diff options
| author | Frank Barchard <fbarchard@google.com> | 2023-01-17 14:00:35 -0800 |
|---|---|---|
| committer | libyuv LUCI CQ <libyuv-scoped@luci-project-accounts.iam.gserviceaccount.com> | 2023-01-18 00:30:05 +0000 |
| commit | 0faf8dd0e004520a61a603a4d2996d5ecc80dc3f (patch) | |
| tree | 36a46d7763ec9e14c50a5e4af3227e2131b8894e /source/row_neon64.cc | |
| parent | 541d8efbaf0319b1e068d9c31f6dbc653d5c72b8 (diff) | |
| download | libyuv-0faf8dd0e004520a61a603a4d2996d5ecc80dc3f.tar.gz | |
Fix for DivideRow_NEON functions
- was dup of 8h but mul of 4s. now use umull
Bug: libyuv:951
Change-Id: If6cb01f5f006c2235886b81ce120642d7e24a9bb
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4166563
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 | 26 |
1 files changed, 11 insertions, 15 deletions
diff --git a/source/row_neon64.cc b/source/row_neon64.cc index 088174c8..7f04b606 100644 --- a/source/row_neon64.cc +++ b/source/row_neon64.cc @@ -4461,22 +4461,18 @@ void DivideRow_16_NEON(const uint16_t* src_y, int scale, int width) { asm volatile( - "dup v6.8h, %w3 \n" + "dup v4.8h, %w3 \n" "1: \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" + "ldp q2, q3, [%0], #32 \n" + "umull v0.4s, v2.4h, v4.4h \n" + "umull2 v1.4s, v2.8h, v4.8h \n" + "umull v2.4s, v3.4h, v4.4h \n" + "umull2 v3.4s, v3.8h, v4.8h \n" "prfm pldl1keep, [%0, 448] \n" - "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" + "shrn v0.4h, v0.4s, #16 \n" + "shrn2 v0.8h, v1.4s, #16 \n" + "shrn v1.4h, v2.4s, #16 \n" + "shrn2 v1.8h, v3.4s, #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" @@ -4484,7 +4480,7 @@ void DivideRow_16_NEON(const uint16_t* src_y, "+r"(dst_y), // %1 "+r"(width) // %2 : "r"(scale) // %3 - : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6"); + : "cc", "memory", "v0", "v1", "v2", "v3", "v4"); } // Use scale to convert lsb formats to msb, depending how many bits there are: |