Add a standard bit depth Neon asm implementation for pixel_ssd_s
for block size 64x64.
A high bit depth Neon asm implementation already exists just enable
it.
---
source/common/aarch64/asm-primitives.cpp | 2 ++
source/common/aarch64/ssd-a.S | 35 ++++++++++++++++++++++++
2 files changed, 37 insertions(+)
diff --git a/source/common/aarch64/asm-primitives.cpp
b/source/common/aarch64/asm-primitives.cpp
index c1f41cb3c..c1317eb74 100644
--- a/source/common/aarch64/asm-primitives.cpp
+++ b/source/common/aarch64/asm-primitives.cpp
@@ -557,11 +557,13 @@ void setupNeonPrimitives(EncoderPrimitives &p)
p.cu[BLOCK_8x8].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_8x8_neon);
p.cu[BLOCK_16x16].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_16x16_neon);
p.cu[BLOCK_32x32].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_32x32_neon);
+ p.cu[BLOCK_64x64].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_64x64_neon);
p.cu[BLOCK_4x4].ssd_s[ALIGNED] = PFX(pixel_ssd_s_4x4_neon);
p.cu[BLOCK_8x8].ssd_s[ALIGNED] = PFX(pixel_ssd_s_8x8_neon);
p.cu[BLOCK_16x16].ssd_s[ALIGNED] = PFX(pixel_ssd_s_16x16_neon);
p.cu[BLOCK_32x32].ssd_s[ALIGNED] = PFX(pixel_ssd_s_32x32_neon);
+ p.cu[BLOCK_64x64].ssd_s[ALIGNED] = PFX(pixel_ssd_s_64x64_neon);
#if !HIGH_BIT_DEPTH
// pixel_avg_pp
diff --git a/source/common/aarch64/ssd-a.S b/source/common/aarch64/ssd-a.S
index f4d612137..e5da67679 100644
--- a/source/common/aarch64/ssd-a.S
+++ b/source/common/aarch64/ssd-a.S
@@ -406,6 +406,41 @@ function PFX(pixel_ssd_s_32x32_neon)
ret_v0_w0
endfunc
+function PFX(pixel_ssd_s_64x64_neon)
+ add x1, x1, x1
+ sub x1, x1, #64
+ sub x3, x3, #64
+
+ mov w12, #32
+ movi v0.16b, #0
+ movi v1.16b, #0
+.Loop_ssd_ss_64:
+ sub w12, w12, #1
+.rept 2
+ ld1 {v16.16b-v19.16b}, [x0], #64
+ ld1 {v20.16b-v23.16b}, [x0], x1
+ smlal v0.4s, v16.4h, v16.4h
+ smlal2 v1.4s, v16.8h, v16.8h
+ smlal v0.4s, v17.4h, v17.4h
+ smlal2 v1.4s, v17.8h, v17.8h
+ smlal v0.4s, v18.4h, v18.4h
+ smlal2 v1.4s, v18.8h, v18.8h
+ smlal v0.4s, v19.4h, v19.4h
+ smlal2 v1.4s, v19.8h, v19.8h
+ smlal v0.4s, v20.4h, v20.4h
+ smlal2 v1.4s, v20.8h, v20.8h
+ smlal v0.4s, v21.4h, v21.4h
+ smlal2 v1.4s, v21.8h, v21.8h
+ smlal v0.4s, v22.4h, v22.4h
+ smlal2 v1.4s, v22.8h, v22.8h
+ smlal v0.4s, v23.4h, v23.4h
+ smlal2 v1.4s, v23.8h, v23.8h
+.endr
+ cbnz w12, .Loop_ssd_ss_64
+ add v0.4s, v0.4s, v1.4s
+ ret_v0_w0
+endfunc
+
#else // HIGH_BIT_DEPTH
.macro SSE_PP_4x2
--
2.39.5 (Apple Git-154)
>From 338fa1e80701f5c1cc874e95ec90afc81a7af3e1 Mon Sep 17 00:00:00 2001
Message-Id:
<338fa1e80701f5c1cc874e95ec90afc81a7af3e1.1740155166.git.gerdazsejke.m...@arm.com>
In-Reply-To: <cover.1740155166.git.gerdazsejke.m...@arm.com>
References: <cover.1740155166.git.gerdazsejke.m...@arm.com>
From: Gerda Zsejke More <gerdazsejke.m...@arm.com>
Date: Tue, 4 Feb 2025 17:28:37 +0100
Subject: [PATCH v2 2/2] AArch64: Add Neon asm impl. for SBD ssd_s block size
64x64
Add a standard bit depth Neon asm implementation for pixel_ssd_s
for block size 64x64.
