Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
 help / color / mirror / Atom feed
* [FFmpeg-devel] [PATCH 1/2] checkasm: add hevc_deblock chroma test
@ 2023-03-21 18:35 J. Dekker
  2023-03-21 18:35 ` [FFmpeg-devel] [PATCH v2 2/2] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
  0 siblings, 1 reply; 9+ messages in thread
From: J. Dekker @ 2023-03-21 18:35 UTC (permalink / raw)
  To: ffmpeg-devel

Signed-off-by: J. Dekker <jdek@itanimul.li>
---

 Only the chroma part, writing the luma part alongside the NEON asm.

 tests/checkasm/Makefile       |  2 +-
 tests/checkasm/checkasm.c     |  1 +
 tests/checkasm/checkasm.h     |  1 +
 tests/checkasm/hevc_deblock.c | 87 +++++++++++++++++++++++++++++++++++
 4 files changed, 90 insertions(+), 1 deletion(-)
 create mode 100644 tests/checkasm/hevc_deblock.c

diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index a6f06c7007..3e62a22bd6 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -28,7 +28,7 @@ AVCODECOBJS-$(CONFIG_HUFFYUV_DECODER)   += huffyuvdsp.o
 AVCODECOBJS-$(CONFIG_JPEG2000_DECODER)  += jpeg2000dsp.o
 AVCODECOBJS-$(CONFIG_OPUS_DECODER)      += opusdsp.o
 AVCODECOBJS-$(CONFIG_PIXBLOCKDSP)       += pixblockdsp.o
-AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_idct.o hevc_sao.o hevc_pel.o
+AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_deblock.o hevc_idct.o hevc_sao.o hevc_pel.o
 AVCODECOBJS-$(CONFIG_UTVIDEO_DECODER)   += utvideodsp.o
 AVCODECOBJS-$(CONFIG_V210_DECODER)      += v210dec.o
 AVCODECOBJS-$(CONFIG_V210_ENCODER)      += v210enc.o
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index e96d84a7da..c2184d260d 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -116,6 +116,7 @@ static const struct {
     #endif
     #if CONFIG_HEVC_DECODER
         { "hevc_add_res", checkasm_check_hevc_add_res },
+        { "hevc_deblock", checkasm_check_hevc_deblock },
         { "hevc_idct", checkasm_check_hevc_idct },
         { "hevc_pel", checkasm_check_hevc_pel },
         { "hevc_sao", checkasm_check_hevc_sao },
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index 8744a81218..89c643e6a0 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -60,6 +60,7 @@ void checkasm_check_h264dsp(void);
 void checkasm_check_h264pred(void);
 void checkasm_check_h264qpel(void);
 void checkasm_check_hevc_add_res(void);
+void checkasm_check_hevc_deblock(void);
 void checkasm_check_hevc_idct(void);
 void checkasm_check_hevc_pel(void);
 void checkasm_check_hevc_sao(void);
diff --git a/tests/checkasm/hevc_deblock.c b/tests/checkasm/hevc_deblock.c
new file mode 100644
index 0000000000..06c9c1969e
--- /dev/null
+++ b/tests/checkasm/hevc_deblock.c
@@ -0,0 +1,87 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <string.h>
+
+#include "libavutil/intreadwrite.h"
+#include "libavutil/mem_internal.h"
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/hevcdsp.h"
+
+#include "checkasm.h"
+
+static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
+
+#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
+#define BUF_STRIDE (8 * 2)
+#define BUF_LINES (8)
+#define BUF_OFFSET (BUF_STRIDE * BUF_LINES)
+#define BUF_SIZE (BUF_STRIDE * BUF_LINES + BUF_OFFSET * 2)
+
+#define randomize_buffers(buf0, buf1, size)                 \
+    do {                                                    \
+        uint32_t mask = pixel_mask[(bit_depth - 8) >> 1];   \
+        int k;                                              \
+        for (k = 0; k < size; k += 4) {                     \
+            uint32_t r = rnd() & mask;                      \
+            AV_WN32A(buf0 + k, r);                          \
+            AV_WN32A(buf1 + k, r);                          \
+        }                                                   \
+    } while (0)
+
+static void check_deblock_chroma(HEVCDSPContext h, int bit_depth)
+{
+    int32_t tc[2] = { 1, 1 };
+    uint8_t no_p[2] = { 0, 0 };
+    uint8_t no_q[2] = { 0, 0 };
+    LOCAL_ALIGNED_32(uint8_t, buf0, [BUF_SIZE]);
+    LOCAL_ALIGNED_32(uint8_t, buf1, [BUF_SIZE]);
+
+    declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, ptrdiff_t stride, int32_t *tc, uint8_t *no_p, uint8_t *no_q);
+
+    randomize_buffers(buf0, buf1, BUF_SIZE);
+    if (check_func(h.hevc_h_loop_filter_chroma, "hevc_h_loop_filter_chroma_%d", bit_depth)) {
+        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        if (memcmp(buf0, buf1, BUF_SIZE))
+            fail();
+        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+    }
+
+    randomize_buffers(buf0, buf1, BUF_SIZE);
+    if (check_func(h.hevc_v_loop_filter_chroma, "hevc_v_loop_filter_chroma_%d", bit_depth)) {
+        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        if (memcmp(buf0, buf1, BUF_SIZE))
+            fail();
+        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+    }
+}
+
+void checkasm_check_hevc_deblock(void)
+{
+    int bit_depth;
+
+    for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
+        HEVCDSPContext h;
+        ff_hevc_dsp_init(&h, bit_depth);
+        check_deblock_chroma(h, bit_depth);
+    }
+    report("chroma");
+}
-- 
2.37.1 (Apple Git-137.1)

