[flang] Unsupported rounding modes #128240
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Two new ieee_round_type values were added in f18 beyond the four values defined in f03 and f08: ieee_away and ieee_other. Contemporary hardware typically does not have support for these rounding modes, so flang does not support them. ieee_support_rounding calls for these values return false. Current generated code handles some attempts to set the rounding mode to one of these unsupported values by setting the mode to ieee_nearest. Update the code to explicitly do this in all cases.
llvmbot
added
flang
|
@llvm/pr-subscribers-flang-fir-hlfir Author: None (vdonaldson) ChangesTwo new ieee_round_type values were added in f18 beyond the four values defined in f03 and f08: ieee_away and ieee_other. Contemporary hardware typically does not have support for these rounding modes, so flang does not support them. ieee_support_rounding calls for these values return false. Current generated code handles some attempts to set the rounding mode to one of these unsupported values by setting the mode to ieee_nearest. Update the code to explicitly do this in all cases. Patch is 47.80 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/128240.diff 3 Files Affected:
diff --git a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
index d98ee58ace2bc..43770877e36a7 100644
--- a/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
+++ b/flang/lib/Optimizer/Builder/IntrinsicCall.cpp
@@ -5450,13 +5450,33 @@ void IntrinsicLibrary::genIeeeSetRoundingMode(
llvm::ArrayRef<fir::ExtendedValue> args) {
// Set the current floating point rounding mode to the value of arg
// ROUNDING_VALUE. Values are llvm.get.rounding encoding values.
- // Generate an error if the value of optional arg RADIX is not 2.
+ // Modes ieee_to_zero, ieee_nearest, ieee_up, and ieee_down are supported.
+ // Modes ieee_away and ieee_other are not supported, and are treated as
+ // ieee_nearest. Generate an error if the optional RADIX arg is not 2.
assert(args.size() == 1 || args.size() == 2);
if (args.size() == 2)
checkRadix(builder, loc, fir::getBase(args[1]), "ieee_set_rounding_mode");
- auto [fieldRef, ignore] = getFieldRef(builder, loc, fir::getBase(args[0]));
+ auto [fieldRef, fieldTy] = getFieldRef(builder, loc, fir::getBase(args[0]));
mlir::func::FuncOp setRound = fir::factory::getLlvmSetRounding(builder);
mlir::Value mode = builder.create<fir::LoadOp>(loc, fieldRef);
+ static_assert(
+ _FORTRAN_RUNTIME_IEEE_TO_ZERO >= 0 &&
+ _FORTRAN_RUNTIME_IEEE_TO_ZERO <= 3 &&
+ _FORTRAN_RUNTIME_IEEE_NEAREST >= 0 &&
+ _FORTRAN_RUNTIME_IEEE_NEAREST <= 3 && _FORTRAN_RUNTIME_IEEE_UP >= 0 &&
+ _FORTRAN_RUNTIME_IEEE_UP <= 3 && _FORTRAN_RUNTIME_IEEE_DOWN >= 0 &&
+ _FORTRAN_RUNTIME_IEEE_DOWN <= 3 && "unexpected rounding mode mapping");
+ mlir::Value mask = builder.create<mlir::arith::ShLIOp>(
+ loc, builder.createAllOnesInteger(loc, fieldTy),
+ builder.createIntegerConstant(loc, fieldTy, 2));
+ mlir::Value modeIsSupported = builder.create<mlir::arith::CmpIOp>(
