Skip to content

Commit 3f1f08f

Browse files
LebedevRILebedevRI
committed
Revert @llvm.isnan intrinsic patchset.
Please refer to https://lists.llvm.org/pipermail/llvm-dev/2021-September/152440.html (and that whole thread.) TLDR: the original patch had no prior RFC, yet it had some changes that really need a proper RFC discussion. It won't be productive to discuss such an RFC, once it's actually posted, while said patch is already committed, because that introduces bias towards already-committed stuff, and the tree is potentially in broken state meanwhile. While the end result of discussion may lead back to the current design, it may also not lead to the current design. Therefore i take it upon myself to revert the tree back to last known good state. This reverts commit 4c4093e. This reverts commit 0a2b1ba. This reverts commit d987371. This reverts commit 791006f. This reverts commit c22b64e. This reverts commit 72ebcd3. This reverts commit 5fa6039. This reverts commit 9efda54. This reverts commit 94d3ff0.
1 parent 9722e8f commit 3f1f08f

39 files changed

+158
-2997
lines changed

‎clang/lib/CodeGen/CGBuiltin.cpp

+24-4
Original file line numberDiff line numberDiff line change
@@ -3068,17 +3068,37 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
30683068
// ZExt bool to int type.
30693069
return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType())));
30703070
}
3071-
30723071
case Builtin::BI__builtin_isnan: {
30733072
CodeGenFunction::CGFPOptionsRAII FPOptsRAII(*this, E);
30743073
Value *V = EmitScalarExpr(E->getArg(0));
3074+
llvm::Type *Ty = V->getType();
3075+
const llvm::fltSemantics &Semantics = Ty->getFltSemantics();
3076+
if (!Builder.getIsFPConstrained() ||
3077+
Builder.getDefaultConstrainedExcept() == fp::ebIgnore ||
3078+
!Ty->isIEEE()) {
3079+
V = Builder.CreateFCmpUNO(V, V, "cmp");
3080+
return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
3081+
}
30753082

30763083
if (Value *Result = getTargetHooks().testFPKind(V, BuiltinID, Builder, CGM))
30773084
return RValue::get(Result);
30783085

3079-
Function *F = CGM.getIntrinsic(Intrinsic::isnan, V->getType());
3080-
Value *Call = Builder.CreateCall(F, V);
3081-
return RValue::get(Builder.CreateZExt(Call, ConvertType(E->getType())));
3086+
// NaN has all exp bits set and a non zero significand. Therefore:
3087+
// isnan(V) == ((exp mask - (abs(V) & exp mask)) < 0)
3088+
unsigned bitsize = Ty->getScalarSizeInBits();
3089+
llvm::IntegerType *IntTy = Builder.getIntNTy(bitsize);
3090+
Value *IntV = Builder.CreateBitCast(V, IntTy);
3091+
APInt AndMask = APInt::getSignedMaxValue(bitsize);
3092+
Value *AbsV =
3093+
Builder.CreateAnd(IntV, llvm::ConstantInt::get(IntTy, AndMask));
3094+
APInt ExpMask = APFloat::getInf(Semantics).bitcastToAPInt();
3095+
Value *Sub =
3096+
Builder.CreateSub(llvm::ConstantInt::get(IntTy, ExpMask), AbsV);
3097+
// V = sign bit (Sub) <=> V = (Sub < 0)
3098+
V = Builder.CreateLShr(Sub, llvm::ConstantInt::get(IntTy, bitsize - 1));
3099+
if (bitsize > 32)
3100+
V = Builder.CreateTrunc(V, ConvertType(E->getType()));
3101+
return RValue::get(V);
30823102
}
30833103

