diff options
| author | Frank Barchard <fbarchard@google.com> | 2023-01-17 13:06:38 -0800 |
|---|---|---|
| committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2023-01-17 21:40:45 +0000 |
| commit | 541d8efbaf0319b1e068d9c31f6dbc653d5c72b8 (patch) | |
| tree | 51cb0f426435b7b3d9f5553144300772806053e8 /source/row_neon.cc | |
| parent | d5aa3d4a76930cde9527f5be5bc89524b48fc069 (diff) | |
| download | libyuv-541d8efbaf0319b1e068d9c31f6dbc653d5c72b8.tar.gz | |
Fix for divide row functions used by P010ToI010
Bug: libyuv:951
Change-Id: Id323656cb6f99b1be0be7aaa854d3cc15feeba69
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4166562
Reviewed-by: Justin Green <greenjustin@google.com>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Diffstat (limited to 'source/row_neon.cc')
| -rw-r--r-- | source/row_neon.cc | 36 |
1 files changed, 17 insertions, 19 deletions
diff --git a/source/row_neon.cc b/source/row_neon.cc index 0611b9aa..416f112f 100644 --- a/source/row_neon.cc +++ b/source/row_neon.cc @@ -3911,31 +3911,29 @@ void DivideRow_16_NEON(const uint16_t* src_y, int scale, int width) { asm volatile( - "vdup.16 q0, %3 \n" - "1: \n" - "vld1.16 {q1}, [%0]! \n" - "vld1.16 {q2}, [%0]! \n" - "vmovl.u16 q3, d2 \n" - "vmovl.u16 q1, d3 \n" - "vmovl.u16 q4, d4 \n" - "vmovl.u16 q2, d5 \n" - "vshl.u32 q3, q3, q0 \n" - "vshl.u32 q4, q4, q0 \n" - "vshl.u32 q1, q1, q0 \n" - "vshl.u32 q2, q2, q0 \n" - "vmovn.u32 d2, q3 \n" - "vmovn.u32 d3, q1 \n" - "vmovn.u32 d4, q4 \n" - "vmovn.u32 d5, q2 \n" - "vst1.16 {q1}, [%1]! \n" - "vst1.16 {q2}, [%1]! \n" + "vdup.16 q6, %3 \n" + "1: \n" + "vld1.16 {q0, q1}, [%0]! \n" + "vmovl.u16 q2, d0 \n" + "vmovl.u16 q3, d1 \n" + "vmovl.u16 q4, d2 \n" + "vmovl.u16 q5, d3 \n" + "vmul.u32 q2, q2, q6 \n" + "vmul.u32 q3, q3, q6 \n" + "vmul.u32 q4, q4, q6 \n" + "vmul.u32 q5, q5, q6 \n" + "vshrn.u32 d0, q2, #16 \n" + "vshrn.u32 d1, q3, #16 \n" + "vshrn.u32 d2, q4, #16 \n" + "vshrn.u32 d3, q5, #16 \n" + "vst1.16 {q0, q1}, [%1]! \n" // store 16 pixels "subs %2, %2, #16 \n" // 16 src pixels per loop "bgt 1b \n" : "+r"(src_y), // %0 "+r"(dst_y), // %1 "+r"(width) // %2 : "r"(scale) // %3 - : "cc", "memory", "q0", "q1", "q2", "q3", "q4"); + : "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6"); } // Use scale to convert lsb formats to msb, depending how many bits there are: |