+ loc, mlir::arith::CmpIPredicate::eq,
+ builder.create<mlir::arith::AndIOp>(loc, mode, mask),
+ builder.createIntegerConstant(loc, fieldTy, 0));
+ mlir::Value nearest = builder.createIntegerConstant(
+ loc, fieldTy, _FORTRAN_RUNTIME_IEEE_NEAREST);
+ mode = builder.create<mlir::arith::SelectOp>(loc, modeIsSupported, mode,
+ nearest);
mode = builder.create<fir::ConvertOp>(
loc, setRound.getFunctionType().getInput(0), mode);
builder.create<fir::CallOp>(loc, setRound, mode);
diff --git a/flang/test/Lower/Intrinsics/ieee_rint_int.f90 b/flang/test/Lower/Intrinsics/ieee_rint_int.f90
index e4b1a5e26f358..86a4aff5005bc 100644
--- a/flang/test/Lower/Intrinsics/ieee_rint_int.f90
+++ b/flang/test/Lower/Intrinsics/ieee_rint_int.f90
@@ -1,4 +1,4 @@
-! RUN: bbc -emit-fir -o - %s | FileCheck %s
+! RUN: bbc -emit-hlfir -o - %s | FileCheck %s
! CHECK-LABEL: c.func @_QQmain
program p
@@ -7,240 +7,275 @@ program p
use ieee_arithmetic, only: ieee_to_zero, ieee_nearest, ieee_up, ieee_away
! CHECK: %[[V_10:[0-9]+]] = fir.alloca i32 {bindc_name = "n", uniq_name = "_QFEn"}
- ! CHECK: %[[V_11:[0-9]+]] = fir.declare %[[V_10]] {uniq_name = "_QFEn"} : (!fir.ref<i32>) -> !fir.ref<i32>
+ ! CHECK: %[[V_11:[0-9]+]]:2 = hlfir.declare %[[V_10]] {uniq_name = "_QFEn"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[V_12:[0-9]+]] = fir.alloca i128 {bindc_name = "n16", uniq_name = "_QFEn16"}
- ! CHECK: %[[V_13:[0-9]+]] = fir.declare %[[V_12]] {uniq_name = "_QFEn16"} : (!fir.ref<i128>) -> !fir.ref<i128>
+ ! CHECK: %[[V_13:[0-9]+]]:2 = hlfir.declare %[[V_12]] {uniq_name = "_QFEn16"} : (!fir.ref<i128>) -> (!fir.ref<i128>, !fir.ref<i128>)
! CHECK: %[[V_14:[0-9]+]] = fir.alloca i16 {bindc_name = "n2", uniq_name = "_QFEn2"}
- ! CHECK: %[[V_15:[0-9]+]] = fir.declare %[[V_14]] {uniq_name = "_QFEn2"} : (!fir.ref<i16>) -> !fir.ref<i16>
+ ! CHECK: %[[V_15:[0-9]+]]:2 = hlfir.declare %[[V_14]] {uniq_name = "_QFEn2"} : (!fir.ref<i16>) -> (!fir.ref<i16>, !fir.ref<i16>)
! CHECK: %[[V_16:[0-9]+]] = fir.alloca i64 {bindc_name = "n8", uniq_name = "_QFEn8"}
- ! CHECK: %[[V_17:[0-9]+]] = fir.declare %[[V_16]] {uniq_name = "_QFEn8"} : (!fir.ref<i64>) -> !fir.ref<i64>
+ ! CHECK: %[[V_17:[0-9]+]]:2 = hlfir.declare %[[V_16]] {uniq_name = "_QFEn8"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>)
! CHECK: %[[V_18:[0-9]+]] = fir.alloca f32 {bindc_name = "x", uniq_name = "_QFEx"}
- ! CHECK: %[[V_19:[0-9]+]] = fir.declare %[[V_18]] {uniq_name = "_QFEx"} : (!fir.ref<f32>) -> !fir.ref<f32>
+ ! CHECK: %[[V_19:[0-9]+]]:2 = hlfir.declare %[[V_18]] {uniq_name = "_QFEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %[[V_20:[0-9]+]] = fir.alloca f16 {bindc_name = "x2", uniq_name = "_QFEx2"}
- ! CHECK: %[[V_21:[0-9]+]] = fir.declare %[[V_20]] {uniq_name = "_QFEx2"} : (!fir.ref<f16>) -> !fir.ref<f16>
+ ! CHECK: %[[V_21:[0-9]+]]:2 = hlfir.declare %[[V_20]] {uniq_name = "_QFEx2"} : (!fir.ref<f16>) -> (!fir.ref<f16>, !fir.ref<f16>)
! CHECK: %[[V_22:[0-9]+]] = fir.alloca bf16 {bindc_name = "x3", uniq_name = "_QFEx3"}
