File: emulate-wide-int-canonicalization.mlir

package info (click to toggle)
llvm-toolchain-17 1%3A17.0.6-22
  • links: PTS, VCS
  • area: main
  • in suites: trixie
  • size: 1,799,624 kB
  • sloc: cpp: 6,428,607; ansic: 1,383,196; asm: 793,408; python: 223,504; objc: 75,364; f90: 60,502; lisp: 33,869; pascal: 15,282; sh: 9,684; perl: 7,453; ml: 4,937; awk: 3,523; makefile: 2,889; javascript: 2,149; xml: 888; fortran: 619; cs: 573
file content (14 lines) | stat: -rw-r--r-- 588 bytes parent folder | download | duplicates (15) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
 
// RUN: mlir-opt --arith-emulate-wide-int="widest-int-supported=32" --canonicalize %s | FileCheck %s

// Check that we can fold away the 'hi' part calculation when it is know to be zero.
//
// CHECK-LABEL: func @uitofp_i16_ext_f64
// CHECK-SAME:    ([[ARG:%.+]]: i16) -> f64
// CHECK-NEXT:    [[EXT:%.+]] = arith.extui [[ARG]] : i16 to i32
// CHECK-NEXT:    [[FP:%.+]]  = arith.uitofp [[EXT]] : i32 to f64
// CHECK-NEXT:    return [[FP]] : f64
func.func @uitofp_i16_ext_f64(%a : i16) -> f64 {
  %ext = arith.extui %a : i16 to i64
  %r = arith.uitofp %ext : i64 to f64
  return %r : f64
}