30843104
case Builtin::BI__builtin_matrix_transpose: {

‎clang/test/CodeGen/X86/strictfp_builtins.c

+20-17
Original file line numberDiff line numberDiff line change
@@ -17,7 +17,7 @@ int printf(const char *, ...);
1717
// CHECK-NEXT: store i32 [[X:%.*]], i32* [[X_ADDR]], align 4
1818
// CHECK-NEXT: [[TMP0:%.*]] = load i8*, i8** [[STR_ADDR]], align 8
1919
// CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[X_ADDR]], align 4
20-
// CHECK-NEXT: [[CALL:%.*]] = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str, i64 0, i64 0), i8* [[TMP0]], i32 [[TMP1]]) #[[ATTR3:[0-9]+]]
20+
// CHECK-NEXT: [[CALL:%.*]] = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str, i64 0, i64 0), i8* [[TMP0]], i32 [[TMP1]]) [[ATTR4:#.*]]
2121
// CHECK-NEXT: ret void
2222
//
2323
void p(char *str, int x) {
@@ -29,13 +29,13 @@ void p(char *str, int x) {
2929
// CHECK-LABEL: @test_long_double_isinf(
3030
// CHECK-NEXT: entry:
3131
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca x86_fp80, align 16
32-
// CHECK-NEXT: store x86_fp80 [[LD:%.*]], x86_fp80* [[LD_ADDR]], align 16
32+
// CHECK-NEXT: store x86_fp80 [[D:%.*]], x86_fp80* [[LD_ADDR]], align 16
3333
// CHECK-NEXT: [[TMP0:%.*]] = load x86_fp80, x86_fp80* [[LD_ADDR]], align 16
34-
// CHECK-NEXT: [[TMP1:%.*]] = bitcast x86_fp80 [[TMP0]] to i80
35-
// CHECK-NEXT: [[TMP2:%.*]] = shl i80 [[TMP1]], 1
36-
// CHECK-NEXT: [[TMP3:%.*]] = icmp eq i80 [[TMP2]], -18446744073709551616
37-
// CHECK-NEXT: [[TMP4:%.*]] = zext i1 [[TMP3]] to i32
38-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.1, i64 0, i64 0), i32 [[TMP4]]) #[[ATTR3]]
34+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast x86_fp80 [[TMP0]] to i80
35+
// CHECK-NEXT: [[SHL1:%.*]] = shl i80 [[BITCAST]], 1
36+
// CHECK-NEXT: [[CMP:%.*]] = icmp eq i80 [[SHL1]], -18446744073709551616
37+
// CHECK-NEXT: [[RES:%.*]] = zext i1 [[CMP]] to i32
38+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.[[#STRID:1]], i64 0, i64 0), i32 [[RES]]) [[ATTR4]]
3939
// CHECK-NEXT: ret void
4040
//
4141
void test_long_double_isinf(long double ld) {
@@ -47,13 +47,13 @@ void test_long_double_isinf(long double ld) {
4747
// CHECK-LABEL: @test_long_double_isfinite(
4848
// CHECK-NEXT: entry:
4949
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca x86_fp80, align 16
50-
// CHECK-NEXT: store x86_fp80 [[LD:%.*]], x86_fp80* [[LD_ADDR]], align 16
50+
// CHECK-NEXT: store x86_fp80 [[D:%.*]], x86_fp80* [[LD_ADDR]], align 16
5151
// CHECK-NEXT: [[TMP0:%.*]] = load x86_fp80, x86_fp80* [[LD_ADDR]], align 16
52-
// CHECK-NEXT: [[TMP1:%.*]] = bitcast x86_fp80 [[TMP0]] to i80
53-
// CHECK-NEXT: [[TMP2:%.*]] = shl i80 [[TMP1]], 1
54-
// CHECK-NEXT: [[TMP3:%.*]] = icmp ult i80 [[TMP2]], -18446744073709551616
55-
// CHECK-NEXT: [[TMP4:%.*]] = zext i1 [[TMP3]] to i32
56-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str.2, i64 0, i64 0), i32 [[TMP4]]) #[[ATTR3]]
52+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast x86_fp80 [[TMP0]] to i80
53+
// CHECK-NEXT: [[SHL1:%.*]] = shl i80 [[BITCAST]], 1
54+
// CHECK-NEXT: [[CMP:%.*]] = icmp ult i80 [[SHL1]], -18446744073709551616
55+
// CHECK-NEXT: [[RES:%.*]] = zext i1 [[CMP]] to i32
56+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str.[[#STRID:STRID+1]], i64 0, i64 0), i32 [[RES]]) [[ATTR4]]
5757
// CHECK-NEXT: ret void
5858
//
5959
void test_long_double_isfinite(long double ld) {
@@ -65,11 +65,14 @@ void test_long_double_isfinite(long double ld) {
6565
// CHECK-LABEL: @test_long_double_isnan(
6666
// CHECK-NEXT: entry:
6767
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca x86_fp80, align 16
68-
// CHECK-NEXT: store x86_fp80 [[LD:%.*]], x86_fp80* [[LD_ADDR]], align 16
68+
// CHECK-NEXT: store x86_fp80 [[D:%.*]], x86_fp80* [[LD_ADDR]], align 16
6969
// CHECK-NEXT: [[TMP0:%.*]] = load x86_fp80, x86_fp80* [[LD_ADDR]], align 16
70-
// CHECK-NEXT: [[TMP1:%.*]] = call i1 @llvm.isnan.f80(x86_fp80 [[TMP0]]) #[[ATTR3]]
71-
// CHECK-NEXT: [[TMP2:%.*]] = zext i1 [[TMP1]] to i32
72-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.3, i64 0, i64 0), i32 [[TMP2]]) #[[ATTR3]]
70+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast x86_fp80 [[TMP0]] to i80
71+
// CHECK-NEXT: [[ABS:%.*]] = and i80 [[BITCAST]], 604462909807314587353087
72+
// CHECK-NEXT: [[TMP1:%.*]] = sub i80 604453686435277732577280, [[ABS]]
73+
// CHECK-NEXT: [[ISNAN:%.*]] = lshr i80 [[TMP1]], 79
74+
// CHECK-NEXT: [[RES:%.*]] = trunc i80 [[ISNAN]] to i32
75+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.[[#STRID:STRID+1]], i64 0, i64 0), i32 [[RES]]) [[ATTR4]]
7376
// CHECK-NEXT: ret void
7477
//
7578
void test_long_double_isnan(long double ld) {

