[InstCombine] handle trunc to i1 in foldLogOpOfMaskedICmps. #128861
Conversation
|
@llvm/pr-subscribers-llvm-transforms Author: Andreas Jonson (andjo403) Changesproof: https://alive2.llvm.org/ce/z/pu8WmX fixes #128778 Full diff: https://github.com/llvm/llvm-project/pull/128861.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
index 4616ea6ab5487..3649626d18e14 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -3336,12 +3336,6 @@ Value *InstCombinerImpl::foldAndOrOfICmps(ICmpInst *LHS, ICmpInst *RHS,
}
}
- // handle (roughly):
- // (icmp ne (A & B), C) | (icmp ne (A & D), E)
- // (icmp eq (A & B), C) & (icmp eq (A & D), E)
- if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, IsAnd, IsLogical, Builder, Q))
- return V;
-
if (Value *V =
foldAndOrOfICmpEqConstantAndICmp(LHS, RHS, IsAnd, IsLogical, Builder))
return V;
@@ -3516,6 +3510,13 @@ Value *InstCombinerImpl::foldBooleanAndOr(Value *LHS, Value *RHS,
if (!LHS->getType()->isIntOrIntVectorTy(1))
return nullptr;
+ // handle (roughly):
+ // (icmp ne (A & B), C) | (icmp ne (A & D), E)
+ // (icmp eq (A & B), C) & (icmp eq (A & D), E)
+ if (Value *V = foldLogOpOfMaskedICmps(LHS, RHS, IsAnd, IsLogical, Builder,
+ SQ.getWithInstruction(&I)))
+ return V;
+
if (auto *LHSCmp = dyn_cast<ICmpInst>(LHS))
if (auto *RHSCmp = dyn_cast<ICmpInst>(RHS))
if (Value *Res = foldAndOrOfICmps(LHSCmp, RHSCmp, I, IsAnd, IsLogical))
diff --git a/llvm/test/Transforms/InstCombine/eq-of-parts.ll b/llvm/test/Transforms/InstCombine/eq-of-parts.ll
index d07c2e6a5be52..4dc502788db47 100644
--- a/llvm/test/Transforms/InstCombine/eq-of-parts.ll
+++ b/llvm/test/Transforms/InstCombine/eq-of-parts.ll
@@ -1455,10 +1455,8 @@ define i1 @and_trunc_i1(i8 %a1, i8 %a2) {
define i1 @and_trunc_i1_wrong_const(i8 %a1, i8 %a2) {
; CHECK-LABEL: @and_trunc_i1_wrong_const(
; CHECK-NEXT: [[XOR:%.*]] = xor i8 [[A1:%.*]], [[A2:%.*]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[XOR]], 4
-; CHECK-NEXT: [[LOBIT:%.*]] = trunc i8 [[XOR]] to i1
-; CHECK-NEXT: [[LOBIT_INV:%.*]] = xor i1 [[LOBIT]], true
-; CHECK-NEXT: [[AND:%.*]] = and i1 [[CMP]], [[LOBIT_INV]]
+; CHECK-NEXT: [[TMP1:%.*]] = and i8 [[XOR]], -3
+; CHECK-NEXT: [[AND:%.*]] = icmp eq i8 [[TMP1]], 0
; CHECK-NEXT: ret i1 [[AND]]
;
%xor = xor i8 %a1, %a2
diff --git a/llvm/test/Transforms/InstCombine/onehot_merge.ll b/llvm/test/Transforms/InstCombine/onehot_merge.ll
index e60edc7315d46..33e39d78e1ce4 100644
--- a/llvm/test/Transforms/InstCombine/onehot_merge.ll
+++ b/llvm/test/Transforms/InstCombine/onehot_merge.ll
@@ -1163,10 +1163,9 @@ define i1 @two_types_of_bittest(i8 %x, i8 %c) {
define i1 @trunc_bittest_and_icmp_bittest(i8 %x, i8 %c) {
; CHECK-LABEL: @trunc_bittest_and_icmp_bittest(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[C:%.*]]
-; CHECK-NEXT: [[TRUNC:%.*]] = trunc i8 [[X:%.*]] to i1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[T0]]
-; CHECK-NEXT: [[ICMP2:%.*]] = icmp ne i8 [[AND]], 0
-; CHECK-NEXT: [[RET:%.*]] = and i1 [[ICMP2]], [[TRUNC]]
+; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[T0]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[X:%.*]], [[TMP1]]
+; CHECK-NEXT: [[RET:%.*]] = icmp eq i8 [[TMP2]], [[TMP1]]
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %c
@@ -1180,11 +1179,9 @@ define i1 @trunc_bittest_and_icmp_bittest(i8 %x, i8 %c) {
define i1 @trunc_bittest_or_icmp_bittest(i8 %x, i8 %c) {
; CHECK-LABEL: @trunc_bittest_or_icmp_bittest(
; CHECK-NEXT: [[T0:%.*]] = shl nuw i8 1, [[C:%.*]]
-; CHECK-NEXT: [[TRUNC:%.*]] = trunc i8 [[X:%.*]] to i1
-; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[T0]]
-; CHECK-NEXT: [[ICMP2:%.*]] = icmp eq i8 [[AND]], 0
-; CHECK-NEXT: [[NOT:%.*]] = xor i1 [[TRUNC]], true
-; CHECK-NEXT: [[RET:%.*]] = or i1 [[ICMP2]], [[NOT]]
+; CHECK-NEXT: [[TMP1:%.*]] = or i8 [[T0]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = and i8 [[X:%.*]], [[TMP1]]
+; CHECK-NEXT: [[RET:%.*]] = icmp ne i8 [[TMP2]], [[TMP1]]
; CHECK-NEXT: ret i1 [[RET]]
;
%t0 = shl i8 1, %c
|
|
what I can see many of the regressions in llvm-opt-benchmark is that |
|
The code change LG. But I don't understand what you mean in the following sentence:
Do you mean we have to land this patch first before fixing these regressions? |
|
sorry for the bad description what I was trying to say is that some of the regressions that this PR cause eg. https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2173/files#diff-c88279c89225628a242defdd907f1444ed67e844ff858382dbea2edc36237e7fR167035-R167038 and https://github.com/dtcxzyw/llvm-opt-benchmark/pull/2173/files#diff-403632332e00c769093bc90d3e97457d65f8adf3b5824d6cd460e13c166cc164R1747-R1749 |
|
@dtcxzyw what do you think we shall do with this PR? |
|
this PR also fix/hide the problem in #131824 |
This regression is still outstanding: https://alive2.llvm.org/ce/z/4tZwjg |
…e(0,2) (NFC) proof https://alive2.llvm.org/ce/z/xeazCu this is a regression found in #128861 This fold is done when icmp eq/ne x, 1/0 is folded to trunc nuw x iff x is in the range(0,2)
andjo403
force-pushed
the
foldLogOpOfMaskedICmpsTrunc
branch
from
dde1040 to
993fff6
Compare
have added a reduced version of the test in 1657331 |
…in the range(0,2) (NFC) proof https://alive2.llvm.org/ce/z/xeazCu this is a regression found in llvm/llvm-project#128861 This fold is done when icmp eq/ne x, 1/0 is folded to trunc nuw x iff x is in the range(0,2)
proof: https://alive2.llvm.org/ce/z/pu8WmX
fixes #128778