- ! CHECK: %[[V_23:[0-9]+]] = fir.declare %[[V_22]] {uniq_name = "_QFEx3"} : (!fir.ref<bf16>) -> !fir.ref<bf16>
+ ! CHECK: %[[V_23:[0-9]+]]:2 = hlfir.declare %[[V_22]] {uniq_name = "_QFEx3"} : (!fir.ref<bf16>) -> (!fir.ref<bf16>, !fir.ref<bf16>)
! CHECK: %[[V_24:[0-9]+]] = fir.alloca f32 {bindc_name = "x8", uniq_name = "_QFEx8"}
- ! CHECK: %[[V_25:[0-9]+]] = fir.declare %[[V_24]] {uniq_name = "_QFEx8"} : (!fir.ref<f32>) -> !fir.ref<f32>
+ ! CHECK: %[[V_25:[0-9]+]]:2 = hlfir.declare %[[V_24]] {uniq_name = "_QFEx8"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %[[V_26:[0-9]+]] = fir.alloca f32 {bindc_name = "y", uniq_name = "_QFEy"}
- ! CHECK: %[[V_27:[0-9]+]] = fir.declare %[[V_26]] {uniq_name = "_QFEy"} : (!fir.ref<f32>) -> !fir.ref<f32>
+ ! CHECK: %[[V_27:[0-9]+]]:2 = hlfir.declare %[[V_26]] {uniq_name = "_QFEy"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
! CHECK: %[[V_28:[0-9]+]] = fir.alloca f16 {bindc_name = "y2", uniq_name = "_QFEy2"}
- ! CHECK: %[[V_29:[0-9]+]] = fir.declare %[[V_28]] {uniq_name = "_QFEy2"} : (!fir.ref<f16>) -> !fir.ref<f16>
+ ! CHECK: %[[V_29:[0-9]+]]:2 = hlfir.declare %[[V_28]] {uniq_name = "_QFEy2"} : (!fir.ref<f16>) -> (!fir.ref<f16>, !fir.ref<f16>)
! CHECK: %[[V_30:[0-9]+]] = fir.alloca bf16 {bindc_name = "y3", uniq_name = "_QFEy3"}
- ! CHECK: %[[V_31:[0-9]+]] = fir.declare %[[V_30]] {uniq_name = "_QFEy3"} : (!fir.ref<bf16>) -> !fir.ref<bf16>
+ ! CHECK: %[[V_31:[0-9]+]]:2 = hlfir.declare %[[V_30]] {uniq_name = "_QFEy3"} : (!fir.ref<bf16>) -> (!fir.ref<bf16>, !fir.ref<bf16>)
! CHECK: %[[V_32:[0-9]+]] = fir.alloca f32 {bindc_name = "y8", uniq_name = "_QFEy8"}
- ! CHECK: %[[V_33:[0-9]+]] = fir.declare %[[V_32]] {uniq_name = "_QFEy8"} : (!fir.ref<f32>) -> !fir.ref<f32>
+ ! CHECK: %[[V_33:[0-9]+]]:2 = hlfir.declare %[[V_32]] {uniq_name = "_QFEy8"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
integer(2) n2
integer(8) n8
integer(16) n16
real(2) x2, y2
real(3) x3, y3
- ! CHECK: fir.store %cst{{[_0-9]*}} to %[[V_19]] : !fir.ref<f32>
+ ! CHECK: hlfir.assign %cst{{[_0-9]*}} to %[[V_19]]#0 : f32, !fir.ref<f32>
x = -200.7
- ! CHECK: %[[V_34:[0-9]+]] = fir.address_of(@_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.0)
- ! CHECK: %[[V_35:[0-9]+]] = fir.declare %[[V_34]] {fortran_attrs = #fir.var_attrs<parameter>, uniq_name = "_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.0"}
- ! CHECK: %[[V_36:[0-9]+]] = fir.load %[[V_19]] : !fir.ref<f32>
+ ! CHECK: %[[V_34:[0-9]+]] = fir.address_of(@_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.0) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>
+ ! CHECK: %[[V_35:[0-9]+]]:2 = hlfir.declare %[[V_34]]
+ ! CHECK: %[[V_36:[0-9]+]] = fir.load %[[V_19]]#0 : !fir.ref<f32>
! CHECK: %[[V_37:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
! CHECK: %[[V_38:[0-9]+]] = fir.field_index _QM__fortran_builtinsT__builtin_ieee_round_type.mode, !fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>
- ! CHECK: %[[V_39:[0-9]+]] = fir.coordinate_of %[[V_35]], %[[V_38]]
