[llvm] [TypePromotion] Support positive addition amounts in isSafeWrap. (PR #81690)
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 13 16:14:52 PST 2024
https://github.com/topperc created https://github.com/llvm/llvm-project/pull/81690
We can support these by changing the sext promotion to -zext(-C) and replacing a sgt check with ugt. Reframing the logic in terms of how the unsigned range are affected.
More comments in the patch.
>From f6a482099643007f1723f4069673431842342ac7 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Tue, 13 Feb 2024 15:50:24 -0800
Subject: [PATCH] [TypePromotion] Support positive addition amounts in
isSafeWrap.
We can support these by changing the sext promotion to -zext(-C)
and replacing a sgt check with ugt. Reframing the logic in terms of
how the unsigned range are affected.
More comments in the patch.
---
llvm/lib/CodeGen/TypePromotion.cpp | 109 ++++----
llvm/test/CodeGen/AArch64/and-mask-removal.ll | 62 ++---
.../lack-of-signed-truncation-check.ll | 62 +++--
.../AArch64/signed-truncation-check.ll | 64 +++--
.../CodeGen/AArch64/typepromotion-overflow.ll | 5 +-
.../CodeGen/AArch64/typepromotion-signed.ll | 7 +-
.../RISCV/lack-of-signed-truncation-check.ll | 240 +++++++++++-------
.../CodeGen/RISCV/signed-truncation-check.ll | 206 ++++++++++-----
.../CodeGen/RISCV/typepromotion-overflow.ll | 5 +-
.../Transforms/TypePromotion/ARM/icmps.ll | 7 +-
.../Transforms/TypePromotion/ARM/wrapping.ll | 10 +-
11 files changed, 470 insertions(+), 307 deletions(-)
diff --git a/llvm/lib/CodeGen/TypePromotion.cpp b/llvm/lib/CodeGen/TypePromotion.cpp
index 48ad8de778010e..2116027f22c6bd 100644
--- a/llvm/lib/CodeGen/TypePromotion.cpp
+++ b/llvm/lib/CodeGen/TypePromotion.cpp
@@ -272,64 +272,58 @@ bool TypePromotionImpl::isSink(Value *V) {
/// Return whether this instruction can safely wrap.
bool TypePromotionImpl::isSafeWrap(Instruction *I) {
- // We can support a potentially wrapping instruction (I) if:
+ // We can support a potentially wrapping Add/Sub instruction (I) if:
// - It is only used by an unsigned icmp.
// - The icmp uses a constant.
- // - The wrapping value (I) is decreasing, i.e would underflow - wrapping
- // around zero to become a larger number than before.
// - The wrapping instruction (I) also uses a constant.
//
- // We can then use the two constants to calculate whether the result would
- // wrap in respect to itself in the original bitwidth. If it doesn't wrap,
- // just underflows the range, the icmp would give the same result whether the
- // result has been truncated or not. We calculate this by:
- // - Zero extending both constants, if needed, to RegisterBitWidth.
- // - Take the absolute value of I's constant, adding this to the icmp const.
- // - Check that this value is not out of range for small type. If it is, it
- // means that it has underflowed enough to wrap around the icmp constant.
+ // This a common pattern emitted to check if a value is within a range.
//
// For example:
//
- // %sub = sub i8 %a, 2
- // %cmp = icmp ule i8 %sub, 254
+ // %sub = sub i8 %a, C1
+ // %cmp = icmp ule i8 %sub, C2
+ //
+ // or
//
- // If %a = 0, %sub = -2 == FE == 254
- // But if this is evalulated as a i32
- // %sub = -2 == FF FF FF FE == 4294967294
- // So the unsigned compares (i8 and i32) would not yield the same result.
+ // %add = add i8 %a, C1
+ // %cmp = icmp ule i8 %add, C2.
//
- // Another way to look at it is:
- // %a - 2 <= 254
- // %a + 2 <= 254 + 2
- // %a <= 256
- // And we can't represent 256 in the i8 format, so we don't support it.
+ // We will treat an add as though it were a subtract by -C1. To promote
+ // the Add/Sub we will zero extend the LHS and the subtracted amount. For Add,
+ // this means we need to negate the constant, zero extend to RegisterBitWidth,
+ // and negate in the larger type.
//
- // Whereas:
+ // This will produce a value in the range [-zext(C1), zext(X)-zext(C1)] where
+ // C1 is the subtracted amount. This is either a small unsigned number or a
+ // large unsigned number in the promoted type.
//
- // %sub i8 %a, 1
+ // Now we need to correct the compare constant C2. Values >= C1 in the
+ // original add result range have been remapped to large values in the
+ // promoted range. If the compare constant fell into this range we need to
+ // remap it as well. We can do this as -(zext(-C2)).
