llvm-6502/test/CodeGen/PowerPC/2008-07-17-Fneg.ll
Dan Gohman ae3a0be92e Split the Add, Sub, and Mul instruction opcodes into separate
integer and floating-point opcodes, introducing
FAdd, FSub, and FMul.

For now, the AsmParser, BitcodeReader, and IRBuilder all preserve
backwards compatability, and the Core LLVM APIs preserve backwards
compatibility for IR producers. Most front-ends won't need to change
immediately.

This implements the first step of the plan outlined here:
http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72897 91177308-0d34-0410-b5e6-96231b3b80d8
2009-06-04 22:49:04 +00:00

19 lines
626 B
LLVM

; RUN: llvm-as < %s | llc
target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128"
target triple = "powerpc-apple-darwin9"
define hidden i64 @__fixunstfdi(ppc_fp128 %a) nounwind {
entry:
br i1 false, label %bb3, label %bb4
bb3: ; preds = %entry
fsub ppc_fp128 0xM80000000000000000000000000000000, 0xM00000000000000000000000000000000 ; <ppc_fp128>:0 [#uses=1]
fptoui ppc_fp128 %0 to i32 ; <i32>:1 [#uses=1]
zext i32 %1 to i64 ; <i64>:2 [#uses=1]
sub i64 0, %2 ; <i64>:3 [#uses=1]
ret i64 %3
bb4: ; preds = %entry
ret i64 0
}