d0p1/ack
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity
ack/util/mcgg/ir.dat

124 lines
2 KiB
Plaintext
Raw Normal View History

Add the OPTIONS clause to the mcgg grammar; add an optional pass which converts sequences of PUSHes to a single STACKADJUST followed by STOREs. This should dramatically improve code on stack-unfriendly architectures like MIPS.
2018-09-22 09:19:00 +00:00
# The second column indicates the type inference rules:
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
#
# I, F, L, D: int, float, long, double
Add the OPTIONS clause to the mcgg grammar; add an optional pass which converts sequences of PUSHes to a single STACKADJUST followed by STOREs. This should dramatically improve code on stack-unfriendly architectures like MIPS.
2018-09-22 09:19:00 +00:00
# i, f: int, float; promoted to long, double based on size
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
# .: ignore this parameter
# ?: pull/push types from other ? parameters
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Simple terminals
Lots more opcodes. Rearrange the stack layout so that fp->ab is a fixed value (needed for CHAINFP and FPTOAB). Wire up lfrs to calls via a phi when necessary, to allow call-bra-lfr chains.
2016-10-29 09:57:56 +00:00
S ?=.. CONST
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
V ?=.. REG
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
S ?=?. NOP
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S I=.. LABEL
S I=.. BLOCK
V ?=.. PAIR
S ?=.. ANY
S ?=.. LOCAL
V ?=.. PHI
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Magic stack operations
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S ?=?. PUSH
S ?=.. POP
Keep more data around about ir instructions. Implement a half-baked type inference routine to propagate information about floats up the tree, so we know whether to put floats into special registers as early as possible.
2016-09-26 20:12:46 +00:00
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
# Memory operations
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
S ?=I. LOAD
S I=I. LOADB
S I=I. LOADH
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S ?=I? STORE
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
S ?=I? STOREB
S ?=I? STOREH
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Arithemetic operations
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S i=ii ADD
S i=ii SUB
S i=ii MUL
S i=ii DIV
S i=ii DIVU
S i=ii MOD
S i=ii MODU
S i=ii NEG
S f=ff ADDF
S f=ff SUBF
S f=ff MULF
S f=ff DIVF
S f=ff NEGF
S i=ii AND
S i=ii OR
S i=ii EOR
S i=ii NOT
S i=ii ASL
S i=ii ASR
S i=ii LSL
S i=ii LSR
# Bitwise conversions
# (Remember, these don't change the value, merely move it)
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
# (order is important here; the 8-byte version of each must immediate succeed
# the 4-byte version)
S F=I. COPYI
S I=F. COPYF
S D=L. COPYL
S L=D. COPYD
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
# Semantic conversions
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
# (order is important here; the 8-byte version of each must immediate succeed
# the 4-byte version)
S ?=I. FROMUI
S ?=L. FROMUL
S ?=I. FROMSI
S ?=L. FROMSL
Better handling of float conversions; more tests; converting to unsigned ints works now.
2016-11-20 10:27:40 +00:00
S ?=F. FROMUF
S ?=D. FROMUD
S ?=F. FROMSF
S ?=D. FROMSD
More opcodes. sti can now cope with non-standard sizes (really need a better fix for this). Hack in crude support for mismatched stack pushes and pops (ints vs longs).
2016-10-29 10:48:05 +00:00
S L=II FROMIPAIR
S I=L. FROML0
S I=L. FROML1
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
# The H versions are only used if wordsize > 2
S I=I. EXTENDB
S I=I. EXTENDH
S I=I. TRUNCATEB
S I=I. TRUNCATEH
Better (and more correct) floating point conversions; fif; various new opcodes.
2016-10-21 22:48:26 +00:00
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Tristate comparisons
Massive change to how IR types are handled; we use the type code for matching rather than the size. Much cleaner and simpler.
2016-10-23 19:54:14 +00:00
# (order is important here; the 8-byte version of each must immediate succeed
# the 4-byte version)
S I=II COMPARESI
S I=LL COMPARESL
S I=II COMPAREUI
S I=LL COMPAREUL
S I=FF COMPAREF
S I=DD COMPARED
Keep more data around about ir instructions. Implement a half-baked type inference routine to propagate information about floats up the tree, so we know whether to put floats into special registers as early as possible.
2016-09-26 20:12:46 +00:00
More opcodes.
2016-10-22 18:32:51 +00:00
# Tristate to boolean conversion
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S I=I. IFEQ
S I=I. IFLT
S I=I. IFLE
Keep more data around about ir instructions. Implement a half-baked type inference routine to propagate information about floats up the tree, so we know whether to put floats into special registers as early as possible.
2016-09-26 20:12:46 +00:00
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Procedures
Lots more opcodes. Rearrange the stack layout so that fp->ab is a fixed value (needed for CHAINFP and FPTOAB). Wire up lfrs to calls via a phi when necessary, to allow call-bra-lfr chains.
2016-10-29 09:57:56 +00:00
S i=I. CALL
S i=?. GETRET
S ?=i. SETRET
Keep more data around about ir instructions. Implement a half-baked type inference routine to propagate information about floats up the tree, so we know whether to put floats into special registers as early as possible.
2016-09-26 20:12:46 +00:00
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Flow control --- these never return
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
V .=i. JUMP
Made csa and csb work with mcg; adjust the libem functions and the corresponding invocation in the ncg table so the same helpers can be used for both mcg and ncg. Add a new IR opcode, FARJUMP, which jumps to a helper function but saves volatile registers.
2016-11-19 09:55:41 +00:00
V .=i. FARJUMP
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
V .=i. CJUMPEQ
V .=i. CJUMPLT
V .=i. CJUMPLE
V .=.. RET
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
# Special
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S ?=i. STACKADJUST
S i=.. GETFP
Lots more opcodes. Rearrange the stack layout so that fp->ab is a fixed value (needed for CHAINFP and FPTOAB). Wire up lfrs to calls via a phi when necessary, to allow call-bra-lfr chains.
2016-10-29 09:57:56 +00:00
S ?=i. SETFP
Re-re-add the type inference layer, now I know more about how things work. Remove that terrible float promotion code.
2016-10-22 21:04:13 +00:00
S i=.. GETSP
S ?=i. SETSP
S i=i. CHAINFP
Lots more opcodes. Rearrange the stack layout so that fp->ab is a fixed value (needed for CHAINFP and FPTOAB). Wire up lfrs to calls via a phi when necessary, to allow call-bra-lfr chains.
2016-10-29 09:57:56 +00:00
S i=i. FPTOAB
S i=i. FPTOLB
IR codes are now owned by mcgg; ir terminals are inserted into the table during compilation (so you can refer to them).
2016-09-24 16:31:35 +00:00
Reference in a new issue Copy permalink