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_neon64.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_neon64.cc')
-rw-r--r-- | source/row_neon64.cc | 32 |
1 files changed, 16 insertions, 16 deletions
diff --git a/source/row_neon64.cc b/source/row_neon64.cc index 41289fe9..088174c8 100644 --- a/source/row_neon64.cc +++ b/source/row_neon64.cc @@ -4461,30 +4461,30 @@ void DivideRow_16_NEON(const uint16_t* src_y, int scale, int width) { asm volatile( - "dup v0.8h, %w3 \n" + "dup v6.8h, %w3 \n" "1: \n" - "ldp q1, q2, [%0], #32 \n" - "ushll v3.4s, v1.4h, #0 \n" - "ushll v4.4s, v2.4h, #0 \n" + "ldp q0, q1, [%0], #32 \n" + "ushll v2.4s, v0.4h, #0 \n" + "ushll2 v3.4s, v0.8h, #0 \n" + "ushll v4.4s, v1.4h, #0 \n" + "ushll2 v5.4s, v1.8h, #0 \n" "prfm pldl1keep, [%0, 448] \n" - "ushll2 v1.4s, v1.8h, #0 \n" - "ushll2 v2.4s, v2.8h, #0 \n" - "mul v3.4s, v0.4s, v3.4s \n" - "mul v4.4s, v0.4s, v4.4s \n" - "mul v1.4s, v0.4s, v1.4s \n" - "mul v2.4s, v0.4s, v2.4s \n" - "shrn v3.4h, v3.4s, #16 \n" - "shrn v4.4h, v4.4s, #16 \n" - "shrn2 v3.8h, v1.4s, #16 \n" - "shrn2 v4.8h, v2.4s, #16 \n" - "stp q3, q3, [%1], #32 \n" // store 16 pixels + "mul v2.4s, v2.4s, v6.4s \n" + "mul v3.4s, v3.4s, v6.4s \n" + "mul v4.4s, v4.4s, v6.4s \n" + "mul v5.4s, v5.4s, v6.4s \n" + "shrn v0.4h, v2.4s, #16 \n" + "shrn2 v0.8h, v3.4s, #16 \n" + "shrn v1.4h, v4.4s, #16 \n" + "shrn2 v1.8h, v5.4s, #16 \n" + "stp q0, q1, [%1], #32 \n" // store 16 pixels "subs %w2, %w2, #16 \n" // 16 src pixels per loop "b.gt 1b \n" : "+r"(src_y), // %0 "+r"(dst_y), // %1 "+r"(width) // %2 : "r"(scale) // %3 - : "cc", "memory", "v0", "v1", "v2", "v3", "v4"); + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6"); } // Use scale to convert lsb formats to msb, depending how many bits there are: |