[flang-commits] [flang] [flang] Use saturated intrinsic for floating point conversions (PR #130686)
Asher Mancinelli via flang-commits
flang-commits at lists.llvm.org
Mon Mar 10 16:46:16 PDT 2025
https://github.com/ashermancinelli created https://github.com/llvm/llvm-project/pull/130686
The saturated floating point conversion intrinsics more closely match the semantics in the standard.
Case 2 of 16.9.100 is
> INT (A [, KIND])
> If A is of type real, there are two cases: if |A| < 1, INT (A) has the value 0; if |A| ≥ 1, INT (A) is the integer whose magnitude is the largest integer that does not exceed the magnitude of A and whose sign is the same as the sign of A.
Currently, converting a floating point value into an integer type too small to hold the constant will be converted to poison in opt, leaving us with garbage:
```
> cat t.f90
program main
real(kind=16) :: f
integer(kind=4) :: i
f=huge(f)
i=f
print *, i
end program main
# current upstream
> for i in `seq 10`; do; ./a.out; done
-862156992
-1497393344
-739096768
-1649494208
1761228608
-1959270592
-746244288
-1629194432
-231217344
382322496
```
With the saturated fptoui/fptosi intrinsics, we get the appropriate values
```
# mine
> flang -O2 ./t.f90 && ./a.out
2147483647
> perl -e 'printf "%d\n", (2 ** 31) - 1'
2147483647
```
One notable difference: NaNs being converted to ints will become zero, unlike current flang (and other compilers).
>From 68ca2073d6ba4a0ed3dbf31f98b3c2231c533655 Mon Sep 17 00:00:00 2001
From: Asher Mancinelli <ashermancinelli at gmail.com>
Date: Thu, 6 Mar 2025 16:45:02 -0800
Subject: [PATCH 1/2] [fircg] Lower to saturated fp to int conversions
fptosi and fptoui llvm instructions become poison during optimization
when the conversion cannot be performed.
The standard mandates that a real converted to an int must be converted
to the largest integer that does not exceed the magnitude of the original
and keeps the same sign. The saturated floating point conversions match
these semantics more closely than the regular conversion instructions.
---
flang/lib/Optimizer/CodeGen/CodeGen.cpp | 18 +-
flang/test/Fir/convert-to-llvm.fir | 10 +-
flang/test/Integration/fp-convert.f90 | 236 ++++++++++++++++++++++++
3 files changed, 255 insertions(+), 9 deletions(-)
create mode 100644 flang/test/Integration/fp-convert.f90
diff --git a/flang/lib/Optimizer/CodeGen/CodeGen.cpp b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
index a2743edd7844a..2302c08fae508 100644
--- a/flang/lib/Optimizer/CodeGen/CodeGen.cpp
+++ b/flang/lib/Optimizer/CodeGen/CodeGen.cpp
@@ -835,10 +835,20 @@ struct ConvertOpConversion : public fir::FIROpConversion<fir::ConvertOp> {
return mlir::success();
}
if (mlir::isa<mlir::IntegerType>(toTy)) {
- if (toTy.isUnsignedInteger())
- rewriter.replaceOpWithNewOp<mlir::LLVM::FPToUIOp>(convert, toTy, op0);
- else
- rewriter.replaceOpWithNewOp<mlir::LLVM::FPToSIOp>(convert, toTy, op0);
+ // NOTE: We are checking the fir type here because toTy is an LLVM type
+ // which is signless, and we need to use the intrinsic that matches the
+ // sign of the output in fir.