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* [FFmpeg-devel] [PATCH v2 2/2] lavc/aarch64: add hevc deblock chroma 8-12bit
  2023-03-21 18:35 [FFmpeg-devel] [PATCH 1/2] checkasm: add hevc_deblock chroma test J. Dekker
@ 2023-03-21 18:35 ` J. Dekker
  2023-03-21 20:30   ` Martin Storsjö
  0 siblings, 1 reply; 9+ messages in thread
From: J. Dekker @ 2023-03-21 18:35 UTC (permalink / raw)
  To: ffmpeg-devel

Signed-off-by: J. Dekker <jdek@itanimul.li>
---

 Implements comments from Martin, split body into its own function,
 interleaved stores, however `.elif' is not valid on macOS so left
 that out.

 libavcodec/aarch64/Makefile               |   3 +-
 libavcodec/aarch64/hevcdsp_deblock_neon.S | 187 ++++++++++++++++++++++
 libavcodec/aarch64/hevcdsp_init_aarch64.c |  18 +++
 3 files changed, 207 insertions(+), 1 deletion(-)
 create mode 100644 libavcodec/aarch64/hevcdsp_deblock_neon.S

diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
index 02fb51c3ab..216191640c 100644
--- a/libavcodec/aarch64/Makefile
+++ b/libavcodec/aarch64/Makefile
@@ -65,7 +65,8 @@ NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_16bpp_neon.o       \
                                            aarch64/vp9lpf_neon.o               \
                                            aarch64/vp9mc_16bpp_neon.o          \
                                            aarch64/vp9mc_neon.o
-NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_idct_neon.o         \
+NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_deblock_neon.o      \
+                                           aarch64/hevcdsp_idct_neon.o         \
                                            aarch64/hevcdsp_init_aarch64.o      \
                                            aarch64/hevcdsp_qpel_neon.o         \
                                            aarch64/hevcdsp_sao_neon.o
diff --git a/libavcodec/aarch64/hevcdsp_deblock_neon.S b/libavcodec/aarch64/hevcdsp_deblock_neon.S
new file mode 100644
index 0000000000..0f0e829fb5
--- /dev/null
+++ b/libavcodec/aarch64/hevcdsp_deblock_neon.S
@@ -0,0 +1,187 @@
+/* -*-arm64-*-
+ * vim: syntax=arm64asm
+ *
+ * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
+ * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+
+#include "libavutil/aarch64/asm.S"
+#include "neon.S"
+
+.macro hevc_loop_filter_chroma_start bitdepth
+        mov             x4, x30
+        ldr             w14, [x2]
+        ldr             w15, [x2, #4]
+.if \bitdepth > 8
+        lsl             w14, w14, #(\bitdepth - 8)
+        lsl             w15, w15, #(\bitdepth - 8)
+.endif
+        adds            w2, w14, w15
+        b.eq            1f
+        dup             v16.4h, w14
+        dup             v17.4h, w15
+        trn1            v16.2d, v16.2d, v17.2d
+.if \bitdepth == 10
+        mvni            v19.8h, #0xFC, lsl #8 // movi #0x03FF
+.endif
+.if \bitdepth == 12
+        mvni            v19.8h, #0xF0, lsl #8 // movi #0x0FFF
+.endif
+.if \bitdepth > 8
+        movi            v18.8h, #0
+.endif
+        neg             v17.8h, v16.8h
+.endm
+
+.macro hevc_loop_filter_chroma_body bitdepth
+.if \bitdepth <= 8
+        uxtl            v0.8h, v0.8b // p1
+        uxtl            v1.8h, v1.8b // p0
+        uxtl            v2.8h, v2.8b // q0
+        uxtl            v3.8h, v3.8b // q1
+.endif
+        sub             v5.8h, v2.8h, v1.8h // q0 - p0
+        sub             v6.8h, v0.8h, v3.8h // p1 - q1
+        shl             v5.8h, v5.8h, #2
+        add             v5.8h, v6.8h, v5.8h
+        srshr           v5.8h, v5.8h, #3
+        smin            v5.8h, v5.8h, v16.8h
+        smax            v5.8h, v5.8h, v17.8h
+        sqadd           v1.8h, v1.8h, v5.8h // p0 + delta
+        sqsub           v2.8h, v2.8h, v5.8h // q0 - delta
+.if \bitdepth <= 8
+        sqxtun          v1.8b, v1.8h
+        sqxtun          v2.8b, v2.8h
+.else
+        smin            v1.8h, v1.8h, v19.8h
+        smin            v2.8h, v2.8h, v19.8h
+        smax            v1.8h, v1.8h, v18.8h
+        smax            v2.8h, v2.8h, v18.8h
+.endif
+.endm
+
+function hevc_loop_filter_chroma_body_8_neon, export=0
+        hevc_loop_filter_chroma_body 8
