aboutsummaryrefslogtreecommitdiff
path: root/source/row_rvv.cc
AgeCommit message (Collapse)Author
2023-09-05[RVV] Support AR64ToAB64 and RGBA-family color conversionsBruce Lai
Add scalar code for AR64ToAB64, ARGBToRGBA, ARGBToBGRA, ARGBToABGR, RGBAToARGB, BGRAToARGB, and ABGRToARGB. They are originally implemented by ARGBShffle. This CL independetly implements them, and only enables for risc-v now. This CL also add RVV implementation for `RGBA-family <-> RGBA-family` color conversions. * Run on SiFive internal FPGA(VLEN=128): Test Case Speedup AR64ToAB64_Opt x4.6 ARGBToRGBA_Opt x6 ARGBToBGRA_Opt x6 ARGBToABGR_Opt x6 RGBAToARGB_Opt x6 Change-Id: Ie0630901046084aa259699fcdeccc64170d7103f Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4797451 Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-08-04RVV check __riscv_v_intrinsic versionFrank Barchard
Bug: libyuv:965 Change-Id: I9b02abd13ab3345288655fa7a16383f59cf66bb8 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4750230 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Mirko Bonadei <mbonadei@chromium.org>
2023-07-25[RVV] Enable ARGBBlendRow_RVV/BlendPlaneRow_RVVBruce Lai
* Run on SiFive internal FPGA: Test case Speedup ARGBBlend_Opt 4.60 BlendPlane_Opt 5.96 I420Blend_Opt 5.83 - Also, add code to use ScaleRowDown2Box_RVV in I420Blend Change-Id: Icc75e05d26b3427a98269d2a33c4474074033264 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4681100 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Wan-Teh Chang <wtc@google.com> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-07-24[RVV] Enable NV{12,21}To{ARGB,RGB24}Row_RVVDarren Hsieh
* Run on SiFive internal FPGA(w/ -march=rv64gcv): Test Case Speedup NV12ToARGB_Opt 12.0 NV21ToARGB_Opt 12.1 NV12ToABGR_Opt 12.6 NV21ToABGR_Opt 12.0 NV12ToRGB24_Opt 12.5 NV21ToRGB24_Opt 11.7 NV12ToRAW_Opt 12.1 NV21ToRAW_Opt 11.4 Change-Id: Icae2bac2b4ebbd4c5a89e847fde9a74fe6481878 Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4707804 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-06-30Fix warnings for missing prototypesFrank Barchard
- Add static to internal scale and rotate functions - Remove unittest that tested an internal scale function - Remove unused private functions - Include missing scale_argb.h header - Bump version and apply clang format Bug: libyuv:830 Change-Id: I45bab0423b86334f9707f935aedd0c6efc442dd4 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4658956 Reviewed-by: Mirko Bonadei <mbonadei@chromium.org>
2023-06-27ARGBExtractAlpha rename variables to match formatFrank Barchard
Bug: libyuv:956 Change-Id: I31070791754fc69b72c6dcc61be2e038d2676ed9 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4646636 Reviewed-by: Wan-Teh Chang <wtc@google.com>
2023-06-19[RVV] Fix TestARGBInterpolate test failBruce Lai
Root cause: Because InterpolateRow_RVV doesn't setup rounding mode to round-to-nearest-up when y1_fraction == 128. The rounding mode register is set to round-down in ARGBAttenuateRow_RVV. It cause InterpolateRow_RVV(y1_fraction == 128) runs on round-down mode. Running on round-down mode make output result differs from round-to-nearest-up mode. Solved by: ensure to use correct rounding mode in InterpolateRow_RVV. Also, removing unnecessary rounding mode setup in ARGBAttenuateRow_RVV. Bug: libyuv:956 Change-Id: Ib5265d42bad76b036e42b8f91ee42a9afe1f768d Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4624492 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-06-17[RVV] Update ARGBAttenuateRow_RVV implementationBruce Lai
Bug: libyuv:956 Change-Id: Ib539c2196767e88fa6e419ed2f22d95b6deaf406 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4623172 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-06-17[RVV] Fix compile warning in row_rvvBruce Lai