+ //
+ // For example:
+ //
+ // %sub = sub i8 %a, 2
// %cmp = icmp ule i8 %sub, 254
//
- // If %a = 0, %sub = -1 == FF == 255
- // As i32:
- // %sub = -1 == FF FF FF FF == 4294967295
+ // becomes
//
- // In this case, the unsigned compare results would be the same and this
- // would also be true for ult, uge and ugt:
- // - (255 < 254) == (0xFFFFFFFF < 254) == false
- // - (255 <= 254) == (0xFFFFFFFF <= 254) == false
- // - (255 > 254) == (0xFFFFFFFF > 254) == true
- // - (255 >= 254) == (0xFFFFFFFF >= 254) == true
+ // %zext = zext %a to i32
+ // %sub = sub i32 %zext, 2
+ // %cmp = icmp ule i32 %sub, 4294967294
//
- // To demonstrate why we can't handle increasing values:
+ // Another example:
//
- // %add = add i8 %a, 2
- // %cmp = icmp ult i8 %add, 127
+ // %sub = sub i8 %a, 1
+ // %cmp = icmp ule i8 %sub, 254
//
- // If %a = 254, %add = 256 == (i8 1)
- // As i32:
- // %add = 256
+ // becomes
//
- // (1 < 127) != (256 < 127)
+ // %zext = zext %a to i32
+ // %sub = sub i32 %zext, 1
+ // %cmp = icmp ule i32 %sub, 254
unsigned Opc = I->getOpcode();
if (Opc != Instruction::Add && Opc != Instruction::Sub)
@@ -356,15 +350,10 @@ bool TypePromotionImpl::isSafeWrap(Instruction *I) {
APInt OverflowConst = cast<ConstantInt>(I->getOperand(1))->getValue();
if (Opc == Instruction::Sub)
OverflowConst = -OverflowConst;
- if (!OverflowConst.isNonPositive())
- return false;
SafeWrap.insert(I);
- // Using C1 = OverflowConst and C2 = ICmpConst, we can either prove that:
- // zext(x) + sext(C1) <u zext(C2) if C1 < 0 and C1 >s C2
- // zext(x) + sext(C1) <u sext(C2) if C1 < 0 and C1 <=s C2
- if (OverflowConst.sgt(ICmpConst)) {
+ if (OverflowConst.ugt(ICmpConst)) {
LLVM_DEBUG(dbgs() << "IR Promotion: Allowing safe overflow for sext "
<< "const of " << *I << "\n");
return true;
@@ -487,18 +476,24 @@ void IRPromoter::PromoteTree() {
continue;
if (auto *Const = dyn_cast<ConstantInt>(Op)) {
- // For subtract, we don't need to sext the constant. We only put it in
+ // For subtract, we only need to zext the constant. We only put it in
// SafeWrap because SafeWrap.size() is used elsewhere.
- // For cmp, we need to sign extend a constant appearing in either
- // operand. For add, we should only sign extend the RHS.
- Constant *NewConst =
- ConstantInt::get(Const->getContext(),
- (SafeWrap.contains(I) &&
- (I->getOpcode() == Instruction::ICmp || i == 1) &&
- I->getOpcode() != Instruction::Sub)
- ? Const->getValue().sext(PromotedWidth)
- : Const->getValue().zext(PromotedWidth));
- I->setOperand(i, NewConst);
+ // For Add and ICmp we need to find how far the constant is from the
+ // top of its original unsigned range and place it the same distance
+ // from the top of its new unsigned range. We can do this by negating
+ // the constant, zero extending it, then negating in the new type.
+ APInt NewConst;
+ if (SafeWrap.contains(I)) {
+ if (I->getOpcode() == Instruction::ICmp)
+ NewConst = -((-Const->getValue()).zext(PromotedWidth));
+ else if (I->getOpcode() == Instruction::Add && i == 1)
+ NewConst = -((-Const->getValue()).zext(PromotedWidth));
+ else
+ NewConst = Const->getValue().zext(PromotedWidth);
+ } else
+ NewConst = Const->getValue().zext(PromotedWidth);
+
+ I->setOperand(i, ConstantInt::get(Const->getContext(), NewConst));
} else if (isa<UndefValue>(Op))
I->setOperand(i, ConstantInt::get(ExtTy, 0));
}
diff --git a/llvm/test/CodeGen/AArch64/and-mask-removal.ll b/llvm/test/CodeGen/AArch64/and-mask-removal.ll
index 17ff0159701689..4ed482c8f2ba39 100644
--- a/llvm/test/CodeGen/AArch64/and-mask-removal.ll
+++ b/llvm/test/CodeGen/AArch64/and-mask-removal.ll
@@ -65,9 +65,8 @@ if.end: ; preds = %if.then, %entry
define zeroext i1 @test8_0(i8 zeroext %x) align 2 {
; CHECK-LABEL: test8_0:
; CHECK: ; %bb.0: ; %entry
-; CHECK-NEXT: add w8, w0, #74
-; CHECK-NEXT: and w8, w8, #0xff
-; CHECK-NEXT: cmp w8, #236
+; CHECK-NEXT: sub w8, w0, #182
+; CHECK-NEXT: cmn w8, #20
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
entry:
@@ -295,20 +294,20 @@ ret_true:
define zeroext i1 @test16_2(i16 zeroext %x) align 2 {
; CHECK-SD-LABEL: test16_2:
; CHECK-SD: ; %bb.0: ; %entry
-; CHECK-SD-NEXT: mov w8, #16882 ; =0x41f2
-; CHECK-SD-NEXT: mov w9, #40700 ; =0x9efc
+; CHECK-SD-NEXT: mov w8, #-48654 ; =0xffff41f2
+; CHECK-SD-NEXT: mov w9, #-24836 ; =0xffff9efc
; CHECK-SD-NEXT: add w8, w0, w8
-; CHECK-SD-NEXT: cmp w9, w8, uxth
-; CHECK-SD-NEXT: cset w0, hi
+; CHECK-SD-NEXT: cmp w8, w9
+; CHECK-SD-NEXT: cset w0, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: test16_2:
; CHECK-GI: ; %bb.0: ; %entry
-; CHECK-GI-NEXT: mov w8, #16882 ; =0x41f2
-; CHECK-GI-NEXT: mov w9, #40699 ; =0x9efb
+; CHECK-GI-NEXT: mov w8, #-48654 ; =0xffff41f2
+; CHECK-GI-NEXT: mov w9, #-24837 ; =0xffff9efb
; CHECK-GI-NEXT: add w8, w0, w8
-; CHECK-GI-NEXT: cmp w9, w8, uxth
-; CHECK-GI-NEXT: cset w0, hs
+; CHECK-GI-NEXT: cmp w8, w9
+; CHECK-GI-NEXT: cset w0, ls
; CHECK-GI-NEXT: ret
entry:
%0 = add i16 %x, 16882
@@ -349,20 +348,20 @@ ret_true:
define zeroext i1 @test16_4(i16 zeroext %x) align 2 {
; CHECK-SD-LABEL: test16_4:
; CHECK-SD: ; %bb.0: ; %entry
-; CHECK-SD-NEXT: mov w8, #29985 ; =0x7521
+; CHECK-SD-NEXT: mov w8, #-35551 ; =0xffff7521
; CHECK-SD-NEXT: mov w9, #15676 ; =0x3d3c
; CHECK-SD-NEXT: add w8, w0, w8
-; CHECK-SD-NEXT: cmp w9, w8, uxth
-; CHECK-SD-NEXT: cset w0, lo
+; CHECK-SD-NEXT: cmp w8, w9
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: test16_4:
; CHECK-GI: ; %bb.0: ; %entry
-; CHECK-GI-NEXT: mov w8, #29985 ; =0x7521
+; CHECK-GI-NEXT: mov w8, #-35551 ; =0xffff7521
; CHECK-GI-NEXT: mov w9, #15677 ; =0x3d3d
