diff options
| author | I-Jui (Ray) Sung <ijsung@google.com> | 2017-04-14 11:18:15 -0700 |
|---|---|---|
| committer | I-Jui (Ray) Sung <ijsung@google.com> | 2017-04-14 14:33:16 -0700 |
| commit | 262b4f97d0592311a16fc53ede08443ab138e5b0 (patch) | |
| tree | 4f5e907ffcbbf4aa1e249833f5182cf3d4937da4 /cpu_ref | |
| parent | 8467325c6d80de30bb86b7665cc41dff0c276e8d (diff) | |
| download | rs-262b4f97d0592311a16fc53ede08443ab138e5b0.tar.gz | |
Revive forEachDstAtop implementations for NEON/ASIMD
ScriptIntrinsicBlend's forEachDstAtop had broken NEON and ASIMD
implementations and were bypassed.
This CL fixed these two and turned them back on in
RsdCpuScriptIntrinsicBlend::kernel().
Bug: b/22047392
Test: CTS and RsTest on Angler
Change-Id: Ifa41d74222606f1c04b7b4e3fe1b43eb932ead89
(cherry picked from commit e0c6f1d59fa8e995cdd80a41cc8ed1d6af35652e)
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 |