bbbe8ecc17
We previously had a heuristic that if a value V was used multiple times
in a single PHI, then to avoid potentially rematerializing into many predecessors
we bail out. The phi uses only counted as a single use in the shouldLocalize() hook
because it counted the PHI as a single instruction use, not factoring in it may
have many incoming edges.
It turns out this heuristic is slightly too pessimistic, and allowing a small number
of these uses to be localized can improve code size due to shortening live ranges,
especially if those ranges span a call.
This change results in some improvements in size on CTMark -Os:
```
Program size.__text
before after diff
kimwitu++/kc 451676.00 451860.00 0.0%
mafft/pairlocalalign 241460.00 241540.00 0.0%
tramp3d-v4/tramp3d-v4 389216.00 389208.00 -0.0%
7zip/7zip-benchmark 587528.00 587464.00 -0.0%
Bullet/bullet 457424.00 457348.00 -0.0%
consumer-typeset/consumer-typeset 405472.00 405376.00 -0.0%
SPASS/SPASS 410288.00 410120.00 -0.0%
lencod/lencod 426396.00 426108.00 -0.1%
ClamAV/clamscan 380108.00 379756.00 -0.1%
sqlite3/sqlite3 283664.00 283372.00 -0.1%
Geomean difference -0.0%
```
I experimented with different variations and thresholds. Using 3 instead
of 2 resulted in a further 0.1% improvement on ClamAV but also regressed
sqlite3 by the same %.
85 lines
3.8 KiB
LLVM
85 lines
3.8 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -mem-intrinsic-expand-size=3 %s -o - | FileCheck -check-prefix=LOOP %s
|
|
; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -mem-intrinsic-expand-size=5 %s -o - | FileCheck -check-prefix=UNROLL %s
|
|
|
|
declare void @llvm.memmove.p1.p1.i32(ptr addrspace(1), ptr addrspace(1), i32, i1)
|
|
|
|
define amdgpu_cs void @memmove_p1i8(ptr addrspace(1) %dst, ptr addrspace(1) %src) {
|
|
; LOOP-LABEL: memmove_p1i8:
|
|
; LOOP: ; %bb.0:
|
|
; LOOP-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[0:1]
|
|
; LOOP-NEXT: s_and_saveexec_b64 s[0:1], vcc
|
|
; LOOP-NEXT: s_xor_b64 s[4:5], exec, s[0:1]
|
|
; LOOP-NEXT: s_cbranch_execz .LBB0_3
|
|
; LOOP-NEXT: ; %bb.1: ; %copy_forward
|
|
; LOOP-NEXT: s_mov_b64 s[6:7], 0
|
|
; LOOP-NEXT: s_mov_b32 s2, 0
|
|
; LOOP-NEXT: s_mov_b32 s3, 0xf000
|
|
; LOOP-NEXT: s_mov_b64 s[0:1], 0
|
|
; LOOP-NEXT: v_mov_b32_e32 v4, s6
|
|
; LOOP-NEXT: v_mov_b32_e32 v5, s7
|
|
; LOOP-NEXT: .LBB0_2: ; %copy_forward_loop
|
|
; LOOP-NEXT: ; =>This Inner Loop Header: Depth=1
|
|
; LOOP-NEXT: v_add_i32_e32 v6, vcc, v2, v4
|
|
; LOOP-NEXT: v_addc_u32_e32 v7, vcc, v3, v5, vcc
|
|
; LOOP-NEXT: s_waitcnt expcnt(0)
|
|
; LOOP-NEXT: buffer_load_ubyte v8, v[6:7], s[0:3], 0 addr64
|
|
; LOOP-NEXT: v_add_i32_e32 v6, vcc, v0, v4
|
|
; LOOP-NEXT: v_addc_u32_e32 v7, vcc, v1, v5, vcc