+ if (toFirTy.isUnsignedInteger()) {
+ auto intrinsicName =
+ mlir::StringAttr::get(convert.getContext(), "llvm.fptoui.sat");
+ rewriter.replaceOpWithNewOp<mlir::LLVM::CallIntrinsicOp>(
+ convert, toTy, intrinsicName, op0);
+ } else {
+ auto intrinsicName =
+ mlir::StringAttr::get(convert.getContext(), "llvm.fptosi.sat");
+ rewriter.replaceOpWithNewOp<mlir::LLVM::CallIntrinsicOp>(
+ convert, toTy, intrinsicName, op0);
+ }
return mlir::success();
}
} else if (mlir::isa<mlir::IntegerType>(fromTy)) {
diff --git a/flang/test/Fir/convert-to-llvm.fir b/flang/test/Fir/convert-to-llvm.fir
index c7037019ee701..7e0acfc3a8db5 100644
--- a/flang/test/Fir/convert-to-llvm.fir
+++ b/flang/test/Fir/convert-to-llvm.fir
@@ -711,11 +711,11 @@ func.func @convert_from_float(%arg0 : f32) {
// CHECK: %{{.*}} = llvm.fpext %[[ARG0]] : f32 to f64
// CHECK: %{{.*}} = llvm.fpext %[[ARG0]] : f32 to f80
// CHECK: %{{.*}} = llvm.fpext %[[ARG0]] : f32 to f128
-// CHECK: %{{.*}} = llvm.fptosi %[[ARG0]] : f32 to i1
-// CHECK: %{{.*}} = llvm.fptosi %[[ARG0]] : f32 to i8
-// CHECK: %{{.*}} = llvm.fptosi %[[ARG0]] : f32 to i16
-// CHECK: %{{.*}} = llvm.fptosi %[[ARG0]] : f32 to i32
-// CHECK: %{{.*}} = llvm.fptosi %[[ARG0]] : f32 to i64
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i1
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i8
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i16
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i32
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i64
// -----
diff --git a/flang/test/Integration/fp-convert.f90 b/flang/test/Integration/fp-convert.f90
new file mode 100644
index 0000000000000..a042b28827b9a
--- /dev/null
+++ b/flang/test/Integration/fp-convert.f90
@@ -0,0 +1,236 @@
+! RUN: %flang -funsigned %s -o %t && %t | FileCheck %s
+! RUN: %flang -funsigned -emit-llvm -S -o - %s | FileCheck %s --check-prefix=LLVMIR
+
+module fp_convert_m
+ implicit none
+ interface set_and_print
+ module procedure set_and_print_r16
+ module procedure set_and_print_r8
+ end interface
+contains
+ subroutine set_and_print_r16(value)
+ real(kind=16), intent(in) :: value
+ integer(kind=1) :: i8
+ integer(kind=2) :: i16
+ integer(kind=4) :: i32
+ integer(kind=8) :: i64
+ integer(kind=16) :: i128
+ unsigned(kind=1) :: u8
+ unsigned(kind=2) :: u16
+ unsigned(kind=4) :: u32
+ unsigned(kind=8) :: u64
+ unsigned(kind=16) :: u128
+ print *, "Original real(16) value:", value
+ i8 = int(value, kind=1)
+ i16 = int(value, kind=2)
+ i32 = int(value, kind=4)
+ i64 = int(value, kind=8)
+ i128 = int(value, kind=16)
+ u8 = uint(value, kind=1)
+ u16 = uint(value, kind=2)
+ u32 = uint(value, kind=4)
+ u64 = uint(value, kind=8)
+ u128 = uint(value, kind=16)
+ print *, "Converted to 8-bit integer:", i8
+ print *, "Converted to 16-bit integer:", i16
+ print *, "Converted to 32-bit integer:", i32
+ print *, "Converted to 64-bit integer:", i64
+ print *, "Converted to 128-bit integer:", i128
+ print *, "Converted to 8-bit unsigned integer:", u8
+ print *, "Converted to 16-bit unsigned integer:", u16
+ print *, "Converted to 32-bit unsigned integer:", u32
+ print *, "Converted to 64-bit unsigned integer:", u64
+ print *, "Converted to 128-bit unsigned integer:", u128
+ end subroutine
+
+ subroutine set_and_print_r8(value)
+ real(kind=8), intent(in) :: value
+ integer(kind=1) :: i8
+ integer(kind=2) :: i16
+ integer(kind=4) :: i32
+ integer(kind=8) :: i64
+ integer(kind=16) :: i128
+ unsigned(kind=1) :: u8
+ unsigned(kind=2) :: u16
+ unsigned(kind=4) :: u32
+ unsigned(kind=8) :: u64
+ unsigned(kind=16) :: u128
+ print *, "Original real(8) value:", value
+ i8 = int(value, kind=1)
+ i16 = int(value, kind=2)
+ i32 = int(value, kind=4)
+ i64 = int(value, kind=8)
+ i128 = int(value, kind=16)
+ u8 = uint(value, kind=1)
+ u16 = uint(value, kind=2)
+ u32 = uint(value, kind=4)
+ u64 = uint(value, kind=8)
+ u128 = uint(value, kind=16)
+ print *, "Converted to 8-bit integer:", i8
+ print *, "Converted to 16-bit integer:", i16
+ print *, "Converted to 32-bit integer:", i32
