1 ; RUN: llc -O3 -disable-peephole -mcpu=corei7-avx -mattr=+avx < %s | FileCheck %s
\r
3 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
\r
4 target triple = "x86_64-unknown-unknown"
\r
6 ; Stack reload folding tests - we use the 'big vectors' pattern to guarantee spilling to stack.
\r
8 ; Many of these tests are primarily to check memory folding with specific instructions. Using a basic
\r
9 ; load/cvt/store pattern to test for this would mean that it wouldn't be the memory folding code thats
\r
10 ; being tested - the load-execute version of the instruction from the tables would be matched instead.
\r
12 define void @stack_fold_vmulpd(<64 x double>* %a, <64 x double>* %b, <64 x double>* %c) {
\r
13 ;CHECK: vmulpd {{[0-9]*}}(%rsp), {{%ymm[0-9][0-9]*}}, {{%ymm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload
\r
14 %1 = load <64 x double>* %a
\r
15 %2 = load <64 x double>* %b
\r
16 %3 = fadd <64 x double> %1, %2
\r
17 %4 = fsub <64 x double> %1, %2
\r
18 %5 = fmul <64 x double> %3, %4
\r
19 store <64 x double> %5, <64 x double>* %c
\r
23 define void @stack_fold_cvtdq2ps(<128 x i32>* %a, <128 x i32>* %b, <128 x float>* %c) {
\r
24 ;CHECK: vcvtdq2ps {{[0-9]*}}(%rsp), {{%ymm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload
\r
25 %1 = load <128 x i32>* %a
\r
26 %2 = load <128 x i32>* %b
\r
27 %3 = and <128 x i32> %1, %2
\r
28 %4 = xor <128 x i32> %1, %2
\r
29 %5 = sitofp <128 x i32> %3 to <128 x float>
\r
30 %6 = sitofp <128 x i32> %4 to <128 x float>
\r
31 %7 = fadd <128 x float> %5, %6
\r
32 store <128 x float> %7, <128 x float>* %c
\r