+        ret
+endfunc
+
+function hevc_loop_filter_chroma_body_10_neon, export=0
+hevc_loop_filter_chroma_body_12_neon:
+        hevc_loop_filter_chroma_body 10
+        ret
+endfunc
+
+// void ff_hevc_h_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
+
+.macro hevc_h_loop_filter_chroma bitdepth
+function ff_hevc_h_loop_filter_chroma_\bitdepth\()_neon, export=1
+        hevc_loop_filter_chroma_start \bitdepth
+        sub             x0, x0, x1, lsl #1
+.if \bitdepth > 8
+        ld1             {v0.8h}, [x0], x1
+        ld1             {v1.8h}, [x0], x1
+        ld1             {v2.8h}, [x0], x1
+        ld1             {v3.8h}, [x0]
+.else
+        ld1             {v0.8b}, [x0], x1
+        ld1             {v1.8b}, [x0], x1
+        ld1             {v2.8b}, [x0], x1
+        ld1             {v3.8b}, [x0]
+.endif
+        sub             x0, x0, x1, lsl #1
+        bl              hevc_loop_filter_chroma_body_\bitdepth\()_neon
+.if \bitdepth > 8
+        st1             {v1.8h}, [x0], x1
+        st1             {v2.8h}, [x0]
+.else
+        st1             {v1.8b}, [x0], x1
+        st1             {v2.8b}, [x0]
+.endif
+1:      ret             x4
+endfunc
+.endm
+
+.macro hevc_v_loop_filter_chroma bitdepth
+function ff_hevc_v_loop_filter_chroma_\bitdepth\()_neon, export=1
+        hevc_loop_filter_chroma_start \bitdepth
+.if \bitdepth > 8
+        sub             x0, x0, #8
+        add             x3, x0, x1
+        lsl             x1, x1, #1
+        ld1             {v20.8h}, [x0], x1
+        ld1             {v21.8h}, [x3], x1
+        ld1             {v0.8h},  [x0], x1
+        ld1             {v1.8h},  [x3], x1
+        ld1             {v2.8h},  [x0], x1
+        ld1             {v3.8h},  [x3], x1
+        ld1             {v22.8h}, [x0], x1
+        ld1             {v23.8h}, [x3], x1
+        transpose_8x8H  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+.else
+        sub             x0, x0, #4
+        add             x3, x0, x1
+        lsl             x1, x1, #1
+        ld1             {v20.8b}, [x0], x1
+        ld1             {v21.8b}, [x3], x1
+        ld1             {v0.8b},  [x0], x1
+        ld1             {v1.8b},  [x3], x1
+        ld1             {v2.8b},  [x0], x1
+        ld1             {v3.8b},  [x3], x1
+        ld1             {v22.8b}, [x0], x1
+        ld1             {v23.8b}, [x3], x1
+        transpose_8x8B  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+.endif
+        sub             x0, x0, x1, lsl #2
+        sub             x3, x3, x1, lsl #2
+        bl              hevc_loop_filter_chroma_body_\bitdepth\()_neon
+.if \bitdepth > 8
+        transpose_8x8H  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+        st1             {v20.8h}, [x0], x1
+        st1             {v21.8h}, [x3], x1
+        st1             {v0.8h},  [x0], x1
+        st1             {v1.8h},  [x3], x1
+        st1             {v2.8h},  [x0], x1
+        st1             {v3.8h},  [x3], x1
+        st1             {v22.8h}, [x0], x1
+        st1             {v23.8h}, [x3]
+.else
+        xtn             v0.8b, v0.8h // restore
+        xtn             v3.8b, v3.8h
+        transpose_8x8B  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+        st1             {v20.8b}, [x0], x1
+        st1             {v21.8b}, [x3], x1
+        st1             {v0.8b},  [x0], x1
+        st1             {v1.8b},  [x3], x1
+        st1             {v2.8b},  [x0], x1
+        st1             {v3.8b},  [x3], x1
+        st1             {v22.8b}, [x0], x1
+        st1             {v23.8b}, [x3]
+.endif
+1:      ret             x4
+endfunc
+.endm
+
+hevc_h_loop_filter_chroma 8
+hevc_h_loop_filter_chroma 10
+hevc_h_loop_filter_chroma 12
+
+hevc_v_loop_filter_chroma 8
+hevc_v_loop_filter_chroma 10
+hevc_v_loop_filter_chroma 12
diff --git a/libavcodec/aarch64/hevcdsp_init_aarch64.c b/libavcodec/aarch64/hevcdsp_init_aarch64.c
index 1deefca0a2..a923bae35c 100644
--- a/libavcodec/aarch64/hevcdsp_init_aarch64.c
+++ b/libavcodec/aarch64/hevcdsp_init_aarch64.c
@@ -25,6 +25,18 @@
 #include "libavutil/aarch64/cpu.h"
 #include "libavcodec/hevcdsp.h"
 
