diff options
| author | Darren Hsieh <darren.hsieh@sifive.com> | 2023-04-11 00:05:48 -0700 |
|---|---|---|
| committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2023-04-28 18:34:46 +0000 |
| commit | 1b3c4c12d4b7972b6656438a37949309bfb2c18a (patch) | |
| tree | 78a2d5ad44167b9dc5a7328b9b53b5e6218a5b74 /include | |
| parent | 7c6a7e5737ec0afa12f132e8d1831d5ffd9ad623 (diff) | |
| download | libyuv-1b3c4c12d4b7972b6656438a37949309bfb2c18a.tar.gz | |
Add Split/Merge RGB/ARGB/XRGB Row_RVV
* 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>
Diffstat (limited to 'include')
| -rw-r--r-- | include/libyuv/row.h | 38 |
1 files changed, 38 insertions, 0 deletions
diff --git a/include/libyuv/row.h b/include/libyuv/row.h index 176b3781..53a8d8a6 100644 --- a/include/libyuv/row.h +++ b/include/libyuv/row.h @@ -764,6 +764,12 @@ extern "C" { #define HAS_ARGBTOAR64ROW_RVV #define HAS_ARGBTORAWROW_RVV #define HAS_ARGBTORGB24ROW_RVV +#define HAS_MERGEARGBROW_RVV +#define HAS_MERGERGBROW_RVV +#define HAS_MERGEXRGBROW_RVV +#define HAS_SPLITARGBROW_RVV +#define HAS_SPLITRGBROW_RVV +#define HAS_SPLITXRGBROW_RVV #define HAS_RAWTOARGBROW_RVV #define HAS_RAWTORGB24ROW_RVV #define HAS_RAWTORGBAROW_RVV @@ -2282,6 +2288,11 @@ void SplitRGBRow_NEON(const uint8_t* src_rgb, uint8_t* dst_g, uint8_t* dst_b, int width); +void SplitRGBRow_RVV(const uint8_t* src_rgb, + uint8_t* dst_r, + uint8_t* dst_g, + uint8_t* dst_b, + int width); void SplitRGBRow_Any_SSSE3(const uint8_t* src_ptr, uint8_t* dst_r, uint8_t* dst_g, @@ -2308,6 +2319,11 @@ void MergeRGBRow_NEON(const uint8_t* src_r, const uint8_t* src_b, uint8_t* dst_rgb, int width); +void MergeRGBRow_RVV(const uint8_t* src_r, + const uint8_t* src_g, + const uint8_t* src_b, + uint8_t* dst_rgb, + int width); void MergeRGBRow_Any_SSSE3(const uint8_t* y_buf, const uint8_t* u_buf, const uint8_t* v_buf, @@ -2342,6 +2358,12 @@ void MergeARGBRow_NEON(const uint8_t* src_r, const uint8_t* src_a, uint8_t* dst_argb, int width); +void MergeARGBRow_RVV(const uint8_t* src_r, + const uint8_t* src_g, + const uint8_t* src_b, + const uint8_t* src_a, + uint8_t* dst_argb, + int width); void MergeARGBRow_Any_SSE2(const uint8_t* y_buf, const uint8_t* u_buf, const uint8_t* v_buf, @@ -2390,6 +2412,12 @@ void SplitARGBRow_NEON(const uint8_t* src_rgba, uint8_t* dst_b, uint8_t* dst_a, int width); +void SplitARGBRow_RVV(const uint8_t* src_rgba, + uint8_t* dst_r, + uint8_t* dst_g, + uint8_t* dst_b, + uint8_t* dst_a, + int width); void SplitARGBRow_Any_SSE2(const uint8_t* src_ptr, uint8_t* dst_r, uint8_t* dst_g, @@ -2434,6 +2462,11 @@ void MergeXRGBRow_NEON(const uint8_t* src_r, const uint8_t* src_b, uint8_t* dst_argb, int width); +void MergeXRGBRow_RVV(const uint8_t* src_r, + const uint8_t* src_g, + const uint8_t* src_b, + uint8_t* dst_argb, + int width); void MergeXRGBRow_Any_SSE2(const uint8_t* y_buf, const uint8_t* u_buf, const uint8_t* v_buf, @@ -2474,6 +2507,11 @@ void SplitXRGBRow_NEON(const uint8_t* src_rgba, uint8_t* dst_g, uint8_t* dst_b, int width); +void SplitXRGBRow_RVV(const uint8_t* src_rgba, + uint8_t* dst_r, + uint8_t* dst_g, + uint8_t* dst_b, + int width); void SplitXRGBRow_Any_SSE2(const uint8_t* src_ptr, uint8_t* dst_r, uint8_t* dst_g, |