[llvm] 5553f27 - [AMDGPU][test] Fix the wrong triples in lower-work-group-id-intrinsics-{hsa,pal}.ll. NFC (#93501)
via llvm-commits
llvm-commits at lists.llvm.org
Wed May 29 02:42:44 PDT 2024
Author: Lu Weining
Date: 2024-05-29T17:42:41+08:00
New Revision: 5553f27d5a45e702415fa2f91d842bf4a1f4a8b5
URL: https://github.com/llvm/llvm-project/commit/5553f27d5a45e702415fa2f91d842bf4a1f4a8b5
DIFF: https://github.com/llvm/llvm-project/commit/5553f27d5a45e702415fa2f91d842bf4a1f4a8b5.diff
LOG: [AMDGPU][test] Fix the wrong triples in lower-work-group-id-intrinsics-{hsa,pal}.ll. NFC (#93501)
- hsa -> amdhsa
- Use amdgcn-amd-amd{hsa,pal} for lower-work-group-id-intrinsics-{hsa,pal}.ll respectively
Added:
Modified:
llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-hsa.ll
llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-pal.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-hsa.ll b/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-hsa.ll
index 9547f08d3eba6..1429251fc6421 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-hsa.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-hsa.ll
@@ -1,17 +1,17 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH,GFX9ARCH-SDAG %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH,GFX9ARCH-GISEL %s
-; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
-; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH,GFX9ARCH-SDAG %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH,GFX9ARCH-GISEL %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
define amdgpu_kernel void @workgroup_ids_kernel() {
; GFX9-LABEL: workgroup_ids_kernel:
; GFX9: ; %bb.0: ; %.entry
-; GFX9-NEXT: v_mov_b32_e32 v0, s0
-; GFX9-NEXT: v_mov_b32_e32 v1, s1
-; GFX9-NEXT: v_mov_b32_e32 v2, s2
+; GFX9-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-NEXT: v_mov_b32_e32 v2, s6
; GFX9-NEXT: buffer_store_dwordx3 v[0:2], off, s[0:3], 0
; GFX9-NEXT: s_endpgm
;
@@ -72,27 +72,20 @@ define amdgpu_kernel void @workgroup_ids_kernel() {
define amdgpu_kernel void @caller() {
; GFX9-SDAG-LABEL: caller:
; GFX9-SDAG: ; %bb.0:
-; GFX9-SDAG-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
-; GFX9-SDAG-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
-; GFX9-SDAG-NEXT: s_mov_b32 s38, -1
-; GFX9-SDAG-NEXT: s_mov_b32 s39, 0xe00000
-; GFX9-SDAG-NEXT: s_add_u32 s36, s36, s7
-; GFX9-SDAG-NEXT: s_addc_u32 s37, s37, 0
-; GFX9-SDAG-NEXT: s_add_u32 s8, s2, 36
-; GFX9-SDAG-NEXT: s_addc_u32 s9, s3, 0
-; GFX9-SDAG-NEXT: s_getpc_b64 s[2:3]
-; GFX9-SDAG-NEXT: s_add_u32 s2, s2, callee at gotpcrel32@lo+4
-; GFX9-SDAG-NEXT: s_addc_u32 s3, s3, callee at gotpcrel32@hi+12
-; GFX9-SDAG-NEXT: s_load_dwordx2 s[14:15], s[2:3], 0x0
-; GFX9-SDAG-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-SDAG-NEXT: s_add_u32 flat_scratch_lo, s10, s13
+; GFX9-SDAG-NEXT: s_addc_u32 flat_scratch_hi, s11, 0
+; GFX9-SDAG-NEXT: s_add_u32 s0, s0, s13
+; GFX9-SDAG-NEXT: s_addc_u32 s1, s1, 0
+; GFX9-SDAG-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-SDAG-NEXT: s_getpc_b64 s[8:9]
+; GFX9-SDAG-NEXT: s_add_u32 s8, s8, callee at gotpcrel32@lo+4
+; GFX9-SDAG-NEXT: s_addc_u32 s9, s9, callee at gotpcrel32@hi+12