+void ff_hevc_v_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                         const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_v_loop_filter_chroma_10_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_v_loop_filter_chroma_12_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                         const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_10_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_12_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
 void ff_hevc_add_residual_4x4_8_neon(uint8_t *_dst, const int16_t *coeffs,
                                      ptrdiff_t stride);
 void ff_hevc_add_residual_4x4_10_neon(uint8_t *_dst, const int16_t *coeffs,
@@ -117,6 +129,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
     if (!have_neon(av_get_cpu_flags())) return;
 
     if (bit_depth == 8) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_8_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_8_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_8_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_8_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_8_neon;
@@ -167,6 +181,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->put_hevc_qpel_bi[9][0][1]   = ff_hevc_put_hevc_qpel_bi_h16_8_neon;
     }
     if (bit_depth == 10) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_10_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_10_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_10_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_10_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_10_neon;
@@ -180,6 +196,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->idct_dc[3]                  = ff_hevc_idct_32x32_dc_10_neon;
     }
     if (bit_depth == 12) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_12_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_12_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_12_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_12_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_12_neon;
-- 
2.37.1 (Apple Git-137.1)

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [FFmpeg-devel] [PATCH v2 2/2] lavc/aarch64: add hevc deblock chroma 8-12bit
  2023-03-21 18:35 ` [FFmpeg-devel] [PATCH v2 2/2] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
@ 2023-03-21 20:30   ` Martin Storsjö
  2023-03-22  0:07     ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro J. Dekker
  0 siblings, 1 reply; 9+ messages in thread
From: Martin Storsjö @ 2023-03-21 20:30 UTC (permalink / raw)
  To: FFmpeg development discussions and patches

On Tue, 21 Mar 2023, J. Dekker wrote:

> Signed-off-by: J. Dekker <jdek@itanimul.li>
> ---
>
> Implements comments from Martin, split body into its own function,
> interleaved stores, however `.elif' is not valid on macOS so left
> that out.

I didn't review the patches themselves right now yet, but noticed this 
comment - there's no ".elif", it's ".elseif".

