Brox Chen 364b97f23b
[AMDGPU][True16][CodeGen] 16bit spill support in true16 mode (#128060)
Enables 16-bit values to be spilled to scratch.

Note, the memory instructions used are defined as reading and writing
VGPR_32, but do not clobber the unspecified 16-bits of those registers,
and so spills and reloads of lo and hi halves of the registers work.
2025-02-26 16:17:20 -05:00

59 lines
3.1 KiB
YAML

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -verify-machineinstrs -mcpu=gfx1100 -mattr=+real-true16 -run-pass=regallocfast -o - %s | FileCheck -check-prefix=SPILLED %s
# RUN: llc -march=amdgcn -verify-machineinstrs -mcpu=gfx1100 -mattr=+real-true16 -run-pass=regallocfast,prologepilog -o - %s | FileCheck -check-prefix=EXPANDED %s
---
name: spill_restore_vgpr16
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
; SPILLED-LABEL: name: spill_restore_vgpr16
; SPILLED: bb.0:
; SPILLED-NEXT: successors: %bb.1(0x80000000)
; SPILLED-NEXT: {{ $}}
; SPILLED-NEXT: S_NOP 0, implicit-def renamable $vgpr0_lo16, implicit-def renamable $vgpr0_hi16
; SPILLED-NEXT: SI_SPILL_V16_SAVE killed $vgpr0_hi16, %stack.1, $sgpr32, 0, implicit $exec :: (store (s16) into %stack.1, addrspace 5)
; SPILLED-NEXT: SI_SPILL_V16_SAVE killed $vgpr0_lo16, %stack.0, $sgpr32, 0, implicit $exec :: (store (s16) into %stack.0, 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_lo16 = SI_SPILL_V16_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s16) from %stack.0, addrspace 5)
; SPILLED-NEXT: $vgpr0_hi16 = SI_SPILL_V16_RESTORE %stack.1, $sgpr32, 0, implicit $exec :: (load (s16) from %stack.1, addrspace 5)
; SPILLED-NEXT: S_NOP 0, implicit killed renamable $vgpr0_lo16, implicit killed renamable $vgpr0_hi16
;
; EXPANDED-LABEL: name: spill_restore_vgpr16
; EXPANDED: bb.0:
; EXPANDED-NEXT: successors: %bb.1(0x80000000)
; EXPANDED-NEXT: {{ $}}
; EXPANDED-NEXT: S_NOP 0, implicit-def renamable $vgpr0_lo16, implicit-def renamable $vgpr0_hi16
; EXPANDED-NEXT: SCRATCH_STORE_SHORT_SADDR_t16 killed $vgpr0_hi16, $sgpr32, 2, 0, implicit $exec, implicit $flat_scr :: (store (s16) into %stack.1, addrspace 5)
; EXPANDED-NEXT: SCRATCH_STORE_SHORT_SADDR_t16 killed $vgpr0_lo16, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s16) into %stack.0, 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_lo16 = SCRATCH_LOAD_SHORT_D16_SADDR_t16 $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s16) from %stack.0, addrspace 5)
; EXPANDED-NEXT: $vgpr0_hi16 = SCRATCH_LOAD_SHORT_D16_SADDR_t16 $sgpr32, 2, 0, implicit $exec, implicit $flat_scr :: (load (s16) from %stack.1, addrspace 5)
; EXPANDED-NEXT: S_NOP 0, implicit killed renamable $vgpr0_lo16, implicit killed renamable $vgpr0_hi16
bb.0:
S_NOP 0, implicit-def %0:vgpr_16, implicit-def %1:vgpr_16
S_CBRANCH_SCC1 implicit undef $scc, %bb.1
bb.1:
S_NOP 1
bb.2:
S_NOP 0, implicit %0, implicit %1
...