+; GFX9-SDAG-NEXT: s_load_dwordx2 s[14:15], s[8:9], 0x0
; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v2, 20, v2
; GFX9-SDAG-NEXT: v_lshlrev_b32_e32 v1, 10, v1
-; GFX9-SDAG-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX9-SDAG-NEXT: s_mov_b64 s[0:1], s[36:37]
; GFX9-SDAG-NEXT: v_or3_b32 v31, v0, v1, v2
-; GFX9-SDAG-NEXT: s_mov_b32 s12, s6
-; GFX9-SDAG-NEXT: s_mov_b64 s[2:3], s[38:39]
-; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s6
+; GFX9-SDAG-NEXT: s_mov_b64 s[8:9], s[6:7]
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s12
; GFX9-SDAG-NEXT: s_mov_b32 s32, 0
; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-SDAG-NEXT: s_swappc_b64 s[30:31], s[14:15]
@@ -100,27 +93,20 @@ define amdgpu_kernel void @caller() {
;
; GFX9-GISEL-LABEL: caller:
; GFX9-GISEL: ; %bb.0:
-; GFX9-GISEL-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
-; GFX9-GISEL-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
-; GFX9-GISEL-NEXT: s_mov_b32 s38, -1
-; GFX9-GISEL-NEXT: s_mov_b32 s39, 0xe00000
-; GFX9-GISEL-NEXT: s_add_u32 s36, s36, s7
-; GFX9-GISEL-NEXT: s_addc_u32 s37, s37, 0
-; GFX9-GISEL-NEXT: s_add_u32 s8, s2, 36
-; GFX9-GISEL-NEXT: s_addc_u32 s9, s3, 0
-; GFX9-GISEL-NEXT: s_mov_b64 s[10:11], s[4:5]
-; GFX9-GISEL-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX9-GISEL-NEXT: s_getpc_b64 s[0:1]
-; GFX9-GISEL-NEXT: s_add_u32 s0, s0, callee at gotpcrel32@lo+4
-; GFX9-GISEL-NEXT: s_addc_u32 s1, s1, callee at gotpcrel32@hi+12
-; GFX9-GISEL-NEXT: s_load_dwordx2 s[14:15], s[0:1], 0x0
+; GFX9-GISEL-NEXT: s_add_u32 flat_scratch_lo, s10, s13
+; GFX9-GISEL-NEXT: s_addc_u32 flat_scratch_hi, s11, 0
+; GFX9-GISEL-NEXT: s_add_u32 s0, s0, s13
+; GFX9-GISEL-NEXT: s_addc_u32 s1, s1, 0
+; GFX9-GISEL-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-GISEL-NEXT: s_getpc_b64 s[8:9]
+; GFX9-GISEL-NEXT: s_add_u32 s8, s8, callee at gotpcrel32@lo+4
+; GFX9-GISEL-NEXT: s_addc_u32 s9, s9, callee at gotpcrel32@hi+12
+; GFX9-GISEL-NEXT: s_load_dwordx2 s[14:15], s[8:9], 0x0
; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v1, 10, v1
; GFX9-GISEL-NEXT: v_lshlrev_b32_e32 v2, 20, v2
-; GFX9-GISEL-NEXT: s_mov_b64 s[0:1], s[36:37]
; GFX9-GISEL-NEXT: v_or3_b32 v31, v0, v1, v2
-; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s6
-; GFX9-GISEL-NEXT: s_mov_b64 s[2:3], s[38:39]
-; GFX9-GISEL-NEXT: s_mov_b32 s12, s6
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s12
+; GFX9-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7]
; GFX9-GISEL-NEXT: s_mov_b32 s32, 0
; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-GISEL-NEXT: s_swappc_b64 s[30:31], s[14:15]
@@ -128,81 +114,61 @@ define amdgpu_kernel void @caller() {
;
; GFX9ARCH-SDAG-LABEL: caller:
; GFX9ARCH-SDAG: ; %bb.0:
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s38, -1
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s39, 0xe00000
-; GFX9ARCH-SDAG-NEXT: s_add_u32 s36, s36, s6
-; GFX9ARCH-SDAG-NEXT: s_addc_u32 s37, s37, 0
-; GFX9ARCH-SDAG-NEXT: s_add_u32 s8, s2, 36
-; GFX9ARCH-SDAG-NEXT: s_addc_u32 s9, s3, 0
-; GFX9ARCH-SDAG-NEXT: s_getpc_b64 s[2:3]
-; GFX9ARCH-SDAG-NEXT: s_add_u32 s2, s2, callee at gotpcrel32@lo+4
