caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
7d2b9fafabd6f1a67e05ab2fa0abb5f7c8a451e7
clang-p2996/llvm/test/CodeGen/AMDGPU/spill320.mir
Jeffrey Byrnes 1f08d3bc3a [AMDGPU] Further reduce attaching of implicit operands to spills 
Extension of https://reviews.llvm.org/D141101 to even further reduce the amount of implicit operands we attach. The main benefit is to improve cability of post-ra scheduler, and reduce unneeded dependency resolution (e.g. inserting snops).

Unfortunately, we run into regressions if we completely minimize the amount implicit operands (naively), we run into some regressions (e.g. dual_movs are replaced with multiple calls to v_mov). This is even more reason to switch to LiveRegUnits.

Nonetheless, this patch removes the operands which we can for free (more or less).

Change-Id: Ib4f409202b36bdbc59eed615bc2d19fa8bd8c057

Differential Revision: https://reviews.llvm.org/D141557

Change-Id: I8b039e3c0d39436b384083f8beb947ee1b1730b2
2023-01-19 14:31:07 -08:00

128 lines
6.3 KiB
YAML
Raw Blame History

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regallocfast,si-lower-sgpr-spills -o - %s | FileCheck -check-prefix=EXPANDED %s
# Make sure spill/restore of 320 bit registers works.
---
name: spill_restore_sgpr320
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
; SPILLED-LABEL: name: spill_restore_sgpr320
; SPILLED: bb.0:
; SPILLED-NEXT: successors: %bb.1(0x80000000)
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: S_NOP 0, implicit-def renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; SPILLED-NEXT: SI_SPILL_S320_SAVE killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13, %stack.0, implicit $exec, implicit $sgpr32 :: (store (s320) into %stack.0, align 4, addrspace 5)
; SPILLED-NEXT: S_CBRANCH_SCC1 %bb.1, implicit undef $scc
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: bb.1:
; SPILLED-NEXT: successors: %bb.2(0x80000000)
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: S_NOP 1
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: bb.2:
; SPILLED-NEXT: $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13 = SI_SPILL_S320_RESTORE %stack.0, implicit $exec, implicit $sgpr32 :: (load (s320) from %stack.0, align 4, addrspace 5)
; SPILLED-NEXT: S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; EXPANDED-LABEL: name: spill_restore_sgpr320
; EXPANDED: bb.0:
; EXPANDED-NEXT: successors: %bb.1(0x80000000)
; EXPANDED-NEXT: liveins: $vgpr0
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: S_NOP 0, implicit-def renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr4, 0, $vgpr0, implicit-def $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13, implicit $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr5, 1, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr6, 2, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr7, 3, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr8, 4, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr9, 5, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr10, 6, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr11, 7, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 $sgpr12, 8, $vgpr0
; EXPANDED-NEXT: $vgpr0 = V_WRITELANE_B32 killed $sgpr13, 9, $vgpr0, implicit killed $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; EXPANDED-NEXT: S_CBRANCH_SCC1 %bb.1, implicit undef $scc
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: bb.1:
; EXPANDED-NEXT: successors: %bb.2(0x80000000)
; EXPANDED-NEXT: liveins: $vgpr0
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: S_NOP 1
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: bb.2:
; EXPANDED-NEXT: liveins: $vgpr0
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: $sgpr4 = V_READLANE_B32 $vgpr0, 0, implicit-def $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
; EXPANDED-NEXT: $sgpr5 = V_READLANE_B32 $vgpr0, 1
; EXPANDED-NEXT: $sgpr6 = V_READLANE_B32 $vgpr0, 2
; EXPANDED-NEXT: $sgpr7 = V_READLANE_B32 $vgpr0, 3
; EXPANDED-NEXT: $sgpr8 = V_READLANE_B32 $vgpr0, 4
; EXPANDED-NEXT: $sgpr9 = V_READLANE_B32 $vgpr0, 5
; EXPANDED-NEXT: $sgpr10 = V_READLANE_B32 $vgpr0, 6
; EXPANDED-NEXT: $sgpr11 = V_READLANE_B32 $vgpr0, 7
; EXPANDED-NEXT: $sgpr12 = V_READLANE_B32 $vgpr0, 8
; EXPANDED-NEXT: $sgpr13 = V_READLANE_B32 $vgpr0, 9
; EXPANDED-NEXT: S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13
bb.0:
S_NOP 0, implicit-def %0:sgpr_320
S_CBRANCH_SCC1 implicit undef $scc, %bb.1
bb.1:
S_NOP 1
bb.2:
S_NOP 0, implicit %0
...
---
name: spill_restore_vgpr320
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
; SPILLED-LABEL: name: spill_restore_vgpr320
; SPILLED: bb.0:
; SPILLED-NEXT: successors: %bb.1(0x80000000)
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: S_NOP 0, implicit-def renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9
; SPILLED-NEXT: SI_SPILL_V320_SAVE killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9, %stack.0, $sgpr32, 0, implicit $exec :: (store (s320) into %stack.0, align 4, addrspace 5)
; SPILLED-NEXT: S_CBRANCH_SCC1 %bb.1, implicit undef $scc
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: bb.1:
; SPILLED-NEXT: successors: %bb.2(0x80000000)
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: S_NOP 1
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: bb.2:
; SPILLED-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = SI_SPILL_V320_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s320) from %stack.0, align 4, addrspace 5)
; SPILLED-NEXT: S_NOP 0, implicit killed renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9
; EXPANDED-LABEL: name: spill_restore_vgpr320
; EXPANDED: bb.0:
; EXPANDED-NEXT: successors: %bb.1(0x80000000)
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: S_NOP 0, implicit-def renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9
; EXPANDED-NEXT: SI_SPILL_V320_SAVE killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9, %stack.0, $sgpr32, 0, implicit $exec :: (store (s320) into %stack.0, align 4, addrspace 5)
; EXPANDED-NEXT: S_CBRANCH_SCC1 %bb.1, implicit undef $scc
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: bb.1:
; EXPANDED-NEXT: successors: %bb.2(0x80000000)
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: S_NOP 1
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: bb.2:
; EXPANDED-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = SI_SPILL_V320_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s320) from %stack.0, align 4, addrspace 5)
; EXPANDED-NEXT: S_NOP 0, implicit killed renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9
bb.0:
S_NOP 0, implicit-def %0:vreg_320
S_CBRANCH_SCC1 implicit undef $scc, %bb.1
bb.1:
S_NOP 1
bb.2:
S_NOP 0, implicit %0
...
Reference in New Issue View Git Blame Copy Permalink