e36a0d7ef4
compiler: clean up code generation for alien boxing/unboxing a bit
2009-09-03 21:22:43 -05:00
c480bec303
convert comparison branch code in compiler to use locals
2009-09-03 21:19:39 -05:00
4d5a4222b6
More SIMD work
...
- Rename SIMD types and register representations: <type>-<count> rather than <count><type>-array
- Make a functor to define 256-bit vector types, use it to define float-8 type
- Make SIMD instructions pure-insns so that they participate in value numbering
2009-09-03 20:58:56 -05:00
036ff77306
add compiler comparison codes for floating-point unordered comparisons; update x86 backend to generate proper code for all floating-point comparisons
2009-09-03 20:32:05 -05:00
906a0d212a
Detect SSE version and enable the correct set of SIMD intrinsics
2009-09-03 03:28:38 -05:00
ff8c70dbe0
Initial implementation of SSE vector intrinsics:
...
- cpu.architecture: add SSE vector representations
- compiler.cfg.intrinsics.alien: remove an attempt at optimization that value numbering handles now
- compiler.cfg.representations: support instructions where the representation is set in the 'rep' slot, and support conversions between single and double floats
- alien-float, set-alien-float now use the single float representation, and the conversion is implicit; this fixes a long-standing bug where a register could get clobbered because of how %set-alien-float was defined on x86
- math.vectors.specialization: add support for SIMD specialization (where the vector word's body is replaced by another quotation), also specialize the 'sum' word
- math.vectors.simd: 4float-array, 2double-array, 4double-array types, and specializers for the math.vectors words
2009-09-03 02:33:07 -05:00
85a2bfab6c
compiler: eliminate boilerplate by centralizing info in declarative INSN: syntax
2009-09-02 06:22:37 -05:00
9595be4bf9
%box-displaced-alien: fix clobberage found by Doug
2009-08-30 05:11:08 -05:00
2bb6293217
compiler: add fixnum-min/max intrinsics; ~10% speedup on benchmark.yuv-to-rgb
2009-08-28 19:02:59 -05:00
d957ae4e44
Performance improvements to make struct-arrays benchmark faster
...
- improved optimization of ##unbox-any-c-ptr on ##box-displaced-alien; convert it to ##unbox-c-ptr where possible using class info stored in the ##bda instruction
- make fcos, fsin, etc inline again; everything in math.libm inline again, except for fsqrt which is an intrinsic
- convert min and max on floats to float-min and float-max
- make min and max not inline, so that the above can work
- struct-arrays: rice a bit so that more fixnums come up
2009-08-28 05:21:16 -05:00
98f93f799b
cpu.ppc: fix ##box-displaced-alien
2009-08-27 04:43:45 -05:00
f662e6403a
compiler: new inline intrinsic for <displaced-alien> where the inputs have known types; value numbering now eliminates unnecessary allocation of displaced aliens if the result is immediately unboxed again
2009-08-27 00:06:19 -05:00
0df8aadce2
cpu.x86: use SQRTSD instruction for math.libm:fsqrt word
2009-08-25 23:22:15 -05:00
3f3d57032b
Delete empty unit tests files, remove 1- and 1+, reorder IN: lines in a lot of places, minor refactoring
2009-08-13 19:21:44 -05:00
725280d424
Split off the notion of a register representation from a register class
2009-08-07 17:44:50 -05:00
45770c6250
Move a bunch of GC check generation logic to platform-independent side
2009-07-30 21:28:27 -05:00
be363d1a5b
compiler.cfg: Get inline GC checks working again, using a dataflow analysis to compute uninitialized stack locations in compiler.cfg.stacks.uninitialized. Re-enable intrinsics which use inline allocation
2009-07-30 09:19:44 -05:00
d71e2f9577
cpu.x86: Fix shuffle bug. Shuffling bugs occurring in code that runs before optimizer/stack checker is online are only caught at runtime during bootstrap, what a pain
2009-07-30 05:12:40 -05:00
d81dec5d45
cpu.x86: fix a bug in small-register logic on 32-bit. Also, on 32-bit, we don't need to do any special register shuffling to work with 16-bit operands since all registers have 16-bit variants. So now only 8-bit operands on x86-32 require special treatment
2009-07-30 05:04:46 -05:00
8ca17d053c
cpu.x86: use full set of 8-bit, 16-bit and 32-bit registers on x86-64 to avoid clumsy save/restore logic
2009-07-29 21:56:37 -05:00
73862a9a03
cpu.x86.assembler: move operands to operands sub-vocabulary, clean up small-reg-* code in compiler backend
2009-07-29 21:44:08 -05:00
bfb2a4c1fc
cpu.x86: compile a load of zero, and adds, subs where dst = src1 more efficiently
2009-07-27 22:27:54 -05:00
3fb4fc1bde
Improve code generation for shift word: add intrinsics for fixnum-shift-fast in the case where the shift count is not constant, transform 1 swap shift into a more overflow check with open-coded fast case, transform bitand into fixnum-bitand in more cases
2009-07-16 23:50:48 -05:00
e76dce8aff
Overflowing fixnum intrinsics now expand into several CFG nodes. This speeds up the common case since only the uncommon case is now a stack syncpoint
2009-07-16 18:29:40 -05:00
768e2a5148
compiler.cfg: split off condition codes into a comparisons sub-vocabulary
2009-07-13 14:42:52 -05:00
45a2105449
cpu.x86.assembler: IMUL2 instruction was busted for immediate operands
...
