diff options
| author | I-Jui (Ray) Sung <ijsung@google.com> | 2017-04-14 21:30:33 +0000 |
|---|---|---|
| committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2017-04-14 21:30:34 +0000 |
| commit | 0919e8578e4f4bd0a7d48fb719b88a8b98a29aa9 (patch) | |
| tree | bd53b46775a964e0f4101f34300cec02568ec6dd /cpu_ref | |
| parent | 34687bab1eb625d26861401cdcd1fb9b4d07999b (diff) | |
| parent | e0c6f1d59fa8e995cdd80a41cc8ed1d6af35652e (diff) | |
| download | rs-0919e8578e4f4bd0a7d48fb719b88a8b98a29aa9.tar.gz | |
Merge "Revive forEachDstAtop implementations for NEON/ASIMD"
Diffstat (limited to 'cpu_ref')
| -rw-r--r-- | cpu_ref/rsCpuIntrinsicBlend.cpp | 4 | ||||
| -rw-r--r-- | cpu_ref/rsCpuIntrinsics_advsimd_Blend.S | 2 | ||||
| -rw-r--r-- | cpu_ref/rsCpuIntrinsics_neon_Blend.S | 2 |
3 files changed, 3 insertions, 5 deletions
diff --git a/cpu_ref/rsCpuIntrinsicBlend.cpp b/cpu_ref/rsCpuIntrinsicBlend.cpp index 1507d452..ce300927 100644 --- a/cpu_ref/rsCpuIntrinsicBlend.cpp +++ b/cpu_ref/rsCpuIntrinsicBlend.cpp @@ -119,9 +119,7 @@ void RsdCpuScriptIntrinsicBlend::kernel(const RsExpandKernelDriverInfo *info, uint32_t x2 = xend; #if defined(ARCH_ARM_USE_INTRINSICS) - // Bug: 22047392 - Skip optimized version for BLEND_DST_ATOP until this - // been fixed. - if (gArchUseSIMD && info->slot != BLEND_DST_ATOP) { + if (gArchUseSIMD) { if (rsdIntrinsicBlend_K(out, in, info->slot, x1, x2) >= 0) return; } diff --git a/cpu_ref/rsCpuIntrinsics_advsimd_Blend.S b/cpu_ref/rsCpuIntrinsics_advsimd_Blend.S index 90dbd73c..1c721b63 100644 --- a/cpu_ref/rsCpuIntrinsics_advsimd_Blend.S +++ b/cpu_ref/rsCpuIntrinsics_advsimd_Blend.S @@ -350,7 +350,7 @@ uqrshrn v2.8b, v2.8h, #8 uqrshrn2 v2.16b, v14.8h, #8 - mvn v3.16b, v3.16b + mov v3.16b, v11.16b .endm #define params_MULTIPLY zipped=0 diff --git a/cpu_ref/rsCpuIntrinsics_neon_Blend.S b/cpu_ref/rsCpuIntrinsics_neon_Blend.S index f0145e54..a1fa1b50 100644 --- a/cpu_ref/rsCpuIntrinsics_neon_Blend.S +++ b/cpu_ref/rsCpuIntrinsics_neon_Blend.S @@ -353,7 +353,7 @@ vqrshrn.u16 d4, q2, #8 vqrshrn.u16 d5, q14, #8 - vmvn q3, q3 + vmov q3, q11 .endm #define params_MULTIPLY zipped=0 |