| Simon Pilgrim | 3a77180 | 2015-06-07 18:34:25 +0000 | [diff] [blame^] | 1 | ; RUN: llc -O3 -disable-peephole -mtriple=x86_64-unknown-unknown -mcpu=x86-64 < %s | FileCheck %s |
| 2 | |
| 3 | target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" |
| 4 | target triple = "x86_64-unknown-unknown" |
| 5 | |
| 6 | ; Stack reload folding tests. |
| 7 | ; |
| 8 | ; By including a nop call with sideeffects we can force a partial register spill of the |
| 9 | ; relevant registers and check that the reload is correctly folded into the instruction. |
| 10 | |
| 11 | ;TODO stack_fold_bsf_i16 |
| 12 | declare i16 @llvm.cttz.i16(i16, i1) |
| 13 | |
| 14 | define i32 @stack_fold_bsf_i32(i32 %a0) { |
| 15 | ;CHECK-LABEL: stack_fold_bsf_i32 |
| 16 | ;CHECK: bsfl {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 4-byte Folded Reload |
| 17 | %1 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"() |
| 18 | %2 = call i32 @llvm.cttz.i32(i32 %a0, i1 -1) |
| 19 | ret i32 %2 |
| 20 | } |
| 21 | declare i32 @llvm.cttz.i32(i32, i1) |
| 22 | |
| 23 | define i64 @stack_fold_bsf_i64(i64 %a0) { |
| 24 | ;CHECK-LABEL: stack_fold_bsf_i64 |
| 25 | ;CHECK: bsfq {{-?[0-9]*}}(%rsp), %rax {{.*#+}} 8-byte Folded Reload |
| 26 | %1 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"() |
| 27 | %2 = call i64 @llvm.cttz.i64(i64 %a0, i1 -1) |
| 28 | ret i64 %2 |
| 29 | } |
| 30 | declare i64 @llvm.cttz.i64(i64, i1) |
| 31 | |
| 32 | ;TODO stack_fold_bsr_i16 |
| 33 | declare i16 @llvm.ctlz.i16(i16, i1) |
| 34 | |
| 35 | define i32 @stack_fold_bsr_i32(i32 %a0) { |
| 36 | ;CHECK-LABEL: stack_fold_bsr_i32 |
| 37 | ;CHECK: bsrl {{-?[0-9]*}}(%rsp), %eax {{.*#+}} 4-byte Folded Reload |
| 38 | %1 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"() |
| 39 | %2 = call i32 @llvm.ctlz.i32(i32 %a0, i1 -1) |
| 40 | ret i32 %2 |
| 41 | } |
| 42 | declare i32 @llvm.ctlz.i32(i32, i1) |
| 43 | |
| 44 | define i64 @stack_fold_bsr_i64(i64 %a0) { |
| 45 | ;CHECK-LABEL: stack_fold_bsr_i64 |
| 46 | ;CHECK: bsrq {{-?[0-9]*}}(%rsp), %rax {{.*#+}} 8-byte Folded Reload |
| 47 | %1 = tail call i64 asm sideeffect "nop", "=x,~{rax},~{rbx},~{rcx},~{rdx},~{rsi},~{rdi},~{rbp},~{r8},~{r9},~{r10},~{r11},~{r12},~{r13},~{r14},~{r15}"() |
| 48 | %2 = call i64 @llvm.ctlz.i64(i64 %a0, i1 -1) |
| 49 | ret i64 %2 |
| 50 | } |
| 51 | declare i64 @llvm.ctlz.i64(i64, i1) |