-; GFX9ARCH-SDAG-NEXT: s_addc_u32 s3, s3, callee at gotpcrel32@hi+12
-; GFX9ARCH-SDAG-NEXT: s_load_dwordx2 s[6:7], s[2:3], 0x0
-; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9ARCH-SDAG-NEXT: s_add_u32 flat_scratch_lo, s10, s12
+; GFX9ARCH-SDAG-NEXT: s_addc_u32 flat_scratch_hi, s11, 0
+; GFX9ARCH-SDAG-NEXT: s_add_u32 s0, s0, s12
+; GFX9ARCH-SDAG-NEXT: s_addc_u32 s1, s1, 0
+; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9ARCH-SDAG-NEXT: s_getpc_b64 s[8:9]
+; GFX9ARCH-SDAG-NEXT: s_add_u32 s8, s8, callee at gotpcrel32@lo+4
+; GFX9ARCH-SDAG-NEXT: s_addc_u32 s9, s9, callee at gotpcrel32@hi+12
+; GFX9ARCH-SDAG-NEXT: s_load_dwordx2 s[12:13], s[8:9], 0x0
; GFX9ARCH-SDAG-NEXT: v_lshlrev_b32_e32 v2, 20, v2
; GFX9ARCH-SDAG-NEXT: v_lshlrev_b32_e32 v1, 10, v1
-; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[0:1], s[36:37]
; GFX9ARCH-SDAG-NEXT: v_or3_b32 v31, v0, v1, v2
-; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[8:9], s[6:7]
; GFX9ARCH-SDAG-NEXT: v_mov_b32_e32 v0, ttmp9
; GFX9ARCH-SDAG-NEXT: s_mov_b32 s32, 0
; GFX9ARCH-SDAG-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9ARCH-SDAG-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9ARCH-SDAG-NEXT: s_swappc_b64 s[30:31], s[12:13]
; GFX9ARCH-SDAG-NEXT: s_endpgm
;
; GFX9ARCH-GISEL-LABEL: caller:
; GFX9ARCH-GISEL: ; %bb.0:
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s36, SCRATCH_RSRC_DWORD0
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s37, SCRATCH_RSRC_DWORD1
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s38, -1
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s39, 0xe00000
-; GFX9ARCH-GISEL-NEXT: s_add_u32 s36, s36, s6
-; GFX9ARCH-GISEL-NEXT: s_addc_u32 s37, s37, 0
-; GFX9ARCH-GISEL-NEXT: s_add_u32 s8, s2, 36
-; GFX9ARCH-GISEL-NEXT: s_addc_u32 s9, s3, 0
-; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[10:11], s[4:5]
-; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX9ARCH-GISEL-NEXT: s_getpc_b64 s[0:1]
-; GFX9ARCH-GISEL-NEXT: s_add_u32 s0, s0, callee at gotpcrel32@lo+4
-; GFX9ARCH-GISEL-NEXT: s_addc_u32 s1, s1, callee at gotpcrel32@hi+12
-; GFX9ARCH-GISEL-NEXT: s_load_dwordx2 s[6:7], s[0:1], 0x0
+; GFX9ARCH-GISEL-NEXT: s_add_u32 flat_scratch_lo, s10, s12
+; GFX9ARCH-GISEL-NEXT: s_addc_u32 flat_scratch_hi, s11, 0
+; GFX9ARCH-GISEL-NEXT: s_add_u32 s0, s0, s12
+; GFX9ARCH-GISEL-NEXT: s_addc_u32 s1, s1, 0
+; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9ARCH-GISEL-NEXT: s_getpc_b64 s[8:9]
+; GFX9ARCH-GISEL-NEXT: s_add_u32 s8, s8, callee at gotpcrel32@lo+4
+; GFX9ARCH-GISEL-NEXT: s_addc_u32 s9, s9, callee at gotpcrel32@hi+12
+; GFX9ARCH-GISEL-NEXT: s_load_dwordx2 s[12:13], s[8:9], 0x0
; GFX9ARCH-GISEL-NEXT: v_lshlrev_b32_e32 v1, 10, v1
; GFX9ARCH-GISEL-NEXT: v_lshlrev_b32_e32 v2, 20, v2
-; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[0:1], s[36:37]
; GFX9ARCH-GISEL-NEXT: v_or3_b32 v31, v0, v1, v2
; GFX9ARCH-GISEL-NEXT: v_mov_b32_e32 v0, ttmp9
-; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[2:3], s[38:39]
+; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7]
; GFX9ARCH-GISEL-NEXT: s_mov_b32 s32, 0
