[TMP1]], [[TMP2]]
492+
; CHECK-NEXT: [[TMP4:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP3]])
493+
; CHECK-NEXT: ret i64 [[TMP4]]
494+
;
495+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %a)
496+
%2 = tail call i64 @llvm.bitreverse.i64(i64 %b)
497+
%3 = and i64 %1, %2
498+
%4 = tail call i64 @llvm.bitreverse.i64(i64 %3)
499+
ret i64 %4
500+
}
501+
502+
define i64 @rev_all_operand64_multiuse_both(i64 %a, i64 %b) #0 {
503+
; CHECK-LABEL: @rev_all_operand64_multiuse_both(
504+
; CHECK-NEXT: [[TMP1:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[A:%.*]])
505+
; CHECK-NEXT: [[TMP2:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[B:%.*]])
506+
; CHECK-NEXT: [[TMP3:%.*]] = and i64 [[TMP1]], [[TMP2]]
507+
; CHECK-NEXT: [[TMP4:%.*]] = tail call i64 @llvm.bitreverse.i64(i64 [[TMP3]])
508+
; CHECK-NEXT: call void @use_i64(i64 [[TMP1]])
509+
; CHECK-NEXT: call void @use_i64(i64 [[TMP2]])
510+
; CHECK-NEXT: ret i64 [[TMP4]]
511+
;
512+
%1 = tail call i64 @llvm.bitreverse.i64(i64 %a)
513+
%2 = tail call i64 @llvm.bitreverse.i64(i64 %b)
514+
%3 = and i64 %1, %2
515+
%4 = tail call i64 @llvm.bitreverse.i64(i64 %3)
516+
517+
call void @use_i64(i64 %1)
518+
call void @use_i64(i64 %2)
519+
ret i64 %4
520+
}

0 commit comments

Comments
 (0)