; CHECK-GI-NEXT: add w8, w0, w8
-; CHECK-GI-NEXT: cmp w9, w8, uxth
-; CHECK-GI-NEXT: cset w0, ls
+; CHECK-GI-NEXT: cmp w8, w9
+; CHECK-GI-NEXT: cset w0, hs
; CHECK-GI-NEXT: ret
entry:
%0 = add i16 %x, -35551
@@ -431,20 +430,20 @@ ret_true:
define zeroext i1 @test16_7(i16 zeroext %x) align 2 {
; CHECK-SD-LABEL: test16_7:
; CHECK-SD: ; %bb.0: ; %entry
-; CHECK-SD-NEXT: mov w8, #9272 ; =0x2438
-; CHECK-SD-NEXT: mov w9, #22619 ; =0x585b
+; CHECK-SD-NEXT: mov w8, #-56264 ; =0xffff2438
+; CHECK-SD-NEXT: mov w9, #-42917 ; =0xffff585b
; CHECK-SD-NEXT: add w8, w0, w8
-; CHECK-SD-NEXT: cmp w9, w8, uxth
-; CHECK-SD-NEXT: cset w0, lo
+; CHECK-SD-NEXT: cmp w8, w9
+; CHECK-SD-NEXT: cset w0, hi
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: test16_7:
; CHECK-GI: ; %bb.0: ; %entry
-; CHECK-GI-NEXT: mov w8, #9272 ; =0x2438
-; CHECK-GI-NEXT: mov w9, #22620 ; =0x585c
+; CHECK-GI-NEXT: mov w8, #-56264 ; =0xffff2438
+; CHECK-GI-NEXT: mov w9, #-42916 ; =0xffff585c
; CHECK-GI-NEXT: add w8, w0, w8
-; CHECK-GI-NEXT: cmp w9, w8, uxth
-; CHECK-GI-NEXT: cset w0, ls
+; CHECK-GI-NEXT: cmp w8, w9
+; CHECK-GI-NEXT: cset w0, hs
; CHECK-GI-NEXT: ret
entry:
%0 = add i16 %x, 9272
@@ -508,16 +507,17 @@ define i64 @pr58109(i8 signext %0) {
define i64 @pr58109b(i8 signext %0, i64 %a, i64 %b) {
; CHECK-SD-LABEL: pr58109b:
; CHECK-SD: ; %bb.0:
-; CHECK-SD-NEXT: add w8, w0, #1
-; CHECK-SD-NEXT: tst w8, #0xfe
-; CHECK-SD-NEXT: csel x0, x1, x2, eq
+; CHECK-SD-NEXT: and w8, w0, #0xff
+; CHECK-SD-NEXT: sub w8, w8, #255
+; CHECK-SD-NEXT: cmn w8, #254
+; CHECK-SD-NEXT: csel x0, x1, x2, lo
; CHECK-SD-NEXT: ret
;
; CHECK-GI-LABEL: pr58109b:
; CHECK-GI: ; %bb.0:
-; CHECK-GI-NEXT: add w8, w0, #1
-; CHECK-GI-NEXT: and w8, w8, #0xff
-; CHECK-GI-NEXT: cmp w8, #2
+; CHECK-GI-NEXT: mov w8, #-255 ; =0xffffff01
+; CHECK-GI-NEXT: add w8, w8, w0, uxtb
+; CHECK-GI-NEXT: cmn w8, #254
; CHECK-GI-NEXT: csel x0, x1, x2, lo
; CHECK-GI-NEXT: ret
%2 = add i8 %0, 1
diff --git a/llvm/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll b/llvm/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
index 56a18b00a974cb..8e464d16dd90c5 100644
--- a/llvm/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
+++ b/llvm/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
@@ -187,10 +187,11 @@ define i1 @add_ulecmp_i16_i8(i16 %x) nounwind {
define i1 @add_ugecmp_i16_i8(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_i16_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: sxtb w8, w0
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, w0, uxth
-; CHECK-NEXT: cset w0, ne
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-65281 // =0xffff00ff
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
+; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp uge i16 %tmp0, 256 ; 1U << 8
@@ -256,10 +257,11 @@ define i1 @add_ugecmp_i64_i8(i64 %x) nounwind {
define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
; CHECK-LABEL: add_ugtcmp_i16_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: sxtb w8, w0
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, w0, uxth
-; CHECK-NEXT: cset w0, ne
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-65281 // =0xffff00ff
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
+; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ugt i16 %tmp0, 255 ; (1U << 8) - 1
@@ -301,9 +303,10 @@ define i1 @add_ugecmp_bad_i16_i8_cmp(i16 %x, i16 %y) nounwind {
define i1 @add_ugecmp_bad_i8_i16(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i8_i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #128
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #127
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-65409 // =0xffff007f
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
@@ -315,9 +318,10 @@ define i1 @add_ugecmp_bad_i8_i16(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i16_i8_c0notpoweroftwo:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #192
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #255
+; CHECK-NEXT: mov w8, #-65344 // =0xffff00c0
+; CHECK-NEXT: mov w9, #-65281 // =0xffff00ff
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 192 ; (1U << (8-1)) + (1U << (8-1-1))
@@ -329,9 +333,10 @@ define i1 @add_ugecmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i16_i8_c1notpoweroftwo:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #128
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #767
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-64769 // =0xffff02ff
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
@@ -343,9 +348,10 @@ define i1 @add_ugecmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_magic(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i16_i8_magic:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #64
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #255
+; CHECK-NEXT: mov w8, #-65472 // =0xffff0040
+; CHECK-NEXT: mov w9, #-65281 // =0xffff00ff
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 64 ; 1U << (8-1-1)
@@ -357,9 +363,10 @@ define i1 @add_ugecmp_bad_i16_i8_magic(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i4(i16 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i16_i4:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #8
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #15
+; CHECK-NEXT: mov w8, #-65528 // =0xffff0008
+; CHECK-NEXT: mov w9, #-65521 // =0xffff000f
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 8 ; 1U << (4-1)
@@ -371,9 +378,12 @@ define i1 @add_ugecmp_bad_i16_i4(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i24_i8(i24 %x) nounwind {
