aboutsummaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* Avoid using unsigned values for signedChris Robinson2018-05-152-1/+11
|
* Avoid unnecessary function-like macrosChris Robinson2018-05-151-5/+5
|
* Use a step counter for gain steppingChris Robinson2018-05-144-76/+100
| | | | | This should provide more stable stepping, preventing floating-point errors from accumulating on each step/sample.
* Add a faster double-to-int converter for x87 buildsChris Robinson2018-05-141-1/+25
|
* Use more accurate elevations for virtual HRTF speaker positionsChris Robinson2018-05-141-8/+8
|
* More accurately convert between degrees and radiansChris Robinson2018-05-141-2/+2
|
* Make a faster float2int method for x87 targetsChris Robinson2018-05-121-1/+24
|
* Use fastf2i instead of manually rounding with float2intChris Robinson2018-05-121-4/+4
|
* Fix MSVCChris Robinson2018-05-121-3/+3
|
* Fix non-SEE (32-bit) GCC buildsChris Robinson2018-05-121-1/+1
|
* Add and use a method for fast float roundingChris Robinson2018-05-123-1/+56
| | | | | Unlike fastf2i, this keeps the result as a float instead of converting to integer.
* Simplify calculating the HRTF B-Format IR lengthChris Robinson2018-05-041-19/+19
|
* Don't assume the FPU is round-to-zero in the pitch shifterChris Robinson2018-05-041-21/+3
|
* Another fix attempt for 32-bit MSVCChris Robinson2018-05-041-5/+5
|
* Try to fix 32-bit MSVC buildsChris Robinson2018-05-041-1/+1
|
* Don't assume round-to-zero for fastf2iChris Robinson2018-05-044-79/+47
|
* Add a specific function for truncating float-to-int conversionsChris Robinson2018-05-038-40/+40
|
* Avoid fastf2i in the converter initChris Robinson2018-05-031-1/+2
|
* Use a fixed-point scale for the pitch shifter frequency indexChris Robinson2018-05-031-5/+10
|
* Avoid excessive if block depthsChris Robinson2018-05-021-12/+6
|
* Avoid duplication for getting the cpuidChris Robinson2018-05-021-55/+24
|
* Pass in the number of channels per order to InitNearFieldCtrlChris Robinson2018-05-011-26/+14
|
* An output device buffer is likelyChris Robinson2018-04-291-1/+1
|
* Avoid potentially calling log10f(0)Chris Robinson2018-04-291-1/+1
|
* Rename some struct members for clarityChris Robinson2018-04-282-77/+77
|
* Fix a comment about a float's mantissaChris Robinson2018-04-281-3/+4
|
* Remove some unnecessary floor callsChris Robinson2018-04-281-4/+2
|
* Increase the band-split IR for decoding ambisonics to HRTFChris Robinson2018-04-271-8/+16
|
* Rename some variables for clarityChris Robinson2018-04-271-31/+29
|
* Transpose the band-splitter all-pass sectionChris Robinson2018-04-261-9/+11
|
* Use doubles for the pitch shifter's FFTs and processingChris Robinson2018-04-241-43/+68
|
* Update ChangeLogChris Robinson2018-04-241-27/+25
|
* Don't specialize biquad processing for a single sampleChris Robinson2018-04-241-37/+28
|
* Use transposed direct form 2 for the UHJ all-pass filtersChris Robinson2018-04-222-17/+17
| | | | | This has one extra multiply, but avoids two moves and uses almost half as much memory for the encoder state.
* Pre-calculate the squares of the UHJ filter coefficientsChris Robinson2018-04-221-53/+31
|
* Change some if checks to asserts since they must be trueChris Robinson2018-04-221-14/+12
|
* Include header files in CMake's source listsChris Robinson2018-04-212-52/+99
|
* Move the bnad-splitter filters to a separate sourceChris Robinson2018-04-218-136/+152
|
* Only use fast float-to-int workarounds for x87Chris Robinson2018-04-211-0/+18
| | | | | | At least SSE and ARM have opcodes that handle float-to-int conversions well enough. Also, Clang doesn't inline lrintf, incurring function call overhead for what should be a single opcode.
* Add some more ASSUME statementsChris Robinson2018-04-214-0/+7
|
* Improve ASSUME for ClangChris Robinson2018-04-211-0/+9
| | | | | | | | For some reason, the { if(!x)__builtin_unreachable(); } construct does not provide the same optimization opportunity for Clang (even though the condition being false would trigger undefined behavior by reaching unreachable code, it still performs checks and such for the condition potentially being false). Using __builtin_assume seems to work better.
* Adjust the stereo panning front gainChris Robinson2018-04-191-2/+2
| | | | This gives it a (more) precise -4.5dB gain drop for front-center panned sounds.
* Remove unnecessary undefsChris Robinson2018-04-193-3/+0
|
* Clear ALSA's PCM handle after closing itChris Robinson2018-04-191-0/+1
|
* Add some ASSUME statements that ensure mixing at least 1 sampleChris Robinson2018-04-186-0/+25
|
* Add an ASSUME macro that requires a true conditionChris Robinson2018-04-174-0/+23
|
* Accumulate the B-Format HRTF responses using doublesChris Robinson2018-04-171-9/+24
| | | | | The final result is still truncated to single-precision float, but this should keep the responses more stable as it accumulates the various inputs.
* Don't minimize the HRTF per-response delayChris Robinson2018-04-162-11/+7
| | | | | | The reverts both fa9ab9af7cb559ff9ecc1846f2996265bfbea1ec and 79604c3c0e0f3f71832a09348cc273a38882cc3e. As helpful as it was for the high frequencies, the overall response's gain suffered.
* Preliminary ChangeLog updateChris Robinson2018-04-151-0/+50
|
* Update .gitignoreChris Robinson2018-04-141-6/+4
|
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-12 13:16:58 +0000