summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* arm/neon: timestamp counter (not for master)neonDavid Schleef2010-07-091-0/+18
| | | | | Add timestamp counter for Cortex-A8. Requires a kernel patch to use.
* memcpy_speed: preload distance check (not for master)David Schleef2010-07-091-1/+161
|
* neon: preload hinting, not for masterDavid Schleef2010-07-091-0/+15
|
* neon: Add load_constant (not for master)David Schleef2010-07-091-1/+20
|
* add performance numbers. not for masterDavid Schleef2010-07-091-0/+40
|
* neon: Fix aligned loadDavid Schleef2010-07-091-1/+1
|
* memcpy_speed: inline orc_memcpyDavid Schleef2010-07-091-1/+28
|
* memcpy_speed: flush/populate cache for hot/cold testingDavid Schleef2010-07-091-0/+42
|
* neon: Add short-array unaligned speedupDavid Schleef2010-07-092-2/+66
|
* neon: Add load/store of 256-bit vectorsDavid Schleef2010-07-091-2/+61
|
* neon: Only preload for each cache lineDavid Schleef2010-07-091-9/+11
|
* neon: Add handling of different size regionsDavid Schleef2010-07-092-21/+99
|
* arm/neon: use subs for loop counterDavid Schleef2010-07-094-14/+10
|
* neon: add preloadDavid Schleef2010-07-093-1/+30
|
* neon: add loop unrollingDavid Schleef2010-07-091-1/+6
|
* neon: Fix rules for quad usageDavid Schleef2010-07-092-64/+142
|
* memcpy_speed: take unalignment parameter on cmd lineDavid Schleef2010-07-091-6/+14
|
* neon: Fix store of l vectorsDavid Schleef2010-07-091-2/+2
|
* neon: Add quad load/storeDavid Schleef2010-07-092-14/+103
|
* neon: Use faster opcodes for aligned storeDavid Schleef2010-07-091-6/+6
|
* Add simple constant foldingDavid Schleef2010-07-091-1/+10
|
* neon: Use unaligned ops instead of mask registerDavid Schleef2010-07-092-3/+42
|
* Improve compile debuggingDavid Schleef2010-07-091-1/+4
|
* test: remove adjustment of performance numberDavid Schleef2010-07-091-1/+1
|
* test: switch neon test back to neonDavid Schleef2010-07-091-1/+1
|
* win32: Fix initialization of once mutexDavid Schleef2010-07-091-1/+9
|
* arm: Fix register movingDavid Schleef2010-07-052-1/+2
|
* arm: fix some compile mismatchesDavid Schleef2010-07-052-12/+20
|
* arm: Fix a bunch of rulesDavid Schleef2010-07-011-25/+32
|
* arm: fix compile errorDavid Schleef2010-07-011-1/+1
|
* arm: mark rules known to be buggyDavid Schleef2010-06-301-44/+49
|
* arm: Implement 2d supportDavid Schleef2010-06-301-5/+92
|
* arm: implement load/store of b, lDavid Schleef2010-06-303-6/+83
|
* arm: arm backend code gen fixesDavid Schleef2010-06-304-24/+63
|
* neon: check for NEON support on CPUDavid Schleef2010-06-304-4/+65
|
* sse: handle NANs and denormalsDavid Schleef2010-06-297-3/+114
| | | | | | Set the DAZ and FTZ flags in the MXCSR, to set proper denormal behavior. Implement NAN handling for maxf and minf.
* opcodes: Make flushing denormals the default behaviorDavid Schleef2010-06-291-4/+12
| | | | | | | | | | Flushing denormals is a better default behavior for SIMD engines, since they tend handle denormals slowly (SSE, Altivec) or not at all (NEON). Unfortunately, that behavior is expensive to emulate on IEEE-compliant FPUs, since you need to check for denormal before and after every operation. At this point, the emulator flushes denormals and C code does not.
* x86: Fix register popping for x86_64David Schleef2010-06-281-1/+1
|
* opcodes: Add code for handling denormalsDavid Schleef2010-06-281-7/+15
|
* opcodes: check for NaN in maxf/minfDavid Schleef2010-06-281-2/+2
|
* tests: better float comparisonDavid Schleef2010-06-282-7/+35
| | | | Allow denormals to be flushed to 0. Check for NaNs.
* neon: Fix accumulator opcodesDavid Schleef2010-06-282-24/+97
|
* neon: implement 2d arraysDavid Schleef2010-06-281-2/+63
|
* Fix segfaults if no targets are availableDavid Schleef2010-06-281-1/+12
|
* Add support for OpenBSD in CPU detectionDavid Schleef2010-06-282-6/+32
| | | | From Brad <brad@comstyle.com>.
* tools: add orc-bugreportDavid Schleef2010-06-262-1/+516
|
* orcc: Add --quiet and --benchmark parameters to generated test programsSebastian Dröge2010-06-261-8/+51
|
* orcc: Improve test program generationSebastian Dröge2010-06-261-9/+8
| | | | Output something if a check has passed instead of being completely silent
* target: add function to return flag namesDavid Schleef2010-06-264-3/+51
|
* arm: disable unused CPU detectionDavid Schleef2010-06-241-0/+4
|