1. Fix compile warning in row_rvv.cc 2. Avoid compile row_rvv.cc/scale_rvv.cc when using GCC There is no RVV segment load & store on GCC. Hence, avoid compiling rvv code on GCC temporarily. 3. Add several compile options to cmake build flow -Wno-sign-compare -Wno-unused-function -Wunused-variable -Wuninitialized Bug: libyuv:956 Change-Id: I9577f98190fc9b28fb6fde65d82d0c67ce54f9ee Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4615441 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-06-15[RVV] Enable ARGBExtractAlphaRow/ARGBCopyYToAlphaRowBruce Lai
* Run on SiFive internal FPGA: TestARGBExtractAlpha(~3.2x vs scalar) TestARGBCopyYToAlpha(~1.6x vs scalar) Change-Id: I36525c67e8ac3f71ea9d1a58c7dc15a4009d9da1 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4617955 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-30[RVV] Use LMUL=2 for I4{44,22}To{ARGB,RGB24,RGBA} conversionBruce Lai
Replace vv+m1(LMUL=1) with vx+m2(LMUL=2). Some kernels' asm code might contain register spill(1~2). Change-Id: Ie3655f250d17f37c1ba9039474ece43ede98ede0 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4573159 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-30[RVV] Enable CopyRow_RVV, InterpolateRow_RVV, {Merge,Split}UVRow_RVVDarren Hsieh
* Run on SiFive internal FPGA: MergeUVPlane_Opt(~6x vs scalar) SplitUVPlane_Opt(~6x vs scalar) TestCopyPlane(~8x vs scalar) ARGBInterpolate0_Opt(~10x vs scalar) ARGBInterpolate64_Opt(~9x vs scalar) ARGBInterpolate168_Opt(~9x vs scalar) ARGBInterpolate192_Opt(~8.5x vs scalar) ARGBInterpolate255_Opt(~8x vs scalar) Bug: libyuv:956 Change-Id: I8372341865f75f42e30371ef943d5c2e4be7b79a Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4574186 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-05-22Manually remove rounding value inside yb(yuvconstant) in row_rvv.ccBruce Lai
After libyuv:961 is completed, yb(yuvconstant) will no longer contain rounding bias +32 for fixed-point. This CL removes rounding bias(-32) manmually in row_rvv.cc. Hence, all fixed-point related codes' rounding mode is changed to round-to-nearest-up "0" in row_rvv.cc. Also, replace vwmul+vnsrl w/ vmulh in I400ToARGBRow_RVV. Bug: libyuv:956, libyuv:961 Change-Id: I10e34668a2332e38393e9d68414f07aafb6c7cf7 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4550591 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-20Enable {J400/I400}ToARGBRow_RVVWan-Teh Chang
Run on SiFive internal FPGA*: I400ToARGB_Opt (~8x vs scalar) J400ToARGB_Opt (~10x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Bug: libyuv:956, libyuv:961 Change-Id: If4e21ec85c4ff79083ec16a6faae0e457129a8de Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4544972 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Wan-Teh Chang <wtc@google.com>
2023-05-17ARGBToI420Alpha function to convert ARGB to I420 with AlphaFrank Barchard
Bug: b/281866362 Change-Id: Ic1093a887fb483f134c78909cf1ee7495e7345ba Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4534100 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Wan-Teh Chang <wtc@google.com>
2023-05-16Enable I{422,444}AlphaToARGBRow_RVV & ARGBAttentuateRow_RVVBruce Lai
Run on SiFive internal FPGA: I444AlphaToARGB_Opt (~16x vs scalar) I422AlphaToARGB_Opt (~10x vs scalar) ARGBAttenuate_Opt (~3x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Change-Id: I0046eb7af8104bc8e13cee1cb91a19f90940d5b0 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4535657 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-11Enable ARGBToYMatrixRow_RVV/RGBAToYMatrixRow_RVV/RGBToYMatrixRow_RVVBruce Lai
Run on SiFive internal FPGA: ARGBToJ400_Opt (~6x vs scalar) RGBAToJ400_Opt (~6x vs scalar) RGB24ToJ400_Opt (~5.5x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Change-Id: Ia3ce8cea7962fbd8618cc23e850a7913c9cabf4f Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4521783 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-05-10Enable I444To{ARGB,RGB24}Row_RVVDarren Hsieh
Run on SiFive internal FPGA: I444ToARGB_Opt (~16x vs scalar) I444ToRGB24_Opt (~10x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Change-Id: Idae7dc46ef648beaa14b58ba3eb56b67b17c9b3b Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4520761 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-10Enable I422To{ARGB,RGBA,RGB24}Row_RVVDarren Hsieh