|
|
; LOOP-NEXT: v_add_i32_e32 v4, vcc, 1, v4
|
|
; LOOP-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
|
|
; LOOP-NEXT: v_cmp_ne_u32_e32 vcc, 4, v4
|
|
; LOOP-NEXT: s_waitcnt vmcnt(0)
|
|
; LOOP-NEXT: buffer_store_byte v8, v[6:7], s[0:3], 0 addr64
|
|
; LOOP-NEXT: s_cbranch_vccnz .LBB0_2
|
|
; LOOP-NEXT: .LBB0_3: ; %Flow17
|
|
; LOOP-NEXT: s_andn2_saveexec_b64 s[0:1], s[4:5]
|
|
; LOOP-NEXT: s_cbranch_execz .LBB0_6
|
|
; LOOP-NEXT: ; %bb.4: ; %copy_backwards
|
|
; LOOP-NEXT: v_add_i32_e32 v0, vcc, 3, v0
|
|
; LOOP-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
|
|
; LOOP-NEXT: v_add_i32_e32 v2, vcc, 3, v2
|
|
; LOOP-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
|
|
; LOOP-NEXT: s_mov_b32 s0, -4
|
|
; LOOP-NEXT: s_mov_b32 s6, 0
|
|
; LOOP-NEXT: s_mov_b32 s7, 0xf000
|
|
; LOOP-NEXT: s_mov_b64 s[4:5], 0
|
|
; LOOP-NEXT: v_mov_b32_e32 v4, s0
|
|
; LOOP-NEXT: .LBB0_5: ; %copy_backwards_loop
|
|
; LOOP-NEXT: ; =>This Inner Loop Header: Depth=1
|
|
; LOOP-NEXT: s_waitcnt expcnt(0)
|
|
; LOOP-NEXT: buffer_load_ubyte v5, v[2:3], s[4:7], 0 addr64
|
|
; LOOP-NEXT: v_add_i32_e32 v4, vcc, 1, v4
|
|
; LOOP-NEXT: s_and_b64 vcc, vcc, exec
|
|
; LOOP-NEXT: s_waitcnt vmcnt(0)
|
|
; LOOP-NEXT: buffer_store_byte v5, v[0:1], s[4:7], 0 addr64
|
|
; LOOP-NEXT: v_add_i32_e64 v0, s[0:1], -1, v0
|
|
; LOOP-NEXT: v_addc_u32_e64 v1, s[0:1], -1, v1, s[0:1]
|
|
; LOOP-NEXT: v_add_i32_e64 v2, s[0:1], -1, v2
|
|
; LOOP-NEXT: v_addc_u32_e64 v3, s[0:1], -1, v3, s[0:1]
|
|
; LOOP-NEXT: s_cbranch_vccz .LBB0_5
|
|
; LOOP-NEXT: .LBB0_6: ; %memmove_done
|
|
; LOOP-NEXT: s_endpgm
|
|
;
|
|
; UNROLL-LABEL: memmove_p1i8:
|
|
; UNROLL: ; %bb.0:
|
|
; UNROLL-NEXT: s_mov_b32 s2, 0
|
|
; UNROLL-NEXT: s_mov_b32 s3, 0xf000
|
|
; UNROLL-NEXT: s_mov_b64 s[0:1], 0
|
|
; UNROLL-NEXT: buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64
|
|
; UNROLL-NEXT: buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:1
|
|
; UNROLL-NEXT: buffer_load_ubyte v6, v[2:3], s[0:3], 0 addr64 offset:2
|
|
; UNROLL-NEXT: buffer_load_ubyte v2, v[2:3], s[0:3], 0 addr64 offset:3
|
|
; UNROLL-NEXT: s_waitcnt vmcnt(3)
|
|
; UNROLL-NEXT: buffer_store_byte v4, v[0:1], s[0:3], 0 addr64
|
|
; UNROLL-NEXT: s_waitcnt vmcnt(3)
|
|
; UNROLL-NEXT: buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:1
|
|
; UNROLL-NEXT: s_waitcnt vmcnt(3)
|
|
; UNROLL-NEXT: buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:2
|
|
; UNROLL-NEXT: s_waitcnt vmcnt(3)
|
|
; UNROLL-NEXT: buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:3
|
|
; UNROLL-NEXT: s_endpgm
|
|
call void @llvm.memmove.p1.p1.i32(ptr addrspace(1) %dst, ptr addrspace(1) %src, i32 4, i1 false)
|
|
ret void
|
|
}
|