A high bit depth Neon asm implementation already exists just enable
it.
---
source/common/aarch64/asm-primitives.cpp | 2 ++
source/common/aarch64/ssd-a.S | 35 ++++++++++++++++++++++++
2 files changed, 37 insertions(+)
diff --git a/source/common/aarch64/asm-primitives.cpp
b/source/common/aarch64/asm-primitives.cpp
index c1f41cb3c..c1317eb74 100644
--- a/source/common/aarch64/asm-primitives.cpp
+++ b/source/common/aarch64/asm-primitives.cpp
@@ -557,11 +557,13 @@ void setupNeonPrimitives(EncoderPrimitives &p)
p.cu[BLOCK_8x8].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_8x8_neon);
p.cu[BLOCK_16x16].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_16x16_neon);
p.cu[BLOCK_32x32].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_32x32_neon);
+ p.cu[BLOCK_64x64].ssd_s[NONALIGNED] = PFX(pixel_ssd_s_64x64_neon);
p.cu[BLOCK_4x4].ssd_s[ALIGNED] = PFX(pixel_ssd_s_4x4_neon);
p.cu[BLOCK_8x8].ssd_s[ALIGNED] = PFX(pixel_ssd_s_8x8_neon);
p.cu[BLOCK_16x16].ssd_s[ALIGNED] = PFX(pixel_ssd_s_16x16_neon);
p.cu[BLOCK_32x32].ssd_s[ALIGNED] = PFX(pixel_ssd_s_32x32_neon);
+ p.cu[BLOCK_64x64].ssd_s[ALIGNED] = PFX(pixel_ssd_s_64x64_neon);
#if !HIGH_BIT_DEPTH
// pixel_avg_pp
diff --git a/source/common/aarch64/ssd-a.S b/source/common/aarch64/ssd-a.S
index f4d612137..e5da67679 100644
--- a/source/common/aarch64/ssd-a.S
+++ b/source/common/aarch64/ssd-a.S
@@ -406,6 +406,41 @@ function PFX(pixel_ssd_s_32x32_neon)
ret_v0_w0
endfunc
+function PFX(pixel_ssd_s_64x64_neon)
+ add x1, x1, x1
+ sub x1, x1, #64
+ sub x3, x3, #64
+
+ mov w12, #32
+ movi v0.16b, #0
+ movi v1.16b, #0
+.Loop_ssd_ss_64:
+ sub w12, w12, #1
+.rept 2
+ ld1 {v16.16b-v19.16b}, [x0], #64
+ ld1 {v20.16b-v23.16b}, [x0], x1
+ smlal v0.4s, v16.4h, v16.4h
+ smlal2 v1.4s, v16.8h, v16.8h
+ smlal v0.4s, v17.4h, v17.4h
+ smlal2 v1.4s, v17.8h, v17.8h
+ smlal v0.4s, v18.4h, v18.4h
+ smlal2 v1.4s, v18.8h, v18.8h
+ smlal v0.4s, v19.4h, v19.4h
+ smlal2 v1.4s, v19.8h, v19.8h
+ smlal v0.4s, v20.4h, v20.4h
+ smlal2 v1.4s, v20.8h, v20.8h
+ smlal v0.4s, v21.4h, v21.4h
+ smlal2 v1.4s, v21.8h, v21.8h
+ smlal v0.4s, v22.4h, v22.4h
+ smlal2 v1.4s, v22.8h, v22.8h
+ smlal v0.4s, v23.4h, v23.4h
+ smlal2 v1.4s, v23.8h, v23.8h
+.endr
+ cbnz w12, .Loop_ssd_ss_64
+ add v0.4s, v0.4s, v1.4s
+ ret_v0_w0
+endfunc
+
#else // HIGH_BIT_DEPTH
.macro SSE_PP_4x2
--
2.39.5 (Apple Git-154)
_______________________________________________
x265-devel mailing list
x265-devel@videolan.org
https://mailman.videolan.org/listinfo/x265-devel