; CHECK-LABEL: add_ugecmp_bad_i24_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #128
-; CHECK-NEXT: and w8, w8, #0xffffff
-; CHECK-NEXT: cmp w8, #255
+; CHECK-NEXT: mov w8, #128 // =0x80
+; CHECK-NEXT: and w9, w0, #0xffffff
+; CHECK-NEXT: movk w8, #65280, lsl #16
+; CHECK-NEXT: add w8, w9, w8
+; CHECK-NEXT: mov w9, #-16776961 // =0xff0000ff
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%tmp0 = add i24 %x, 128 ; 1U << (8-1)
diff --git a/llvm/test/CodeGen/AArch64/signed-truncation-check.ll b/llvm/test/CodeGen/AArch64/signed-truncation-check.ll
index ab42e6463feeed..6a5795ff337dd7 100644
--- a/llvm/test/CodeGen/AArch64/signed-truncation-check.ll
+++ b/llvm/test/CodeGen/AArch64/signed-truncation-check.ll
@@ -200,10 +200,11 @@ define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
define i1 @add_ultcmp_i16_i8(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_i16_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: sxtb w8, w0
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, w0, uxth
-; CHECK-NEXT: cset w0, eq
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-65280 // =0xffff0100
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
+; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i16 %tmp0, 256 ; 1U << 8
@@ -269,10 +270,11 @@ define i1 @add_ultcmp_i64_i8(i64 %x) nounwind {
define i1 @add_ulecmp_i16_i8(i16 %x) nounwind {
; CHECK-LABEL: add_ulecmp_i16_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: sxtb w8, w0
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, w0, uxth
-; CHECK-NEXT: cset w0, eq
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-65280 // =0xffff0100
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
+; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ule i16 %tmp0, 255 ; (1U << 8) - 1
@@ -314,9 +316,9 @@ define i1 @add_ultcmp_bad_i16_i8_cmp(i16 %x, i16 %y) nounwind {
define i1 @add_ultcmp_bad_i8_i16(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i8_i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: and w8, w0, #0xffff
-; CHECK-NEXT: add w8, w8, #128
-; CHECK-NEXT: lsr w0, w8, #16
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: cmn w8, w0, uxth
+; CHECK-NEXT: cset w0, hs
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i16 %tmp0, 128 ; 1U << (8-1)
@@ -327,9 +329,10 @@ define i1 @add_ultcmp_bad_i8_i16(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i16_i8_c0notpoweroftwo:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #192
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #256
+; CHECK-NEXT: mov w8, #-65344 // =0xffff00c0
+; CHECK-NEXT: mov w9, #-65280 // =0xffff0100
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 192 ; (1U << (8-1)) + (1U << (8-1-1))
@@ -341,9 +344,10 @@ define i1 @add_ultcmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i16_i8_c1notpoweroftwo:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #128
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #768
+; CHECK-NEXT: mov w8, #-65408 // =0xffff0080
+; CHECK-NEXT: mov w9, #-64768 // =0xffff0300
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
@@ -355,9 +359,10 @@ define i1 @add_ultcmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_magic(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i16_i8_magic:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #64
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #256
+; CHECK-NEXT: mov w8, #-65472 // =0xffff0040
+; CHECK-NEXT: mov w9, #-65280 // =0xffff0100
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 64 ; 1U << (8-1-1)
@@ -369,9 +374,10 @@ define i1 @add_ultcmp_bad_i16_i8_magic(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i4(i16 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i16_i4:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #8
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #16
+; CHECK-NEXT: mov w8, #-65528 // =0xffff0008
+; CHECK-NEXT: mov w9, #-65520 // =0xffff0010
+; CHECK-NEXT: add w8, w8, w0, uxth
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 8 ; 1U << (4-1)
@@ -383,9 +389,13 @@ define i1 @add_ultcmp_bad_i16_i4(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i24_i8(i24 %x) nounwind {
; CHECK-LABEL: add_ultcmp_bad_i24_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #128
-; CHECK-NEXT: and w8, w8, #0xffffff
-; CHECK-NEXT: cmp w8, #256
+; CHECK-NEXT: mov w8, #128 // =0x80
+; CHECK-NEXT: and w9, w0, #0xffffff
+; CHECK-NEXT: movk w8, #65280, lsl #16
+; CHECK-NEXT: add w8, w9, w8
+; CHECK-NEXT: mov w9, #256 // =0x100
+; CHECK-NEXT: movk w9, #65280, lsl #16
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, lo
; CHECK-NEXT: ret
%tmp0 = add i24 %x, 128 ; 1U << (8-1)
@@ -396,7 +406,7 @@ define i1 @add_ultcmp_bad_i24_i8(i24 %x) nounwind {
define i1 @add_ulecmp_bad_i16_i8(i16 %x) nounwind {
; CHECK-LABEL: add_ulecmp_bad_i16_i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: mov w0, #1
+; CHECK-NEXT: mov w0, #1 // =0x1
; CHECK-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ule i16 %tmp0, -1 ; when we +1 it, it will wrap to 0
diff --git a/llvm/test/CodeGen/AArch64/typepromotion-overflow.ll b/llvm/test/CodeGen/AArch64/typepromotion-overflow.ll
index ccfbf456693d7a..39edc03ced442e 100644
--- a/llvm/test/CodeGen/AArch64/typepromotion-overflow.ll
+++ b/llvm/test/CodeGen/AArch64/typepromotion-overflow.ll
@@ -246,9 +246,8 @@ define i32 @safe_sub_var_imm(ptr nocapture readonly %b) local_unnamed_addr #1 {
; CHECK-LABEL: safe_sub_var_imm:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: ldrb w8, [x0]