‎clang/test/CodeGen/aarch64-strictfp-builtins.c

+20-18
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,3 @@
1-
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
21
// RUN: %clang_cc1 %s -emit-llvm -ffp-exception-behavior=maytrap -fexperimental-strict-floating-point -o - -triple arm64-none-linux-gnu | FileCheck %s
32

43
// Test that the constrained intrinsics are picking up the exception
@@ -16,7 +15,7 @@ int printf(const char *, ...);
1615
// CHECK-NEXT: store i32 [[X:%.*]], i32* [[X_ADDR]], align 4
1716
// CHECK-NEXT: [[TMP0:%.*]] = load i8*, i8** [[STR_ADDR]], align 8
1817
// CHECK-NEXT: [[TMP1:%.*]] = load i32, i32* [[X_ADDR]], align 4
19-
// CHECK-NEXT: [[CALL:%.*]] = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str, i64 0, i64 0), i8* [[TMP0]], i32 [[TMP1]]) #[[ATTR3:[0-9]+]]
18+
// CHECK-NEXT: [[CALL:%.*]] = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str, i64 0, i64 0), i8* [[TMP0]], i32 [[TMP1]]) [[ATTR4:#.*]]
2019
// CHECK-NEXT: ret void
2120
//
2221
void p(char *str, int x) {
@@ -28,13 +27,13 @@ void p(char *str, int x) {
2827
// CHECK-LABEL: @test_long_double_isinf(
2928
// CHECK-NEXT: entry:
3029
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca fp128, align 16
31-
// CHECK-NEXT: store fp128 [[LD:%.*]], fp128* [[LD_ADDR]], align 16
30+
// CHECK-NEXT: store fp128 [[D:%.*]], fp128* [[LD_ADDR]], align 16
3231
// CHECK-NEXT: [[TMP0:%.*]] = load fp128, fp128* [[LD_ADDR]], align 16
33-
// CHECK-NEXT: [[TMP1:%.*]] = bitcast fp128 [[TMP0]] to i128
34-
// CHECK-NEXT: [[TMP2:%.*]] = shl i128 [[TMP1]], 1
35-
// CHECK-NEXT: [[TMP3:%.*]] = icmp eq i128 [[TMP2]], -10384593717069655257060992658440192
36-
// CHECK-NEXT: [[TMP4:%.*]] = zext i1 [[TMP3]] to i32
37-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.1, i64 0, i64 0), i32 [[TMP4]]) #[[ATTR3]]
32+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast fp128 [[TMP0]] to i128
33+
// CHECK-NEXT: [[SHL1:%.*]] = shl i128 [[BITCAST]], 1
34+
// CHECK-NEXT: [[CMP:%.*]] = icmp eq i128 [[SHL1]], -10384593717069655257060992658440192
35+
// CHECK-NEXT: [[RES:%.*]] = zext i1 [[CMP]] to i32
36+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.[[#STRID:1]], i64 0, i64 0), i32 [[RES]]) [[ATTR4]]
3837
// CHECK-NEXT: ret void
3938
//
4039
void test_long_double_isinf(long double ld) {
@@ -46,13 +45,13 @@ void test_long_double_isinf(long double ld) {
4645
// CHECK-LABEL: @test_long_double_isfinite(
4746
// CHECK-NEXT: entry:
4847