+ ! CHECK: %[[V_39:[0-9]+]] = fir.coordinate_of %[[V_35]]#1, %[[V_38]] : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>, !fir.field) -> !fir.ref<i8>
! CHECK: %[[V_40:[0-9]+]] = fir.load %[[V_39]] : !fir.ref<i8>
- ! CHECK: %[[V_41:[0-9]+]] = fir.convert %[[V_40]] : (i8) -> i32
- ! CHECK: fir.call @llvm.set.rounding(%[[V_41]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_42:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_36]]) fastmath<contract> : (f32) -> f32
+ ! CHECK: %[[V_41:[0-9]+]] = arith.shli %c-1{{.*}}, %c2{{.*}} : i8
+ ! CHECK: %[[V_42:[0-9]+]] = arith.andi %[[V_40]], %[[V_41]] : i8
+ ! CHECK: %[[V_43:[0-9]+]] = arith.cmpi eq, %[[V_42]], %c0{{.*}} : i8
+ ! CHECK: %[[V_44:[0-9]+]] = arith.select %[[V_43]], %[[V_40]], %c1{{.*}} : i8
+ ! CHECK: %[[V_45:[0-9]+]] = fir.convert %[[V_44]] : (i8) -> i32
+ ! CHECK: fir.call @llvm.set.rounding(%[[V_45]]) fastmath<contract> : (i32) -> ()
+ ! CHECK: %[[V_46:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_36]]) fastmath<contract> : (f32) -> f32
+ ! CHECK: %[[V_47:[0-9]+]] = fir.convert %[[V_46]] : (f32) -> f32
! CHECK: fir.call @llvm.set.rounding(%[[V_37]]) fastmath<contract> : (i32) -> ()
- ! CHECK: fir.store %[[V_42]] to %[[V_27]] : !fir.ref<f32>
+ ! CHECK: hlfir.assign %[[V_47]] to %[[V_27]]#0 : f32, !fir.ref<f32>
y = ieee_rint(x, ieee_nearest)
- ! CHECK: %[[V_43:[0-9]+]] = fir.declare %[[V_34]] {fortran_attrs = #fir.var_attrs<parameter>, uniq_name = "_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.0"}
- ! CHECK: %[[V_44:[0-9]+]] = fir.load %[[V_19]] : !fir.ref<f32>
- ! CHECK: %[[V_45:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
- ! CHECK: %[[V_46:[0-9]+]] = fir.coordinate_of %[[V_43]], %[[V_38]]
- ! CHECK: %[[V_47:[0-9]+]] = fir.load %[[V_46]] : !fir.ref<i8>
- ! CHECK: %[[V_48:[0-9]+]] = fir.convert %[[V_47]] : (i8) -> i32
- ! CHECK: fir.call @llvm.set.rounding(%[[V_48]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_49:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_44]]) fastmath<contract> : (f32) -> f32
- ! CHECK: fir.call @llvm.set.rounding(%[[V_45]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_50:[0-9]+]] = fir.convert %c-2147483648{{.*}} : (i32) -> f32
- ! CHECK: %[[V_51:[0-9]+]] = arith.negf %[[V_50]] fastmath<contract> : f32
- ! CHECK: %[[V_52:[0-9]+]] = arith.cmpf oge, %[[V_49]], %[[V_50]] fastmath<contract> : f32
- ! CHECK: %[[V_53:[0-9]+]] = arith.cmpf olt, %[[V_49]], %[[V_51]] fastmath<contract> : f32
- ! CHECK: %[[V_54:[0-9]+]] = arith.andi %[[V_52]], %[[V_53]] : i1
- ! CHECK: %[[V_55:[0-9]+]] = fir.if %[[V_54]] -> (i32) {
- ! CHECK: %[[V_128:[0-9]+]] = arith.cmpf one, %[[V_44]], %[[V_49]] fastmath<contract> : f32
- ! CHECK: fir.if %[[V_128]] {
- ! CHECK: %[[V_130:[0-9]+]] = fir.call @_FortranAMapException(%c32{{.*}}) fastmath<contract> : (i32) -> i32
- ! CHECK: %[[V_131:[0-9]+]] = fir.call @feraiseexcept(%[[V_130]]) fastmath<contract> : (i32) -> i32