-; CHECK-NEXT: add w8, w8, #8
-; CHECK-NEXT: and w8, w8, #0xff
-; CHECK-NEXT: cmp w8, #252
+; CHECK-NEXT: sub w8, w8, #248
+; CHECK-NEXT: cmn w8, #4
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
entry:
diff --git a/llvm/test/CodeGen/AArch64/typepromotion-signed.ll b/llvm/test/CodeGen/AArch64/typepromotion-signed.ll
index 212f02d86850b8..98cc6cb243fdaa 100644
--- a/llvm/test/CodeGen/AArch64/typepromotion-signed.ll
+++ b/llvm/test/CodeGen/AArch64/typepromotion-signed.ll
@@ -30,9 +30,10 @@ define i16 @test_ashr(i16 zeroext %arg) local_unnamed_addr #1 {
define i16 @test_sdiv(i16 zeroext %arg) local_unnamed_addr #1 {
; CHECK-LABEL: test_sdiv:
; CHECK: // %bb.0:
-; CHECK-NEXT: add w8, w0, #1
-; CHECK-NEXT: and w8, w8, #0xffff
-; CHECK-NEXT: cmp w8, #2
+; CHECK-NEXT: mov w8, #-65535 // =0xffff0001
+; CHECK-NEXT: mov w9, #-65534 // =0xffff0002
+; CHECK-NEXT: add w8, w0, w8
+; CHECK-NEXT: cmp w8, w9
; CHECK-NEXT: cset w0, hi
; CHECK-NEXT: ret
%arg.off = add i16 %arg, 1
diff --git a/llvm/test/CodeGen/RISCV/lack-of-signed-truncation-check.ll b/llvm/test/CodeGen/RISCV/lack-of-signed-truncation-check.ll
index 6e3a50542939f1..2030296580184d 100644
--- a/llvm/test/CodeGen/RISCV/lack-of-signed-truncation-check.ll
+++ b/llvm/test/CodeGen/RISCV/lack-of-signed-truncation-check.ll
@@ -484,36 +484,44 @@ define i1 @add_ulecmp_i16_i8(i16 %x) nounwind {
define i1 @add_ugecmp_i16_i8(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_i16_i8:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 255
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_i16_i8:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 255
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_i16_i8:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 255
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_i16_i8:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 255
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp uge i16 %tmp0, 256 ; 1U << 8
@@ -672,36 +680,44 @@ define i1 @add_ugecmp_i64_i8(i64 %x) nounwind {
define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
; RV32I-LABEL: add_ugtcmp_i16_i8:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 255
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugtcmp_i16_i8:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 255
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugtcmp_i16_i8:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 255
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugtcmp_i16_i8:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 255
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ugt i16 %tmp0, 255 ; (1U << 8) - 1
@@ -801,36 +817,44 @@ define i1 @add_ugecmp_bad_i16_i8_cmp(i16 %x, i16 %y) nounwind {
define i1 @add_ugecmp_bad_i8_i16(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_bad_i8_i16:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 128
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 127
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_bad_i8_i16:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 128
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 127
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_bad_i8_i16:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 128
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 127
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_bad_i8_i16:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 128
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 127
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp uge i16 %tmp0, 128 ; 1U << (8-1)
@@ -841,36 +865,44 @@ define i1 @add_ugecmp_bad_i8_i16(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_bad_i16_i8_c0notpoweroftwo:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 192
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 192
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 255
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_bad_i16_i8_c0notpoweroftwo:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 192
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 192
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 255
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_bad_i16_i8_c0notpoweroftwo:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 192
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 192
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 255
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_bad_i16_i8_c0notpoweroftwo:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 192
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 192
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 255
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 192 ; (1U << (8-1)) + (1U << (8-1-1))
%tmp1 = icmp uge i16 %tmp0, 256 ; 1U << 8
@@ -881,36 +913,44 @@ define i1 @add_ugecmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_bad_i16_i8_c1notpoweroftwo:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 768
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 767
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_bad_i16_i8_c1notpoweroftwo:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 768
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 767
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_bad_i16_i8_c1notpoweroftwo:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 768
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 767
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_bad_i16_i8_c1notpoweroftwo:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 768
