diff options
| author | Ulrich Weigand <ulrich.weigand@de.ibm.com> | 2012-10-18 13:16:11 +0000 |
|---|---|---|
| committer | Ulrich Weigand <ulrich.weigand@de.ibm.com> | 2012-10-18 13:16:11 +0000 |
| commit | 6c28a7eec8d1794fb5476cd9fe14affb479ed1d9 (patch) | |
| tree | 1f47a43a375ae85f1595dc7ae797c7bd4a094ba2 /test/CodeGen/PowerPC/i64_fp_round.ll | |
| parent | 2360b7ad99eedecaae512373e7be49c2143550cf (diff) | |
| download | external_llvm-6c28a7eec8d1794fb5476cd9fe14affb479ed1d9.zip external_llvm-6c28a7eec8d1794fb5476cd9fe14affb479ed1d9.tar.gz external_llvm-6c28a7eec8d1794fb5476cd9fe14affb479ed1d9.tar.bz2 | |
This patch fixes failures in the SingleSource/Regression/C/uint64_to_float
test case on PowerPC caused by rounding errors when converting from a 64-bit
integer to a single-precision floating point. The reason for this are
double-rounding effects, since on PowerPC we have to convert to an
intermediate double-precision value first, which gets rounded to the
final single-precision result.
The patch fixes the problem by preparing the 64-bit integer so that the
first conversion step to double-precision will always be exact, and the
final rounding step will result in the correctly-rounded single-precision
result. The generated code sequence is equivalent to what GCC would generate.
When -enable-unsafe-fp-math is in effect, that extra effort is omitted
and we accept possible rounding errors (just like GCC does as well).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@166178 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/PowerPC/i64_fp_round.ll')
| -rw-r--r-- | test/CodeGen/PowerPC/i64_fp_round.ll | 27 |
1 files changed, 27 insertions, 0 deletions
diff --git a/test/CodeGen/PowerPC/i64_fp_round.ll b/test/CodeGen/PowerPC/i64_fp_round.ll new file mode 100644 index 0000000..5a0c072 --- /dev/null +++ b/test/CodeGen/PowerPC/i64_fp_round.ll @@ -0,0 +1,27 @@ +; RUN: llc -mcpu=pwr7 < %s | FileCheck %s +target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64" +target triple = "powerpc64-unknown-linux-gnu" + +define float @test(i64 %x) nounwind readnone { +entry: + %conv = sitofp i64 %x to float + ret float %conv +} + +; Verify that we get the code sequence needed to avoid double-rounding. +; Note that only parts of the sequence are checked for here, to allow +; for minor code generation differences. + +; CHECK: sradi [[REGISTER:[0-9]+]], 3, 53 +; CHECK: addi [[REGISTER:[0-9]+]], [[REGISTER]], 1 +; CHECK: cmpldi 0, [[REGISTER]], 1 +; CHECK: isel [[REGISTER:[0-9]+]], {{[0-9]+}}, 3, 1 +; CHECK: std [[REGISTER]], -{{[0-9]+}}(1) + + +; Also check that with -enable-unsafe-fp-math we do not get that extra +; code sequence. Simply verify that there is no "isel" present. + +; RUN: llc -mcpu=pwr7 -enable-unsafe-fp-math < %s | FileCheck %s -check-prefix=UNSAFE +; CHECK-UNSAFE-NOT: isel + |