+ ! CHECK: %[[V_48:[0-9]+]] = fir.address_of(@_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.0) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>
+ ! CHECK: %[[V_49:[0-9]+]]:2 = hlfir.declare %[[V_48]]
+ ! CHECK: %[[V_50:[0-9]+]] = fir.load %[[V_19]]#0 : !fir.ref<f32>
+ ! CHECK: %[[V_51:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
+ ! CHECK: %[[V_52:[0-9]+]] = fir.field_index _QM__fortran_builtinsT__builtin_ieee_round_type.mode, !fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>
+ ! CHECK: %[[V_53:[0-9]+]] = fir.coordinate_of %[[V_49]]#1, %[[V_52]] : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>, !fir.field) -> !fir.ref<i8>
+ ! CHECK: %[[V_54:[0-9]+]] = fir.load %[[V_53]] : !fir.ref<i8>
+ ! CHECK: %[[V_55:[0-9]+]] = arith.shli %c-1{{.*}}, %c2{{.*}} : i8
+ ! CHECK: %[[V_56:[0-9]+]] = arith.andi %[[V_54]], %[[V_55]] : i8
+ ! CHECK: %[[V_57:[0-9]+]] = arith.cmpi eq, %[[V_56]], %c0{{.*}} : i8
+ ! CHECK: %[[V_58:[0-9]+]] = arith.select %[[V_57]], %[[V_54]], %c1{{.*}} : i8
+ ! CHECK: %[[V_59:[0-9]+]] = fir.convert %[[V_58]] : (i8) -> i32
+ ! CHECK: fir.call @llvm.set.rounding(%[[V_59]]) fastmath<contract> : (i32) -> ()
+ ! CHECK: %[[V_60:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_50]]) fastmath<contract> : (f32) -> f32
+ ! CHECK: %[[V_61:[0-9]+]] = fir.convert %[[V_60]] : (f32) -> f32
+ ! CHECK: fir.call @llvm.set.rounding(%[[V_51]]) fastmath<contract> : (i32) -> ()
+ ! CHECK: %[[V_62:[0-9]+]] = fir.convert %c-2147483648{{.*}} : (i32) -> f32
+ ! CHECK: %[[V_63:[0-9]+]] = arith.negf %[[V_62]] fastmath<contract> : f32
+ ! CHECK: %[[V_64:[0-9]+]] = arith.cmpf oge, %[[V_61]], %[[V_62]] fastmath<contract> : f32
+ ! CHECK: %[[V_65:[0-9]+]] = arith.cmpf olt, %[[V_61]], %[[V_63]] fastmath<contract> : f32
+ ! CHECK: %[[V_66:[0-9]+]] = arith.andi %[[V_64]], %[[V_65]] : i1
+ ! CHECK: %[[V_67:[0-9]+]] = fir.if %[[V_66]] -> (i32) {
+ ! CHECK: %[[V_163:[0-9]+]] = arith.cmpf one, %[[V_50]], %[[V_61]] fastmath<contract> : f32
+ ! CHECK: fir.if %[[V_163]] {
+ ! CHECK: %[[V_165:[0-9]+]] = fir.call @_FortranAMapException(%c32{{.*}}) fastmath<contract> : (i32) -> i32
+ ! CHECK: %[[V_166:[0-9]+]] = fir.call @feraiseexcept(%[[V_165]]) fastmath<contract> : (i32) -> i32
! CHECK: }
- ! CHECK: %[[V_129:[0-9]+]] = fir.convert %[[V_49]] : (f32) -> i32
- ! CHECK: fir.result %[[V_129]] : i32
+ ! CHECK: %[[V_164:[0-9]+]] = fir.convert %[[V_61]] : (f32) -> i32
+ ! CHECK: fir.result %[[V_164]] : i32
! CHECK: } else {
- ! CHECK: %[[V_128:[0-9]+]] = fir.call @_FortranAMapException(%c1{{.*}}) fastmath<contract> : (i32) -> i32
- ! CHECK: %[[V_129:[0-9]+]] = fir.call @feraiseexcept(%[[V_128]]) fastmath<contract> : (i32) -> i32
- ! CHECK: %[[V_130:[0-9]+]] = arith.select %[[V_52]], %c2147483647{{.*}}, %c-2147483648{{.*}} : i32
- ! CHECK: fir.result %[[V_130]] : i32
+ ! CHECK: %[[V_163:[0-9]+]] = fir.call @_FortranAMapException(%c1{{.*}}) fastmath<contract> : (i32) -> i32
+ ! CHECK: %[[V_164:[0-9]+]] = fir.call @feraiseexcept(%[[V_163]]) fastmath<contract> : (i32) -> i32