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca fp128, align 16
49-
// CHECK-NEXT: store fp128 [[LD:%.*]], fp128* [[LD_ADDR]], align 16
48+
// CHECK-NEXT: store fp128 [[D:%.*]], fp128* [[LD_ADDR]], align 16
5049
// CHECK-NEXT: [[TMP0:%.*]] = load fp128, fp128* [[LD_ADDR]], align 16
51-
// CHECK-NEXT: [[TMP1:%.*]] = bitcast fp128 [[TMP0]] to i128
52-
// CHECK-NEXT: [[TMP2:%.*]] = shl i128 [[TMP1]], 1
53-
// CHECK-NEXT: [[TMP3:%.*]] = icmp ult i128 [[TMP2]], -10384593717069655257060992658440192
54-
// CHECK-NEXT: [[TMP4:%.*]] = zext i1 [[TMP3]] to i32
55-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str.2, i64 0, i64 0), i32 [[TMP4]]) #[[ATTR3]]
50+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast fp128 [[TMP0]] to i128
51+
// CHECK-NEXT: [[SHL1:%.*]] = shl i128 [[BITCAST]], 1
52+
// CHECK-NEXT: [[CMP:%.*]] = icmp ult i128 [[SHL1]], -10384593717069655257060992658440192
53+
// CHECK-NEXT: [[RES:%.*]] = zext i1 [[CMP]] to i32
54+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([13 x i8], [13 x i8]* @.str.[[#STRID:STRID+1]], i64 0, i64 0), i32 [[RES]]) [[ATTR4]]
5655
// CHECK-NEXT: ret void
5756
//
5857
void test_long_double_isfinite(long double ld) {
@@ -64,11 +63,14 @@ void test_long_double_isfinite(long double ld) {
6463
// CHECK-LABEL: @test_long_double_isnan(
6564
// CHECK-NEXT: entry:
6665
// CHECK-NEXT: [[LD_ADDR:%.*]] = alloca fp128, align 16
67-
// CHECK-NEXT: store fp128 [[LD:%.*]], fp128* [[LD_ADDR]], align 16
66+
// CHECK-NEXT: store fp128 [[D:%.*]], fp128* [[LD_ADDR]], align 16
6867
// CHECK-NEXT: [[TMP0:%.*]] = load fp128, fp128* [[LD_ADDR]], align 16
69-
// CHECK-NEXT: [[TMP1:%.*]] = call i1 @llvm.isnan.f128(fp128 [[TMP0]]) #[[ATTR3]]
70-
// CHECK-NEXT: [[TMP2:%.*]] = zext i1 [[TMP1]] to i32
71-
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.3, i64 0, i64 0), i32 [[TMP2]]) #[[ATTR3]]
68+
// CHECK-NEXT: [[BITCAST:%.*]] = bitcast fp128 [[TMP0]] to i128
69+
// CHECK-NEXT: [[ABS:%.*]] = and i128 [[BITCAST]], 170141183460469231731687303715884105727
70+
// CHECK-NEXT: [[TMP1:%.*]] = sub i128 170135991163610696904058773219554885632, [[ABS]]
71+
// CHECK-NEXT: [[ISNAN:%.*]] = lshr i128 [[TMP1]], 127
72+
// CHECK-NEXT: [[RES:%.*]] = trunc i128 [[ISNAN]] to i32
73+
// CHECK-NEXT: call void @p(i8* getelementptr inbounds ([10 x i8], [10 x i8]* @.str.[[#STRID:STRID+1]], i64 0, i64 0), i32 [[RES]])
7274
// CHECK-NEXT: ret void
7375
//
7476
void test_long_double_isnan(long double ld) {

0 commit comments

Comments
 (0)