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 767
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp uge i16 %tmp0, 768 ; (1U << 8)) + (1U << (8+1))
@@ -921,36 +961,44 @@ define i1 @add_ugecmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i8_magic(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_bad_i16_i8_magic:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 64
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 64
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 255
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_bad_i16_i8_magic:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 64
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 64
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 255
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_bad_i16_i8_magic:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 64
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 64
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 255
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_bad_i16_i8_magic:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 64
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 64
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 255
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 64 ; 1U << (8-1-1)
%tmp1 = icmp uge i16 %tmp0, 256 ; 1U << 8
@@ -961,36 +1009,44 @@ define i1 @add_ugecmp_bad_i16_i8_magic(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i16_i4(i16 %x) nounwind {
; RV32I-LABEL: add_ugecmp_bad_i16_i4:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 8
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 16
-; RV32I-NEXT: xori a0, a0, 1
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 8
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 15
+; RV32I-NEXT: sltu a0, a1, a0
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ugecmp_bad_i16_i4:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 8
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 16
-; RV64I-NEXT: xori a0, a0, 1
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 8
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 15
+; RV64I-NEXT: sltu a0, a1, a0
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ugecmp_bad_i16_i4:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 8
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 16
-; RV32ZBB-NEXT: xori a0, a0, 1
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 8
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 15
+; RV32ZBB-NEXT: sltu a0, a1, a0
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ugecmp_bad_i16_i4:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 8
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 16
-; RV64ZBB-NEXT: xori a0, a0, 1
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 8
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 15
+; RV64ZBB-NEXT: sltu a0, a1, a0
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 8 ; 1U << (4-1)
%tmp1 = icmp uge i16 %tmp0, 16 ; 1U << 4
@@ -1001,20 +1057,24 @@ define i1 @add_ugecmp_bad_i16_i4(i16 %x) nounwind {
define i1 @add_ugecmp_bad_i24_i8(i24 %x) nounwind {
; RV32-LABEL: add_ugecmp_bad_i24_i8:
; RV32: # %bb.0:
-; RV32-NEXT: addi a0, a0, 128
; RV32-NEXT: slli a0, a0, 8
; RV32-NEXT: srli a0, a0, 8
-; RV32-NEXT: sltiu a0, a0, 256
-; RV32-NEXT: xori a0, a0, 1
+; RV32-NEXT: lui a1, 1044480
+; RV32-NEXT: addi a2, a1, 128
+; RV32-NEXT: add a0, a0, a2
+; RV32-NEXT: addi a1, a1, 255
+; RV32-NEXT: sltu a0, a1, a0
; RV32-NEXT: ret
;
; RV64-LABEL: add_ugecmp_bad_i24_i8:
; RV64: # %bb.0:
-; RV64-NEXT: addi a0, a0, 128
; RV64-NEXT: slli a0, a0, 40
; RV64-NEXT: srli a0, a0, 40
-; RV64-NEXT: sltiu a0, a0, 256
-; RV64-NEXT: xori a0, a0, 1
+; RV64-NEXT: lui a1, 1044480
+; RV64-NEXT: addiw a2, a1, 128
+; RV64-NEXT: add a0, a0, a2
+; RV64-NEXT: addiw a1, a1, 255
+; RV64-NEXT: sltu a0, a1, a0
; RV64-NEXT: ret
%tmp0 = add i24 %x, 128 ; 1U << (8-1)
%tmp1 = icmp uge i24 %tmp0, 256 ; 1U << 8
diff --git a/llvm/test/CodeGen/RISCV/signed-truncation-check.ll b/llvm/test/CodeGen/RISCV/signed-truncation-check.ll
index de36bcdb910609..4c0d2c70fe4501 100644
--- a/llvm/test/CodeGen/RISCV/signed-truncation-check.ll
+++ b/llvm/test/CodeGen/RISCV/signed-truncation-check.ll
@@ -540,32 +540,44 @@ define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
define i1 @add_ultcmp_i16_i8(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_i16_i8:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 256
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_i16_i8:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 256
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_i16_i8:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 256
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_i16_i8:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 256
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i16 %tmp0, 256 ; 1U << 8
@@ -720,32 +732,44 @@ define i1 @add_ultcmp_i64_i8(i64 %x) nounwind {
define i1 @add_ulecmp_i16_i8(i16 %x) nounwind {
; RV32I-LABEL: add_ulecmp_i16_i8:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 256
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ulecmp_i16_i8:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 256
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ulecmp_i16_i8:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 256
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ulecmp_i16_i8:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 256
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ule i16 %tmp0, 255 ; (1U << 8) - 1