+ ! CHECK: %[[V_165:[0-9]+]] = arith.select %[[V_64]], %c2147483647{{.*}}, %c-2147483648{{.*}} : i32
+ ! CHECK: fir.result %[[V_165]] : i32
! CHECK: }
- ! CHECK: fir.store %[[V_55]] to %[[V_11]] : !fir.ref<i32>
+ ! CHECK: hlfir.assign %[[V_67]] to %[[V_11]]#0 : i32, !fir.ref<i32>
n = ieee_int(x, ieee_nearest)
! print*, x, ' -> ', y, ' -> ', n
- ! CHECK: fir.store %cst{{[_0-9]*}} to %[[V_21]] : !fir.ref<f16>
+ ! CHECK: hlfir.assign %cst{{[_0-9]*}} to %[[V_21]]#0 : f16, !fir.ref<f16>
x2 = huge(x2)
- ! CHECK: %[[V_56:[0-9]+]] = fir.address_of(@_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.1)
- ! CHECK: %[[V_57:[0-9]+]] = fir.declare %[[V_56]] {fortran_attrs = #fir.var_attrs<parameter>, uniq_name = "_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.1"}
- ! CHECK: %[[V_58:[0-9]+]] = fir.load %[[V_21]] : !fir.ref<f16>
- ! CHECK: %[[V_59:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
- ! CHECK: %[[V_60:[0-9]+]] = fir.coordinate_of %[[V_57]], %[[V_38]]
- ! CHECK: %[[V_61:[0-9]+]] = fir.load %[[V_60]] : !fir.ref<i8>
- ! CHECK: %[[V_62:[0-9]+]] = fir.convert %[[V_61]] : (i8) -> i32
- ! CHECK: fir.call @llvm.set.rounding(%[[V_62]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_63:[0-9]+]] = fir.convert %[[V_58]] : (f16) -> f32
- ! CHECK: %[[V_64:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_63]]) fastmath<contract> : (f32) -> f32
- ! CHECK: %[[V_65:[0-9]+]] = fir.convert %[[V_64]] : (f32) -> f16
- ! CHECK: fir.call @llvm.set.rounding(%[[V_59]]) fastmath<contract> : (i32) -> ()
- ! CHECK: fir.store %[[V_65]] to %[[V_29]] : !fir.ref<f16>
+ ! CHECK: %[[V_68:[0-9]+]] = fir.address_of(@_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.1) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>
+ ! CHECK: %[[V_69:[0-9]+]]:2 = hlfir.declare %[[V_68]]
+ ! CHECK: %[[V_70:[0-9]+]] = fir.load %[[V_21]]#0 : !fir.ref<f16>
+ ! CHECK: %[[V_71:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
+ ! CHECK: %[[V_72:[0-9]+]] = fir.field_index _QM__fortran_builtinsT__builtin_ieee_round_type.mode, !fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>
+ ! CHECK: %[[V_73:[0-9]+]] = fir.coordinate_of %[[V_69]]#1, %[[V_72]] : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_ieee_round_type{_QM__fortran_builtinsT__builtin_ieee_round_type.mode:i8}>>, !fir.field) -> !fir.ref<i8>
+ ! CHECK: %[[V_74:[0-9]+]] = fir.load %[[V_73]] : !fir.ref<i8>
+ ! CHECK: %[[V_75:[0-9]+]] = arith.shli %c-1{{.*}}, %c2{{.*}} : i8
+ ! CHECK: %[[V_76:[0-9]+]] = arith.andi %[[V_74]], %[[V_75]] : i8
+ ! CHECK: %[[V_77:[0-9]+]] = arith.cmpi eq, %[[V_76]], %c0{{.*}} : i8
+ ! CHECK: %[[V_78:[0-9]+]] = arith.select %[[V_77]], %[[V_74]], %c1{{.*}} : i8
+ ! CHECK: %[[V_79:[0-9]+]] = fir.convert %[[V_78]] : (i8) -> i32
+ ! CHECK: fir.call @llvm.set.rounding(%[[V_79]]) fastmath<contract> : (i32) -> ()
+ ! CHECK: %[[V_80:[0-9]+]] = fir.convert %[[V_70]] : (f16) -> f32
+ ! CHECK: %[[V_81:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_80]]) fastmath<contract> : (f32) -> f32
+ ! CHECK: %[[V_82:[0-9]+]] = fir.convert %[[V_81]] : (f32) -> f16