; GFX9ARCH-GISEL-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9ARCH-GISEL-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX9ARCH-GISEL-NEXT: s_swappc_b64 s[30:31], s[12:13]
; GFX9ARCH-GISEL-NEXT: s_endpgm
;
-; GFX12-SDAG-LABEL: caller:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: v_dual_mov_b32 v31, v0 :: v_dual_mov_b32 v0, ttmp9
-; GFX12-SDAG-NEXT: s_mov_b64 s[10:11], s[4:5]
-; GFX12-SDAG-NEXT: s_mov_b32 s7, callee at abs32@hi
-; GFX12-SDAG-NEXT: s_mov_b32 s6, callee at abs32@lo
-; GFX12-SDAG-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX12-SDAG-NEXT: s_mov_b64 s[8:9], s[2:3]
-; GFX12-SDAG-NEXT: s_mov_b32 s32, 0
-; GFX12-SDAG-NEXT: s_swappc_b64 s[30:31], s[6:7]
-; GFX12-SDAG-NEXT: s_endpgm
-;
-; GFX12-GISEL-LABEL: caller:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: v_dual_mov_b32 v31, v0 :: v_dual_mov_b32 v0, ttmp9
-; GFX12-GISEL-NEXT: s_mov_b64 s[10:11], s[4:5]
-; GFX12-GISEL-NEXT: s_mov_b32 s6, callee at abs32@lo
-; GFX12-GISEL-NEXT: s_mov_b32 s7, callee at abs32@hi
-; GFX12-GISEL-NEXT: s_mov_b64 s[4:5], s[0:1]
-; GFX12-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3]
-; GFX12-GISEL-NEXT: s_mov_b32 s32, 0
-; GFX12-GISEL-NEXT: s_swappc_b64 s[30:31], s[6:7]
-; GFX12-GISEL-NEXT: s_endpgm
+; GFX12-LABEL: caller:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX12-NEXT: s_getpc_b64 s[4:5]
+; GFX12-NEXT: s_sext_i32_i16 s5, s5
+; GFX12-NEXT: s_add_co_u32 s4, s4, callee at gotpcrel32@lo+8
+; GFX12-NEXT: s_add_co_ci_u32 s5, s5, callee at gotpcrel32@hi+16
+; GFX12-NEXT: v_dual_mov_b32 v31, v0 :: v_dual_mov_b32 v0, ttmp9
+; GFX12-NEXT: s_load_b64 s[6:7], s[4:5], 0x0
+; GFX12-NEXT: s_mov_b64 s[4:5], s[0:1]
+; GFX12-NEXT: s_mov_b64 s[8:9], s[2:3]
+; GFX12-NEXT: s_mov_b32 s32, 0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: s_swappc_b64 s[30:31], s[6:7]
+; GFX12-NEXT: s_endpgm
%idx = call i32 @llvm.amdgcn.workgroup.id.x()
call void @callee(i32 %idx) #0
ret void
diff --git a/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-pal.ll b/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-pal.ll
index 14fe4e5f48c67..8009f917aef5a 100644
--- a/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-pal.ll
+++ b/llvm/test/CodeGen/AMDGPU/lower-work-group-id-intrinsics-pal.ll
@@ -1,8 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH-SDAG %s
-; RUN: llc -mtriple=amdgcn-amd-hsa -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH-GISEL %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=-architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH-SDAG %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -mattr=+architected-sgprs -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX9ARCH-GISEL %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -global-isel=0 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 -global-isel=1 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
@@ -67,62 +67,37 @@ define amdgpu_cs void @_amdgpu_cs_main() {
}
define amdgpu_cs void @caller() {
-; GFX9-LABEL: caller:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
-; GFX9-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
-; GFX9-NEXT: s_mov_b32 s10, -1
-; GFX9-NEXT: s_mov_b32 s11, 0xe00000