+ print *, "Converted to 64-bit integer:", i64
+ print *, "Converted to 128-bit integer:", i128
+ print *, "Converted to 8-bit unsigned integer:", u8
+ print *, "Converted to 16-bit unsigned integer:", u16
+ print *, "Converted to 32-bit unsigned integer:", u32
+ print *, "Converted to 64-bit unsigned integer:", u64
+ print *, "Converted to 128-bit unsigned integer:", u128
+ end subroutine
+end module fp_convert_m
+
+program fp_convert
+ use ieee_arithmetic, only: ieee_value, ieee_quiet_nan, ieee_positive_inf, ieee_negative_inf
+ use fp_convert_m, only: set_and_print
+ implicit none
+
+ real(kind=8) :: nan, inf, ninf
+ nan = ieee_value(nan, ieee_quiet_nan)
+ inf = ieee_value(inf, ieee_positive_inf)
+ ninf = ieee_value(ninf, ieee_negative_inf)
+
+ call set_and_print(huge(0.0_8))
+ call set_and_print(-huge(0.0_8))
+ call set_and_print(huge(0.0_16))
+ call set_and_print(-huge(0.0_16))
+ call set_and_print(tiny(0.0_8))
+ call set_and_print(-tiny(0.0_8))
+ call set_and_print(tiny(0.0_16))
+ call set_and_print(-tiny(0.0_16))
+ call set_and_print(nan)
+ call set_and_print(inf)
+ call set_and_print(ninf)
+
+end program fp_convert
+
+! LLVMIR: call i8 @llvm.fptosi.sat.i8.f128(fp128 %{{.+}})
+! LLVMIR: call i16 @llvm.fptosi.sat.i16.f128(fp128 %{{.+}})
+! LLVMIR: call i32 @llvm.fptosi.sat.i32.f128(fp128 %{{.+}})
+! LLVMIR: call i64 @llvm.fptosi.sat.i64.f128(fp128 %{{.+}})
+! LLVMIR: call i128 @llvm.fptosi.sat.i128.f128(fp128 %{{.+}})
+! LLVMIR: call i8 @llvm.fptoui.sat.i8.f128(fp128 %{{.+}})
+! LLVMIR: call i16 @llvm.fptoui.sat.i16.f128(fp128 %{{.+}})
+! LLVMIR: call i32 @llvm.fptoui.sat.i32.f128(fp128 %{{.+}})
+! LLVMIR: call i64 @llvm.fptoui.sat.i64.f128(fp128 %{{.+}})
+! LLVMIR: call i128 @llvm.fptoui.sat.i128.f128(fp128 %{{.+}})
+! LLVMIR: call i8 @llvm.fptosi.sat.i8.f64(double %{{.+}})
+! LLVMIR: call i16 @llvm.fptosi.sat.i16.f64(double %{{.+}})
+! LLVMIR: call i32 @llvm.fptosi.sat.i32.f64(double %{{.+}})
+! LLVMIR: call i64 @llvm.fptosi.sat.i64.f64(double %{{.+}})
+! LLVMIR: call i128 @llvm.fptosi.sat.i128.f64(double %{{.+}})
+! LLVMIR: call i8 @llvm.fptoui.sat.i8.f64(double %{{.+}})
+! LLVMIR: call i16 @llvm.fptoui.sat.i16.f64(double %{{.+}})
+! LLVMIR: call i32 @llvm.fptoui.sat.i32.f64(double %{{.+}})
+! LLVMIR: call i64 @llvm.fptoui.sat.i64.f64(double %{{.+}})
+! LLVMIR: call i128 @llvm.fptoui.sat.i128.f64(double %{{.+}})
+
+! CHECK: Converted to 8-bit integer: 127
+! CHECK: Converted to 16-bit integer: 32767
+! CHECK: Converted to 32-bit integer: 2147483647
+! CHECK: Converted to 64-bit integer: 9223372036854775807
+! CHECK: Converted to 128-bit integer: 170141183460469231731687303715884105727
+! CHECK: Converted to 8-bit unsigned integer: 255
+! CHECK: Converted to 16-bit unsigned integer: 65535
+! CHECK: Converted to 32-bit unsigned integer: 4294967295
+! CHECK: Converted to 64-bit unsigned integer: 18446744073709551615
+! CHECK: Converted to 128-bit unsigned integer: 340282366920938463463374607431768211455
+! CHECK: Converted to 8-bit integer: -128
+! CHECK: Converted to 16-bit integer: -32768
+! CHECK: Converted to 32-bit integer: -2147483648
+! CHECK: Converted to 64-bit integer: -9223372036854775808
+! CHECK: Converted to 128-bit integer: -170141183460469231731687303715884105728
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 127
+! CHECK: Converted to 16-bit integer: 32767
+! CHECK: Converted to 32-bit integer: 2147483647
+! CHECK: Converted to 64-bit integer: 9223372036854775807
+! CHECK: Converted to 128-bit integer: 170141183460469231731687303715884105727
+! CHECK: Converted to 8-bit unsigned integer: 255
+! CHECK: Converted to 16-bit unsigned integer: 65535
+! CHECK: Converted to 32-bit unsigned integer: 4294967295
+! CHECK: Converted to 64-bit unsigned integer: 18446744073709551615
+! CHECK: Converted to 128-bit unsigned integer: 340282366920938463463374607431768211455
+! CHECK: Converted to 8-bit integer: -128
+! CHECK: Converted to 16-bit integer: -32768
+! CHECK: Converted to 32-bit integer: -2147483648