When given a register and an immediate, it would generate imul imm,dst,dst however the 64-bit prefix was generated wrong and if dst was an extended register only the first operand would be an extended register. To fix this, change IMUL2 to not work on immediates anymore, and added a new IMUL3 that takes a destination register, source register, and immediate. Also, change compiler.cfg.two-operand to not two-operandize %mul-imm, since this isn't needed anymore.
This fixes the sporadic benchmark.tuple-arrays crash on 64-bit machines.
2009-06-08 21:15:52 -05:00
9ad9600764
Remove %dispatch-label since its tehe same on all platforms; fix %gc on PowerPC
2009-06-07 21:46:28 -05:00
ade5db2405
cpu.x86: fix small register intrinsics on x86-64
2009-06-03 03:22:46 -05:00
2d231f066a
GC checks now save and restore registers
2009-06-02 18:23:47 -05:00
b389dcf441
Redo compiler.codegen.fixup and get %dispatch to work
2009-06-01 02:32:36 -05:00
fc152ef210
Various improvements aimed at getting local optimization regressions fixed:
...
- Rename _gc to ##gc
- Absolute labels are now supported
- Generate _dispatch-label
2009-05-31 23:28:08 -05:00
76d74c16af
Fixing various bugs; alias analysis wasn't handling ##phi nodes, stack analysis incorrectly handled height-changing back edges and ##fixnum-*, clean up ##dispatch generation
2009-05-29 01:39:14 -05:00
318552ba60
Fix tail call PICs on x86-64
2009-05-06 22:44:30 -05:00
581d017b46
Working on inline caching for tail call sites
2009-05-06 19:22:22 -05:00
c93d876075
Better separation of concerns: cpu.{x86,ppc}.assembler no longer depends on compiler.codegen.fixup and cpu.architecture. Rename rt-xt-direct to rt-xt-pic to better explain its purpose
2009-05-06 16:14:53 -05:00
44bfff7c7b
Rename ##load-indirect to ##load-reference since this is more descriptive; value numbering doesn't assign expressions to ##load-reference nodes since this would end up folding literals which were eq? but not =
2009-01-29 01:44:58 -06:00
8a8f0c925c
Use BSR instruction to implement fixnum-log2 intrinsic
2008-12-06 15:31:17 -06:00
a56d480aa6
Various optimizations leading to a 10% speedup on compiling empty EBNF parser:
...
- open-code getenv primitive
- inline tuple predicates in finalization
- faster partial dispatch
- faster built-in type predicates
- faster tuple predicates
- faster lo-tag dispatch
- compile V{ } clone and H{ } clone more efficiently
- add fixnum fast-path to =; avoid indirect branch if two fixnums not eq
- faster >alist on hashtables
2008-12-06 09:16:29 -06:00
82cf6530c6
set-string-nth-fast intrinsic was busted
2008-12-05 23:52:09 -06:00
e256846acd
Tweak string representation; high bit indicates if character has high bits in aux vector. Avoids memory access in common case. Split set-string-nth into two primitives; set-string-nth-fast is open-coded by optimizing compiler. 13% improvement on reverse-complement
2008-12-05 06:38:51 -06:00
e5ed7447ed
Removing more >r/r> usages
2008-12-03 08:46:16 -06:00
e7f4563374
fixnum* intrinsic for x86
2008-11-30 07:26:49 -06:00
f44506089d
More work on overflow instructions: don't need temp register anymore, add -tail variants which don't need stack frame
2008-11-28 06:36:30 -06:00
5634becda1
##fixnum-add, ##fixnum-sub instructions open-code overflow check
2008-11-28 05:33:58 -06:00
ab689c098b
Clean up direct literal code and make a first attempt at PowerPC support
2008-11-24 08:16:14 -06:00
2aaf860f47
Experimental optimizations
2008-11-24 06:40:51 -06:00
20f5541d35
Refactoring FFI for Win64
2008-11-17 13:34:37 -06:00
eb05dd3a12
Optimize a ##dispatch that is applied to the result of a ##sub-imm or ##add-imm; this eliminates an instruction from the common 1 fixnum-fast { ... } dispatch and 8 fixnum-fast { ... } dispatch code sequences appearing in generic word expansions
2008-11-13 04:16:08 -06:00
f7fe84e563
Working on Win64 FFI
2008-11-08 21:40:47 -06:00
7365959f01
Starting work on Win64 port
2008-11-07 20:33:32 -06:00
d2ec46e38f
PowerPC backend almost functional; some new compiler unit tests added,
...
better compilation of 'f eq?'; f becomes an immediate operand
move aux-offset to compiler.constants
2008-11-06 06:27:27 -06:00
53cd75b06c
Add string-nth intrinsic
2008-11-06 01:11:28 -06:00
8b7c47a68b
Clean up x86 backend: move cpu.x86.architecture to cpu.x86, use branchless arithmetic in some intrinsics
2008-11-05 04:15:48 -06:00
Slava Pestov
Joe Groff
sheeple
Doug Coleman
Slava Pestov
Slava Pestov
unknown