@@ -837,32 +861,40 @@ define i1 @add_ultcmp_bad_i16_i8_cmp(i16 %x, i16 %y) nounwind {
define i1 @add_ultcmp_bad_i8_i16(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_bad_i8_i16:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 128
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a1, a1, 128
+; RV32I-NEXT: add a0, a0, a1
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_bad_i8_i16:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 128
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a1, a1, 128
+; RV64I-NEXT: add a0, a0, a1
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_bad_i8_i16:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 128
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a1, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a1
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_bad_i8_i16:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 128
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a1, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a1
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i16 %tmp0, 128 ; 1U << (8-1)
@@ -873,32 +905,44 @@ define i1 @add_ultcmp_bad_i8_i16(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_bad_i16_i8_c0notpoweroftwo:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 192
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 192
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 256
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_bad_i16_i8_c0notpoweroftwo:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 192
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 192
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 256
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_bad_i16_i8_c0notpoweroftwo:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 192
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 192
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 256
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_bad_i16_i8_c0notpoweroftwo:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 192
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 192
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 256
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 192 ; (1U << (8-1)) + (1U << (8-1-1))
%tmp1 = icmp ult i16 %tmp0, 256 ; 1U << 8
@@ -909,32 +953,44 @@ define i1 @add_ultcmp_bad_i16_i8_c0notpoweroftwo(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_bad_i16_i8_c1notpoweroftwo:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 128
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 768
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 128
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 768
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_bad_i16_i8_c1notpoweroftwo:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 128
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 768
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 128
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 768
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_bad_i16_i8_c1notpoweroftwo:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 128
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 768
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 128
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 768
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_bad_i16_i8_c1notpoweroftwo:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 128
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 768
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 128
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 768
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i16 %tmp0, 768 ; (1U << 8)) + (1U << (8+1))
@@ -945,32 +1001,44 @@ define i1 @add_ultcmp_bad_i16_i8_c1notpoweroftwo(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i8_magic(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_bad_i16_i8_magic:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 64
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 256
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 64
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 256
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_bad_i16_i8_magic:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 64
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 256
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 64
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 256
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_bad_i16_i8_magic:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 64
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 256
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 64
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 256
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_bad_i16_i8_magic:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 64
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 256
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 64
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 256
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 64 ; 1U << (8-1-1)
%tmp1 = icmp ult i16 %tmp0, 256 ; 1U << 8
@@ -981,32 +1049,44 @@ define i1 @add_ultcmp_bad_i16_i8_magic(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i16_i4(i16 %x) nounwind {
; RV32I-LABEL: add_ultcmp_bad_i16_i4:
; RV32I: # %bb.0:
-; RV32I-NEXT: addi a0, a0, 8
; RV32I-NEXT: slli a0, a0, 16
; RV32I-NEXT: srli a0, a0, 16
-; RV32I-NEXT: sltiu a0, a0, 16
