Mercurial > hg > CbC > CbC_llvm

comparison test/CodeGen/AMDGPU/private-memory.ll @ 100:7d135dc70f03 LLVM 3.9

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 3.9
author Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
date Tue, 26 Jan 2016 22:53:40 +0900
parents afa8332a0e37
children
comparison
equal deleted inserted replaced
96:6418606d0ead 100:7d135dc70f03
1 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC 1 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
2 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC 2 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
3 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE
3 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC 4 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
5 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA
4 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC 6 ; RUN: llc -show-mc-encoding -mattr=+promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC
5 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC 7 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
6 8
7 declare i32 @llvm.r600.read.tidig.x() nounwind readnone 9 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
8 10
11 ; R600: LDS_WRITE 13 ; R600: LDS_WRITE
12 ; R600: LDS_WRITE 14 ; R600: LDS_WRITE
13 ; R600: LDS_READ 15 ; R600: LDS_READ
14 ; R600: LDS_READ 16 ; R600: LDS_READ
15 17
18 ; HSA-PROMOTE: .amd_kernel_code_t
19 ; HSA-PROMOTE: workgroup_group_segment_byte_size = 5120
20 ; HSA-PROMOTE: .end_amd_kernel_code_t
21
16 ; SI-PROMOTE: ds_write_b32 22 ; SI-PROMOTE: ds_write_b32
17 ; SI-PROMOTE: ds_write_b32 23 ; SI-PROMOTE: ds_write_b32
18 ; SI-PROMOTE: ds_read_b32 24 ; SI-PROMOTE: ds_read_b32
19 ; SI-PROMOTE: ds_read_b32 25 ; SI-PROMOTE: ds_read_b32
26
27 ; HSA-ALLOCA: .amd_kernel_code_t
28 ; FIXME: Creating the emergency stack slots causes us to over-estimate scratch
29 ; by 4 bytes.
30 ; HSA-ALLOCA: workitem_private_segment_byte_size = 24
31 ; HSA-ALLOCA: .end_amd_kernel_code_t
20 32
21 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0 33 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
22 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0 34 ; SI-ALLOCA: buffer_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offen ; encoding: [0x00,0x10,0x70,0xe0
23 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) { 35 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
24 entry: 36 entry: