gh-mirror/mikepaul-LuaJIT
1
0
Fork 0
mirror of https://github.com/LuaJIT/LuaJIT.git synced 2025-02-08 15:34:09 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Use execute rather than loop for mvc and avoid jumps in fast path.

Browse Source 
Not sure if this works, the tests don't exercise the stack code.
This commit is contained in:
Michael Munday 2017-01-10 14:12:06 -05:00
parent 660ddd1db2
commit e933353feb
2 changed files with 31 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/Makefile
View File

@ -56,7 +56,7 @@ CCOPT_mips=
# #
CCDEBUG= CCDEBUG=
# Uncomment the next line to generate debug information: # Uncomment the next line to generate debug information:
#CCDEBUG= -g CCDEBUG= -g
# #
CCWARN= -Wall CCWARN= -Wall
# Uncomment the next line to enable more warnings: # Uncomment the next line to enable more warnings:

58
src/vm_s390x.dasc
View File

@ -2142,47 +2142,26 @@ static void build_subroutines(BuildCtx *ctx)
|->vm_ffi_call: // Call C function via FFI. |->vm_ffi_call: // Call C function via FFI.
| // Caveat: needs special frame unwinding, see below. | // Caveat: needs special frame unwinding, see below.
|.if FFI |.if FFI
| .type CCSTATE, CCallState, r10 | .type CCSTATE, CCallState, r8
| stmg r6, r15, 48(sp) // TODO: need to save r6, but might be better in separate store? | stmg r6, r15, 48(sp)
| lgr CCSTATE, CARG1 | lgr CCSTATE, CARG1
| lg r7, CCSTATE->func // TODO: move further up?
| |
| // Readjust stack. | // Readjust stack.
| sgf sp, CCSTATE->spadj | sgf sp, CCSTATE->spadj
| |
| // Copy stack slots. | // Copy stack slots.
| llgc r0, CCSTATE->nsp | llgc r1, CCSTATE->nsp
| cghi r0, 0 | chi r1, 0
| jle >3 | jh >2
| lay r1, (offsetof(CCallState, stack))(CCSTATE) // Source.
| lay r11, (CCALL_SPS_EXTRA*8)(sp) // Destination.
|1: |1:
| cghi r0, 256
| jl >2
| mvc 0(256, r11), 0(r1)
| aghi r1, 256*8
| aghi r11, 256*8
| aghi r0, -256
| j <1
|2:
| cghi r0, 0
| je >3
| // TODO: exrl mvc rather than loop.
| mvc 0(8, r11), 0(r1)
| aghi r1, 8
| aghi r11, 8
| aghi r0, -1
| j <2
|3:
|
| lmg CARG1, CARG5, CCSTATE->gpr[0] | lmg CARG1, CARG5, CCSTATE->gpr[0]
| // TODO: conditionally load FPRs? | // TODO: conditionally load FPRs?
| ld FARG1, CCSTATE->fpr[0] | ld FARG1, CCSTATE->fpr[0]
| ld FARG2, CCSTATE->fpr[1] | ld FARG2, CCSTATE->fpr[1]
| ld FARG3, CCSTATE->fpr[2] | ld FARG3, CCSTATE->fpr[2]
| ld FARG4, CCSTATE->fpr[3] | ld FARG4, CCSTATE->fpr[3]
|5: | basr r14, r7
| lg r1, CCSTATE->func // TODO: move further up?
| basr r14, r1
| |
| stg CRET1, CCSTATE->gpr[0] | stg CRET1, CCSTATE->gpr[0]
| stg f0, CCSTATE->fpr[0] | stg f0, CCSTATE->fpr[0]
@ -2190,6 +2169,29 @@ static void build_subroutines(BuildCtx *ctx)
| agf sp, CCSTATE->spadj | agf sp, CCSTATE->spadj
| lmg r6, r15, 48(sp) | lmg r6, r15, 48(sp)
| br r14 | br r14
|
|2:
| lay r10, (offsetof(CCallState, stack))(CCSTATE) // Source.
| lay r11, (CCALL_SPS_EXTRA*8)(sp) // Destination.
|3:
| chi r1, 256
| jl >4
| mvc 0(256, r11), 0(r10)
| la r10, 256*8(r10)
| la r11, 256*8(r11)
| ahi r1, -256
| j <3
|
|4:
| ahi r1, -1
| jl <1
| larl r9, >5
| ex r1, 0(r9) // TODO: exrl is faster but needs z10.
| j <1
|
|5:
| // exrl target
| mvc 0(1, r11), 0(r10)
|.endif |.endif
|// Note: vm_ffi_call must be the last function in this object file! |// Note: vm_ffi_call must be the last function in this object file!
| |