273 commits

Author	SHA1	Message	Date
unknown	156e4cac52	fixed problem ordering zero-extension before add	2013-11-20 12:09:53 -05:00
unknown	daa2c258b2	better jump table using function look-up (smaller & faster)	2013-10-11 03:41:21 -04:00
unknown	cd77984576	manual override of GCC's slower decode of SA IW bitmask	2013-10-11 02:27:18 -04:00
unknown	56c5ac9daf	BIG speed-up by moving shuffles out of EX queue, into VU ops	2013-10-11 00:40:41 -04:00
unknown	4c2b671d53	deprecated 2-D opcode-element vector jump table	2013-10-10 23:04:07 -04:00
unknown	f0b8985bda	moved SSE2 declare macro to MAKE/GCC command script	2013-10-07 23:37:06 -04:00
unknown	34c6065842	add SSSE3 shuffling alternative straight off CEN64 wisdom	2013-10-07 22:44:32 -04:00
unknown	adced4b284	fixed VMACU overflow mask, more direct VMULU speed clamp	2013-10-07 06:17:06 -04:00
unknown	690c25008e	much faster unsigned clamp for VMACU/MULU, small VMADN jump	2013-10-07 03:45:31 -04:00
unknown	38a03b2566	found similar fraction delay speed-ups from MUL as with MAC	2013-10-06 09:49:43 -04:00
unknown	2042e783a4	new semi-fraction rounding delay technique to optimize MUL	2013-10-06 09:24:27 -04:00
unknown	b7656289a5	micro-optimizations to basic MAC of fractions	2013-10-06 08:10:58 -04:00
unknown	3870e93fa8	fix MusyX MP3 signed fractions compressor	2013-10-05 23:59:55 -04:00
unknown	9d1e64e935	moved everything about RSP flags to new SSE2-hybrid header	2013-10-03 21:06:41 -04:00
unknown	390f0d2ad0	minimize automated pack/unpack extensions in VADDC/VSUBC	2013-10-02 01:35:06 -04:00
unknown	271f6cae5f	fix interposed clamp problems in VADD/VSUB, wipe old crap	2013-10-02 00:15:56 -04:00
unknown	9af8ba0f57	further compacted clamp into saturated add/sub	2013-09-29 02:11:00 -04:00
unknown	dafee07a5a	switch to smaller dynamic shuffling resource	2013-09-28 22:34:38 -04:00
unknown	a661e72c64	amend code generation bug in GCC 4.8.1 vectorizer	2013-09-28 19:55:57 -04:00
unknown	4bdbbfdff0	rewired low clamper scheme, moved VMRG to a schematic	2013-09-28 16:39:12 -04:00
unknown	33eb07512a	simplified regular signed clamping with straight SSE	2013-09-28 04:38:32 -04:00
unknown	148aa0f7a4	microoptimizations to uncommon clamps via MarathonMan's intrinsics	2013-09-27 22:17:06 -04:00
unknown	b6fc11a983	fix build issue when compiling without SSE2 support	2013-09-26 03:51:46 -04:00
unknown	b0d38d05b0	updates to the source directory structure, a few ANSI tweaks	2013-09-24 02:36:00 -04:00
unknown	d226d4b693	tl;dr	2013-09-23 14:33:34 -04:00
unknown	342488f056	integrated signed clamp with VADD/VSUB	2013-09-23 14:29:18 -04:00
unknown	bdf5e3c068	snuck in sign-extension bug just before last commit >.<	2013-09-23 05:25:52 -04:00
unknown	85fe9d7081	destroyed global result clamp buffer, lots of extra SSE2 ops cut out	2013-09-23 05:14:33 -04:00
unknown	d29cbbe3e2	more MAC micro-optimizations, split clamping to new header	2013-09-23 03:30:24 -04:00
unknown	256ffb4b57	phased out some more excess multiply packs/unpacks	2013-09-22 20:21:35 -04:00
unknown	dde3f6d456	once again, ensure 128-bit VR<--ACC writes, not memcpy bytes	2013-09-22 06:11:30 -04:00
unknown	3c1a9c1cb4	various upgrades and microoptimizations to ADD group	2013-09-22 05:36:43 -04:00
unknown	19b571ff43	wrong byte-iterative copy precision caused scalar acc R/W	2013-09-22 04:52:45 -04:00
unknown	c427e052fe	purified vectors for complex RSP clip selects, fix VCL	2013-09-22 04:39:23 -04:00
unknown	b960b8759f	purify parallel executions for simple select ops	2013-09-21 23:43:45 -04:00
cxd4	3458546391	Merge pull request #1 from tj90241/master ... Do not pass -m3dnow to gcc.	2013-09-21 15:16:34 -07:00
unknown	5c1ab5806b	unify shuffles to new one-dimensional vector op-code jumping	2013-09-21 18:16:13 -04:00
Tyler Stachecki	e5ad629b5a	Make this array static.	2013-09-21 02:21:18 -04:00
unknown	d064b82976	split shuffle-related stuff to its own header	2013-09-20 23:49:05 -04:00
unknown	70a45f8bd7	fix Visual Studio interpretation of VR "re-def"	2013-09-20 15:59:27 -04:00
unknown	62e5d5cecd	factored out 16-bit VMACU segments into clamp mode	2013-09-20 15:57:06 -04:00
unknown	d757e41676	force 16-byte alignment, divide clamp base formula to each op	2013-09-20 11:56:10 -04:00
unknown	bbebbcc81a	uninstalled dynamic vector/scalar coefficient global	2013-09-20 00:36:26 -04:00
unknown	3c96c25950	fix compiler static over-optimized SSE2 build faults	2013-09-19 20:14:03 -04:00
unknown	d69229f3f2	unify scalar whole shuffling to SSE2 generator	2013-09-19 05:13:45 -04:00
unknown	139864c71f	forgot to apply N macro to the divides	2013-09-19 03:23:09 -04:00
unknown	d690c5f1c9	SSE2-shuffled clip/select ops, and that's the last of them.	2013-09-19 02:11:10 -04:00
unknown	d6b475855a	all shuffling for vector add group vectorized	2013-09-18 23:24:08 -04:00
unknown	485a04bc60	divide group shuffling now completely vectorized	2013-09-18 22:00:00 -04:00
unknown	cd99d4306b	no more ugly scalar shuffling in logical vector group	2013-09-18 03:45:53 -04:00