Run on SiFive internal FPGA: I422ToARGB_Opt (~10x vs scalar) I422ToRGBA_Opt (~10x vs scalar) I420ToRGB24_Opt (~8x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 This CL manually sets rounding mode, since we use fixed-point vector narrowing clip. There is no definition about default value for fixed-point rounding mode. https://github.com/riscv/riscv-v-spec/blob/master/v-spec.adoc#38-vector-fixed-point-rounding-mode-register-vxrm The behavior could be different on differet paltforms. To avoid unexpected behavior, we set rounding mode manually. Change-Id: I90f0dcb90c37f7da7caab8eb1df6c9c7a3c874a8 Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4512373 Reviewed-by: Wan-Teh Chang <wtc@google.com> Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-05-04Fix compile error for riscv scalar & simplify cmake cross build flowBruce Lai
1. Fix compile error when build riscv without using vector 2. Fix run_qemu.sh misused v=true for USE_RVV=OFF case 3. [cmake] Fix warning by rename TEST to UNIT_TEST Warning log: CMake Warning (dev) at CMakeLists.txt:57 (if): [54/1931] Policy CMP0064 is not set: Support new TEST if() operator. Run "cmake --help-policy CMP0064" for policy details. Use the cmake_policy command to set the policy and suppress this warning. TEST will be interpreted as an operator when the policy is set to NEW. Since the policy is not set the OLD behavior will be used. This warning is for project developers. Use -Wno-dev to suppress it. 4. [cmake] Simplify logic for cross-build Bug: libyuv:956 Change-Id: I120402fc7d6d86403e7d974180b81f4f9c663e36 Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4486239 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Mirko Bonadei <mbonadei@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-04-28Add Split/Merge RGB/ARGB/XRGB Row_RVVDarren Hsieh
* Run on SiFive internal FPGA: SplitRGBPlane_Opt (~6.87x vs scalar) SplitARGBPlane_Opt (~10.77x vs scalar) SplitXRGBPlane_Opt (~18.69x vs scalar) MergeRGBPlane_Opt (~3.63x vs scalar) MergeARGBPlane_Opt (~3.50x vs scalar) MergeXRGBPlane_Opt (~2.90x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 - include a fix to avoid implict conversion warning between size_t & int. Bug: libyuv:956 Change-Id: Icd79b282b04ea3981e7fd4e6d547da6708d82516 Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4443411 Commit-Queue: Frank Barchard <fbarchard@chromium.org> Reviewed-by: Frank Barchard <fbarchard@chromium.org>
2023-04-20Optimized AR64/AB64 <-> ARGB with RVVBruce Lai
* Run on SiFive internal FPGA: ARGBToAR64_Opt (~13.7x vs scalar) ARGBToAB64_Opt (~5.81x vs scalar) AR64ToARGB_Opt (~15.8x vs scalar) AB64ToARGB_Opt (~2.40x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Bug: libyuv:956 Change-Id: Ida642a5077f59d25fb7c5328f671956b2293dadd Signed-off-by: Bruce Lai <bruce.lai@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4442913 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-04-13Add ARGBToRAWRow_RVV, ARGBToRGB24Row_RVV, RGB24ToARGBRow_RVVDarren Hsieh
* Run on SiFive internal FPGA: ARGBToRAW_Opt (~1.55x vs scalar) ARGBToRGB24_Opt (~1.44x vs scalar) RGB24ToARGB_Opt (~1.77x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Bug: libyuv:956 Change-Id: I26722f6848cd68684d95d9a7ee06ce0416e7985d Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4413083 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
2023-04-07Add RAWToARGBRow_RVV,RAWToRGBARow_RVV,RAWToRGB24Row_RVVDarren Hsieh
* Run on SiFive internal FPGA: RAWToARGB_Opt (~2x vs scalar) RAWToRGBA_Opt (~2x vs scalar) RAWToRGB24_Opt (~1.5x vs scalar) LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=10 Change-Id: I21a13d646589ea2aa3822cb9225f5191068c285b Signed-off-by: Darren Hsieh <darren.hsieh@sifive.com> Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4408357 Reviewed-by: Frank Barchard <fbarchard@chromium.org> Commit-Queue: Frank Barchard <fbarchard@chromium.org>
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-08 06:42:42 +0000