+; RV32I-NEXT: lui a1, 1048560
+; RV32I-NEXT: addi a2, a1, 8
+; RV32I-NEXT: add a0, a0, a2
+; RV32I-NEXT: addi a1, a1, 16
+; RV32I-NEXT: sltu a0, a0, a1
; RV32I-NEXT: ret
;
; RV64I-LABEL: add_ultcmp_bad_i16_i4:
; RV64I: # %bb.0:
-; RV64I-NEXT: addi a0, a0, 8
; RV64I-NEXT: slli a0, a0, 48
; RV64I-NEXT: srli a0, a0, 48
-; RV64I-NEXT: sltiu a0, a0, 16
+; RV64I-NEXT: lui a1, 1048560
+; RV64I-NEXT: addiw a2, a1, 8
+; RV64I-NEXT: add a0, a0, a2
+; RV64I-NEXT: addiw a1, a1, 16
+; RV64I-NEXT: sltu a0, a0, a1
; RV64I-NEXT: ret
;
; RV32ZBB-LABEL: add_ultcmp_bad_i16_i4:
; RV32ZBB: # %bb.0:
-; RV32ZBB-NEXT: addi a0, a0, 8
; RV32ZBB-NEXT: zext.h a0, a0
-; RV32ZBB-NEXT: sltiu a0, a0, 16
+; RV32ZBB-NEXT: lui a1, 1048560
+; RV32ZBB-NEXT: addi a2, a1, 8
+; RV32ZBB-NEXT: add a0, a0, a2
+; RV32ZBB-NEXT: addi a1, a1, 16
+; RV32ZBB-NEXT: sltu a0, a0, a1
; RV32ZBB-NEXT: ret
;
; RV64ZBB-LABEL: add_ultcmp_bad_i16_i4:
; RV64ZBB: # %bb.0:
-; RV64ZBB-NEXT: addi a0, a0, 8
; RV64ZBB-NEXT: zext.h a0, a0
-; RV64ZBB-NEXT: sltiu a0, a0, 16
+; RV64ZBB-NEXT: lui a1, 1048560
+; RV64ZBB-NEXT: addiw a2, a1, 8
+; RV64ZBB-NEXT: add a0, a0, a2
+; RV64ZBB-NEXT: addiw a1, a1, 16
+; RV64ZBB-NEXT: sltu a0, a0, a1
; RV64ZBB-NEXT: ret
%tmp0 = add i16 %x, 8 ; 1U << (4-1)
%tmp1 = icmp ult i16 %tmp0, 16 ; 1U << 4
@@ -1017,18 +1097,24 @@ define i1 @add_ultcmp_bad_i16_i4(i16 %x) nounwind {
define i1 @add_ultcmp_bad_i24_i8(i24 %x) nounwind {
; RV32-LABEL: add_ultcmp_bad_i24_i8:
; RV32: # %bb.0:
-; RV32-NEXT: addi a0, a0, 128
; RV32-NEXT: slli a0, a0, 8
; RV32-NEXT: srli a0, a0, 8
-; RV32-NEXT: sltiu a0, a0, 256
+; RV32-NEXT: lui a1, 1044480
+; RV32-NEXT: addi a2, a1, 128
+; RV32-NEXT: add a0, a0, a2
+; RV32-NEXT: addi a1, a1, 256
+; RV32-NEXT: sltu a0, a0, a1
; RV32-NEXT: ret
;
; RV64-LABEL: add_ultcmp_bad_i24_i8:
; RV64: # %bb.0:
-; RV64-NEXT: addi a0, a0, 128
; RV64-NEXT: slli a0, a0, 40
; RV64-NEXT: srli a0, a0, 40
-; RV64-NEXT: sltiu a0, a0, 256
+; RV64-NEXT: lui a1, 1044480
+; RV64-NEXT: addiw a2, a1, 128
+; RV64-NEXT: add a0, a0, a2
+; RV64-NEXT: addiw a1, a1, 256
+; RV64-NEXT: sltu a0, a0, a1
; RV64-NEXT: ret
%tmp0 = add i24 %x, 128 ; 1U << (8-1)
%tmp1 = icmp ult i24 %tmp0, 256 ; 1U << 8
diff --git a/llvm/test/CodeGen/RISCV/typepromotion-overflow.ll b/llvm/test/CodeGen/RISCV/typepromotion-overflow.ll
index 3740dc675949fa..ec7e0ecce80caa 100644
--- a/llvm/test/CodeGen/RISCV/typepromotion-overflow.ll
+++ b/llvm/test/CodeGen/RISCV/typepromotion-overflow.ll
@@ -283,9 +283,8 @@ define i32 @safe_sub_var_imm(ptr nocapture readonly %b) local_unnamed_addr #1 {
; CHECK-LABEL: safe_sub_var_imm:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: lbu a0, 0(a0)
-; CHECK-NEXT: addi a0, a0, 8
-; CHECK-NEXT: andi a0, a0, 255
-; CHECK-NEXT: sltiu a0, a0, 253
+; CHECK-NEXT: addi a0, a0, -248
+; CHECK-NEXT: sltiu a0, a0, -3
; CHECK-NEXT: xori a0, a0, 1
; CHECK-NEXT: ret
entry:
diff --git a/llvm/test/Transforms/TypePromotion/ARM/icmps.ll b/llvm/test/Transforms/TypePromotion/ARM/icmps.ll
index 842aab121b96fb..fb537a1f64705c 100644
--- a/llvm/test/Transforms/TypePromotion/ARM/icmps.ll
+++ b/llvm/test/Transforms/TypePromotion/ARM/icmps.ll
@@ -4,8 +4,9 @@
define i32 @test_ult_254_inc_imm(i8 zeroext %x) {
; CHECK-LABEL: @test_ult_254_inc_imm(
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[ADD:%.*]] = add i8 [[X:%.*]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ult i8 [[ADD]], -2
+; CHECK-NEXT: [[TMP0:%.*]] = zext i8 [[X:%.*]] to i32
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP0]], -255
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[ADD]], -2
; CHECK-NEXT: [[RES:%.*]] = select i1 [[CMP]], i32 35, i32 47
; CHECK-NEXT: ret i32 [[RES]]
;
@@ -368,7 +369,7 @@ if.end:
define i32 @degenerateicmp() {
; CHECK-LABEL: @degenerateicmp(
; CHECK-NEXT: [[TMP1:%.*]] = sub i32 190, 0
-; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i32 225, [[TMP1]]
+; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt i32 -31, [[TMP1]]
; CHECK-NEXT: [[TMP3:%.*]] = select i1 [[TMP2]], i32 1, i32 0
; CHECK-NEXT: ret i32 [[TMP3]]
;
diff --git a/llvm/test/Transforms/TypePromotion/ARM/wrapping.ll b/llvm/test/Transforms/TypePromotion/ARM/wrapping.ll
index 377708cf71134a..78c5e7323ceab3 100644
--- a/llvm/test/Transforms/TypePromotion/ARM/wrapping.ll
+++ b/llvm/test/Transforms/TypePromotion/ARM/wrapping.ll
@@ -89,8 +89,9 @@ define i32 @overflow_add_const_limit(i8 zeroext %a, i8 zeroext %b) {
define i32 @overflow_add_positive_const_limit(i8 zeroext %a) {
; CHECK-LABEL: @overflow_add_positive_const_limit(
-; CHECK-NEXT: [[ADD:%.*]] = add i8 [[A:%.*]], 1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[ADD]], -128
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[A:%.*]] to i32
+; CHECK-NEXT: [[ADD:%.*]] = add i32 [[TMP1]], -255
+; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[ADD]], -128
; CHECK-NEXT: [[RES:%.*]] = select i1 [[CMP]], i32 8, i32 16
; CHECK-NEXT: ret i32 [[RES]]
;
@@ -144,8 +145,9 @@ define i32 @safe_add_underflow_neg(i8 zeroext %a) {
define i32 @overflow_sub_negative_const_limit(i8 zeroext %a) {
; CHECK-LABEL: @overflow_sub_negative_const_limit(
-; CHECK-NEXT: [[SUB:%.*]] = sub i8 [[A:%.*]], -1
-; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i8 [[SUB]], -128
+; CHECK-NEXT: [[TMP1:%.*]] = zext i8 [[A:%.*]] to i32
+; CHECK-NEXT: [[SUB:%.*]] = sub i32 [[TMP1]], 255
+; CHECK-NEXT: [[CMP:%.*]] = icmp ugt i32 [[SUB]], -128
; CHECK-NEXT: [[RES:%.*]] = select i1 [[CMP]], i32 8, i32 16
; CHECK-NEXT: ret i32 [[RES]]
;
More information about the llvm-commits
mailing list