+! CHECK: Converted to 64-bit integer: -9223372036854775808
+! CHECK: Converted to 128-bit integer: -170141183460469231731687303715884105728
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 0
+! CHECK: Converted to 16-bit integer: 0
+! CHECK: Converted to 32-bit integer: 0
+! CHECK: Converted to 64-bit integer: 0
+! CHECK: Converted to 128-bit integer: 0
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 0
+! CHECK: Converted to 16-bit integer: 0
+! CHECK: Converted to 32-bit integer: 0
+! CHECK: Converted to 64-bit integer: 0
+! CHECK: Converted to 128-bit integer: 0
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 0
+! CHECK: Converted to 16-bit integer: 0
+! CHECK: Converted to 32-bit integer: 0
+! CHECK: Converted to 64-bit integer: 0
+! CHECK: Converted to 128-bit integer: 0
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 0
+! CHECK: Converted to 16-bit integer: 0
+! CHECK: Converted to 32-bit integer: 0
+! CHECK: Converted to 64-bit integer: 0
+! CHECK: Converted to 128-bit integer: 0
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 0
+! CHECK: Converted to 16-bit integer: 0
+! CHECK: Converted to 32-bit integer: 0
+! CHECK: Converted to 64-bit integer: 0
+! CHECK: Converted to 128-bit integer: 0
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
+! CHECK: Converted to 8-bit integer: 127
+! CHECK: Converted to 16-bit integer: 32767
+! CHECK: Converted to 32-bit integer: 2147483647
+! CHECK: Converted to 64-bit integer: 9223372036854775807
+! CHECK: Converted to 128-bit integer: 170141183460469231731687303715884105727
+! CHECK: Converted to 8-bit unsigned integer: 255
+! CHECK: Converted to 16-bit unsigned integer: 65535
+! CHECK: Converted to 32-bit unsigned integer: 4294967295
+! CHECK: Converted to 64-bit unsigned integer: 18446744073709551615
+! CHECK: Converted to 128-bit unsigned integer: 340282366920938463463374607431768211455
+! CHECK: Converted to 8-bit integer: -128
+! CHECK: Converted to 16-bit integer: -32768
+! CHECK: Converted to 32-bit integer: -2147483648
+! CHECK: Converted to 64-bit integer: -9223372036854775808
+! CHECK: Converted to 128-bit integer: -170141183460469231731687303715884105728
+! CHECK: Converted to 8-bit unsigned integer: 0
+! CHECK: Converted to 16-bit unsigned integer: 0
+! CHECK: Converted to 32-bit unsigned integer: 0
+! CHECK: Converted to 64-bit unsigned integer: 0
+! CHECK: Converted to 128-bit unsigned integer: 0
>From 302db7840c714c4d129f283861172765b6e108a9 Mon Sep 17 00:00:00 2001
From: Asher Mancinelli <ashermancinelli at gmail.com>
Date: Mon, 10 Mar 2025 16:19:50 -0700
Subject: [PATCH 2/2] Add unsigned conversion checks
---
flang/test/Fir/convert-to-llvm.fir | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/flang/test/Fir/convert-to-llvm.fir b/flang/test/Fir/convert-to-llvm.fir
index 7e0acfc3a8db5..2960528fb6c24 100644
--- a/flang/test/Fir/convert-to-llvm.fir
+++ b/flang/test/Fir/convert-to-llvm.fir
@@ -701,6 +701,10 @@ func.func @convert_from_float(%arg0 : f32) {
%7 = fir.convert %arg0 : (f32) -> i16
%8 = fir.convert %arg0 : (f32) -> i32
%9 = fir.convert %arg0 : (f32) -> i64
+ %10 = fir.convert %arg0 : (f32) -> ui8
+ %11 = fir.convert %arg0 : (f32) -> ui16
+ %12 = fir.convert %arg0 : (f32) -> ui32
+ %13 = fir.convert %arg0 : (f32) -> ui64
return
}
@@ -716,6 +720,10 @@ func.func @convert_from_float(%arg0 : f32) {
// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i16
// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i32
// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptosi.sat"(%[[ARG0]]) : (f32) -> i64
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptoui.sat"(%[[ARG0]]) : (f32) -> i8
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptoui.sat"(%[[ARG0]]) : (f32) -> i16
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptoui.sat"(%[[ARG0]]) : (f32) -> i32
+// CHECK: %{{.*}} = llvm.call_intrinsic "llvm.fptoui.sat"(%[[ARG0]]) : (f32) -> i64
// -----
More information about the flang-commits
mailing list