-; GFX9-NEXT: s_add_u32 s8, s8, s0
-; GFX9-NEXT: s_addc_u32 s9, s9, 0
-; GFX9-NEXT: s_getpc_b64 s[0:1]
-; GFX9-NEXT: s_add_u32 s0, s0, callee at gotpcrel32@lo+4
-; GFX9-NEXT: s_addc_u32 s1, s1, callee at gotpcrel32@hi+12
-; GFX9-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
-; GFX9-NEXT: s_mov_b64 s[0:1], s[8:9]
-; GFX9-NEXT: s_mov_b64 s[2:3], s[10:11]
-; GFX9-NEXT: s_mov_b32 s32, 0
-; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: s_swappc_b64 s[30:31], s[4:5]
-; GFX9-NEXT: s_endpgm
-;
; GFX9ARCH-SDAG-LABEL: caller:
; GFX9ARCH-SDAG: ; %bb.0:
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s10, -1
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9ARCH-SDAG-NEXT: s_getpc_b64 s[8:9]
+; GFX9ARCH-SDAG-NEXT: s_mov_b32 s8, s0
+; GFX9ARCH-SDAG-NEXT: s_load_dwordx4 s[8:11], s[8:9], 0x10
+; GFX9ARCH-SDAG-NEXT: s_mov_b32 s5, callee at abs32@hi
+; GFX9ARCH-SDAG-NEXT: s_mov_b32 s4, callee at abs32@lo
+; GFX9ARCH-SDAG-NEXT: v_mov_b32_e32 v0, ttmp9
+; GFX9ARCH-SDAG-NEXT: s_mov_b32 s32, 0
+; GFX9ARCH-SDAG-NEXT: s_waitcnt lgkmcnt(0)
; GFX9ARCH-SDAG-NEXT: s_add_u32 s8, s8, s0
; GFX9ARCH-SDAG-NEXT: s_addc_u32 s9, s9, 0
-; GFX9ARCH-SDAG-NEXT: s_getpc_b64 s[0:1]
-; GFX9ARCH-SDAG-NEXT: s_add_u32 s0, s0, callee at gotpcrel32@lo+4
-; GFX9ARCH-SDAG-NEXT: s_addc_u32 s1, s1, callee at gotpcrel32@hi+12
-; GFX9ARCH-SDAG-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[0:1], s[8:9]
; GFX9ARCH-SDAG-NEXT: s_mov_b64 s[2:3], s[10:11]
-; GFX9ARCH-SDAG-NEXT: v_mov_b32_e32 v0, ttmp9
-; GFX9ARCH-SDAG-NEXT: s_mov_b32 s32, 0
-; GFX9ARCH-SDAG-NEXT: s_waitcnt lgkmcnt(0)
; GFX9ARCH-SDAG-NEXT: s_swappc_b64 s[30:31], s[4:5]
; GFX9ARCH-SDAG-NEXT: s_endpgm
;
; GFX9ARCH-GISEL-LABEL: caller:
; GFX9ARCH-GISEL: ; %bb.0:
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s8, SCRATCH_RSRC_DWORD0
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s9, SCRATCH_RSRC_DWORD1
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s10, -1
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s11, 0xe00000
+; GFX9ARCH-GISEL-NEXT: s_getpc_b64 s[8:9]
+; GFX9ARCH-GISEL-NEXT: s_mov_b32 s8, s0
+; GFX9ARCH-GISEL-NEXT: s_load_dwordx4 s[8:11], s[8:9], 0x10
+; GFX9ARCH-GISEL-NEXT: s_mov_b32 s4, callee at abs32@lo
+; GFX9ARCH-GISEL-NEXT: s_mov_b32 s5, callee at abs32@hi
+; GFX9ARCH-GISEL-NEXT: v_mov_b32_e32 v0, ttmp9
+; GFX9ARCH-GISEL-NEXT: s_mov_b32 s32, 0
+; GFX9ARCH-GISEL-NEXT: s_waitcnt lgkmcnt(0)
; GFX9ARCH-GISEL-NEXT: s_add_u32 s8, s8, s0
; GFX9ARCH-GISEL-NEXT: s_addc_u32 s9, s9, 0
-; GFX9ARCH-GISEL-NEXT: s_getpc_b64 s[0:1]
-; GFX9ARCH-GISEL-NEXT: s_add_u32 s0, s0, callee at gotpcrel32@lo+4
-; GFX9ARCH-GISEL-NEXT: s_addc_u32 s1, s1, callee at gotpcrel32@hi+12
-; GFX9ARCH-GISEL-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x0
; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[0:1], s[8:9]
-; GFX9ARCH-GISEL-NEXT: v_mov_b32_e32 v0, ttmp9
; GFX9ARCH-GISEL-NEXT: s_mov_b64 s[2:3], s[10:11]
-; GFX9ARCH-GISEL-NEXT: s_mov_b32 s32, 0
-; GFX9ARCH-GISEL-NEXT: s_waitcnt lgkmcnt(0)
; GFX9ARCH-GISEL-NEXT: s_swappc_b64 s[30:31], s[4:5]
; GFX9ARCH-GISEL-NEXT: s_endpgm
;
More information about the llvm-commits
mailing list