+ ! CHECK: fir.call @llvm.set.rounding(%[[V_71]]) fastmath<contract> : (i32) -> ()
+ ! CHECK: hlfir.assign %[[V_82]] to %[[V_29]]#0 : f16, !fir.ref<f16>
y2 = ieee_rint(x2, ieee_up)
- ! CHECK: %[[V_66:[0-9]+]] = fir.declare %[[V_56]] {fortran_attrs = #fir.var_attrs<parameter>, uniq_name = "_QQro._QM__fortran_builtinsT__builtin_ieee_round_type.1"}
- ! CHECK: %[[V_67:[0-9]+]] = fir.load %[[V_21]] : !fir.ref<f16>
- ! CHECK: %[[V_68:[0-9]+]] = fir.call @llvm.get.rounding() fastmath<contract> : () -> i32
- ! CHECK: %[[V_69:[0-9]+]] = fir.coordinate_of %[[V_66]], %[[V_38]]
- ! CHECK: %[[V_70:[0-9]+]] = fir.load %[[V_69]] : !fir.ref<i8>
- ! CHECK: %[[V_71:[0-9]+]] = fir.convert %[[V_70]] : (i8) -> i32
- ! CHECK: fir.call @llvm.set.rounding(%[[V_71]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_72:[0-9]+]] = fir.convert %[[V_67]] : (f16) -> f32
- ! CHECK: %[[V_73:[0-9]+]] = fir.call @llvm.nearbyint.f32(%[[V_72]]) fastmath<contract> : (f32) -> f32
- ! CHECK: %[[V_74:[0-9]+]] = fir.convert %[[V_73]] : (f32) -> f16
- ! CHECK: fir.call @llvm.set.rounding(%[[V_68]]) fastmath<contract> : (i32) -> ()
- ! CHECK: %[[V_75:[0-9]+]] = fir.convert %c-9223372036854775808{{.*}} : (i64) -> f16
- ! CHECK: %[[V_76:[0-9]+]] = arith.negf %[[V_75]] fastmath<contract> : f16
- ! CHECK: %[[V_77:[0-9]+]] = arith.cmpf oge, %[[V_74]], %[[V_75]] fastmath<contract> : f16
- ! CHECK: %[[V_78:[0-9]+]] = arith.cmpf olt, %[[V_74]], %[[V_76]] fastmath<contract> : f16
- ! CHECK: %[[V_79:[0-9]+]] = arith.andi %[[V_77]], %[[V_78]] : i1
- ! CHECK: %[[V_80:[0-9]+]] = fir.if %[[V_79]] -> (i64) {
- ! CHECK: %[[V_128:[0-9]+]] = arith.cmpf one, %[[V_67]], %[[V_74]] fastmath<contract> : f16
- ! CHECK: fir.if %[[V_128]] {
- ! CHECK: %[[V_130:[0-9]+]] = fir.call @_FortranAMapException(%c32{{.*}}) fastmath<contract> : (i32) -> i32
- ! CHECK: %[[V_131:[0-...
[truncated]
|
jeanPerier
approved these changes
Feb 24, 2025
| // Generate an error if the value of optional arg RADIX is not 2. | ||
| // Modes ieee_to_zero, ieee_nearest, ieee_up, and ieee_down are supported. | ||
| // Modes ieee_away and ieee_other are not supported, and are treated as | ||
| // ieee_nearest. Generate an error if the optional RADIX arg is not 2. |
There was a problem hiding this comment.
Can you document this for the user in Extension.md or Intrinsics.md? (not sure which document makes the most sense).
There was a problem hiding this comment.
I've added a short item in Extensions.md.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Two new ieee_round_type values were added in f18 beyond the four values defined in f03 and f08: ieee_away and ieee_other. Contemporary hardware typically does not have support for these rounding modes, so flang does not support them. ieee_support_rounding calls for these values return false. Current generated code handles some attempts to set the rounding mode to one of these unsupported values by setting the mode to ieee_nearest. Update the code to explicitly do this in all cases.