Add an implementation of luma_hvpp, using Neon I8MM implementation
for the horizontal part, and Armv8.0 Neon implementation for the
vertical part.
---
source/common/aarch64/filter-neon-i8mm.cpp | 22 +++++++++++++++++++++-
1 file changed, 21 insertions(+), 1 deletion(-)
diff --git a/source/common/aarch64/filter-neon-i8mm.cpp
b/source/common/aarch64/filter-neon-i8mm.cpp
index f8334016d..fb42d6672 100644
--- a/source/common/aarch64/filter-neon-i8mm.cpp
+++ b/source/common/aarch64/filter-neon-i8mm.cpp
@@ -755,9 +755,29 @@ void interp4_horiz_pp_i8mm(const uint8_t *src, intptr_t
srcStride, uint8_t *dst,
}
}
+// Declaration for use in interp_hv_pp_i8mm().
+template<int N, int width, int height>
+void interp_vert_sp_neon(const int16_t *src, intptr_t srcStride, uint8_t *dst,
+ intptr_t dstStride, int coeffIdx);
+
+// Implementation of luma_hvpp, using Neon i8mm implementation for the
+// horizontal part, and Armv8.0 Neon implementation for the vertical part.
+template<int width, int height>
+void interp_hv_pp_i8mm(const pixel *src, intptr_t srcStride, pixel *dst,
+ intptr_t dstStride, int idxX, int idxY)
+{
+ const int N_TAPS = 8;
+ ALIGN_VAR_32(int16_t, immed[width * (height + N_TAPS - 1)]);
+
+ interp8_horiz_ps_i8mm<width, height>(src, srcStride, immed, width, idxX,
1);
+ interp_vert_sp_neon<N_TAPS, width, height>(immed + (N_TAPS / 2 - 1) *
width,
+ width, dst, dstStride, idxY);
+}
+
#define LUMA_I8MM(W, H) \
p.pu[LUMA_ ## W ## x ## H].luma_hpp = interp8_horiz_pp_i8mm<W, H>; \
- p.pu[LUMA_ ## W ## x ## H].luma_hps = interp8_horiz_ps_i8mm<W, H>;
+ p.pu[LUMA_ ## W ## x ## H].luma_hps = interp8_horiz_ps_i8mm<W, H>; \
+ p.pu[LUMA_ ## W ## x ## H].luma_hvpp = interp_hv_pp_i8mm<W, H>;
#define CHROMA_420_I8MM(W, H) \
p.chroma[X265_CSP_I420].pu[CHROMA_420_ ## W ## x ## H].filter_hpp = \
--
2.42.1
>From 988b25411508117185a4d88bb519838d2c40fd0b Mon Sep 17 00:00:00 2001
Message-ID: <988b25411508117185a4d88bb519838d2c40fd0b.1725629250.git.hari.lim...@arm.com>
In-Reply-To: <cover.1725629250.git.hari.lim...@arm.com>
References: <cover.1725629250.git.hari.lim...@arm.com>
From: Hari Limaye <hari.lim...@arm.com>
Date: Thu, 16 May 2024 00:07:55 +0100
Subject: [PATCH 11/14] AArch64: Add Armv8.6 Neon I8MM implementation of
interp_hv_pp
Add an implementation of luma_hvpp, using Neon I8MM implementation
for the horizontal part, and Armv8.0 Neon implementation for the
vertical part.
---
source/common/aarch64/filter-neon-i8mm.cpp | 22 +++++++++++++++++++++-
1 file changed, 21 insertions(+), 1 deletion(-)
diff --git a/source/common/aarch64/filter-neon-i8mm.cpp b/source/common/aarch64/filter-neon-i8mm.cpp
index f8334016d..fb42d6672 100644
--- a/source/common/aarch64/filter-neon-i8mm.cpp
+++ b/source/common/aarch64/filter-neon-i8mm.cpp
@@ -755,9 +755,29 @@ void interp4_horiz_pp_i8mm(const uint8_t *src, intptr_t srcStride, uint8_t *dst,
}
}
+// Declaration for use in interp_hv_pp_i8mm().
+template<int N, int width, int height>
+void interp_vert_sp_neon(const int16_t *src, intptr_t srcStride, uint8_t *dst,
+ intptr_t dstStride, int coeffIdx);
+
+// Implementation of luma_hvpp, using Neon i8mm implementation for the
+// horizontal part, and Armv8.0 Neon implementation for the vertical part.
+template<int width, int height>
+void interp_hv_pp_i8mm(const pixel *src, intptr_t srcStride, pixel *dst,
+ intptr_t dstStride, int idxX, int idxY)
+{
+ const int N_TAPS = 8;
+ ALIGN_VAR_32(int16_t, immed[width * (height + N_TAPS - 1)]);
+
+ interp8_horiz_ps_i8mm<width, height>(src, srcStride, immed, width, idxX, 1);
+ interp_vert_sp_neon<N_TAPS, width, height>(immed + (N_TAPS / 2 - 1) * width,
+ width, dst, dstStride, idxY);
+}
+
#define LUMA_I8MM(W, H) \
p.pu[LUMA_ ## W ## x ## H].luma_hpp = interp8_horiz_pp_i8mm<W, H>; \
- p.pu[LUMA_ ## W ## x ## H].luma_hps = interp8_horiz_ps_i8mm<W, H>;
+ p.pu[LUMA_ ## W ## x ## H].luma_hps = interp8_horiz_ps_i8mm<W, H>; \
+ p.pu[LUMA_ ## W ## x ## H].luma_hvpp = interp_hv_pp_i8mm<W, H>;
#define CHROMA_420_I8MM(W, H) \
p.chroma[X265_CSP_I420].pu[CHROMA_420_ ## W ## x ## H].filter_hpp = \
--
2.42.1
_______________________________________________
x265-devel mailing list
x265-devel@videolan.org
https://mailman.videolan.org/listinfo/x265-devel