// Martin

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro
  2023-03-21 20:30   ` Martin Storsjö
@ 2023-03-22  0:07     ` J. Dekker
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test J. Dekker
                         ` (2 more replies)
  0 siblings, 3 replies; 9+ messages in thread
From: J. Dekker @ 2023-03-22  0:07 UTC (permalink / raw)
  To: ffmpeg-devel

Signed-off-by: J. Dekker <jdek@itanimul.li>
---
 libavcodec/aarch64/hevcdsp_idct_neon.S | 19 +++++--------------
 libavcodec/aarch64/neon.S              | 11 +++++++++++
 2 files changed, 16 insertions(+), 14 deletions(-)

diff --git a/libavcodec/aarch64/hevcdsp_idct_neon.S b/libavcodec/aarch64/hevcdsp_idct_neon.S
index 467cb0f48a..3e59dd20bb 100644
--- a/libavcodec/aarch64/hevcdsp_idct_neon.S
+++ b/libavcodec/aarch64/hevcdsp_idct_neon.S
@@ -5,7 +5,7 @@
  *
  * Ported from arm/hevcdsp_idct_neon.S by
  * Copyright (c) 2020 Reimar Döffinger
- * Copyright (c) 2020 J. Dekker
+ * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
  *
  * This file is part of FFmpeg.
  *
@@ -38,13 +38,6 @@ const trans, align=4
         .short          31, 22, 13, 4
 endconst
 
-.macro clip2 in1, in2, min, max
-        smax            \in1, \in1, \min
-        smax            \in2, \in2, \min
-        smin            \in1, \in1, \max
-        smin            \in2, \in2, \max
-.endm
-
 function ff_hevc_add_residual_4x4_8_neon, export=1
         ld1             {v0.8h-v1.8h}, [x1]
         ld1             {v2.s}[0], [x0], x2
@@ -182,7 +175,7 @@ function hevc_add_residual_4x4_16_neon, export=0
         ld1             {v3.d}[1], [x12], x2
         movi            v4.8h, #0
         sqadd           v1.8h, v1.8h, v3.8h
-        clip2           v0.8h, v1.8h, v4.8h, v21.8h
+        clip            v4.8h, v21.8h, v0.8h, v1.8h
         st1             {v0.d}[0], [x0],  x2
         st1             {v0.d}[1], [x0],  x2
         st1             {v1.d}[0], [x0],  x2
@@ -201,7 +194,7 @@ function hevc_add_residual_8x8_16_neon, export=0
         sqadd           v0.8h, v0.8h, v2.8h
         ld1             {v3.8h}, [x12]
         sqadd           v1.8h, v1.8h, v3.8h
-        clip2           v0.8h, v1.8h, v4.8h, v21.8h
+        clip            v4.8h, v21.8h, v0.8h, v1.8h
         st1             {v0.8h}, [x0],  x2
         st1             {v1.8h}, [x12], x2
         bne             1b
@@ -221,8 +214,7 @@ function hevc_add_residual_16x16_16_neon, export=0
         sqadd           v1.8h, v1.8h, v17.8h
         sqadd           v2.8h, v2.8h, v18.8h
         sqadd           v3.8h, v3.8h, v19.8h
-        clip2           v0.8h, v1.8h, v20.8h, v21.8h
-        clip2           v2.8h, v3.8h, v20.8h, v21.8h
+        clip            v20.8h, v21.8h, v0.8h, v1.8h, v2.8h, v3.8h
         st1             {v0.8h-v1.8h}, [x0],  x2
         st1             {v2.8h-v3.8h}, [x12], x2
         bne             1b
@@ -239,8 +231,7 @@ function hevc_add_residual_32x32_16_neon, export=0
         sqadd           v1.8h, v1.8h, v17.8h
         sqadd           v2.8h, v2.8h, v18.8h
         sqadd           v3.8h, v3.8h, v19.8h
-        clip2           v0.8h, v1.8h, v20.8h, v21.8h
-        clip2           v2.8h, v3.8h, v20.8h, v21.8h
+        clip            v20.8h, v21.8h, v0.8h, v1.8h, v2.8h, v3.8h
         st1             {v0.8h-v3.8h}, [x0], x2
         bne             1b
         ret
diff --git a/libavcodec/aarch64/neon.S b/libavcodec/aarch64/neon.S
index 1ad32c359d..bc105e4861 100644
--- a/libavcodec/aarch64/neon.S
+++ b/libavcodec/aarch64/neon.S
@@ -1,6 +1,8 @@
 /*
  * This file is part of FFmpeg.
  *
+ * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
+ *
  * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
@@ -16,6 +18,15 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+.macro clip min, max, regs:vararg
+.irp x, \regs
+        smax            \x, \x, \min
+.endr
+.irp x, \regs
+        smin            \x, \x, \max
+.endr
+.endm
+
 .macro  transpose_8x8B  r0, r1, r2, r3, r4, r5, r6, r7, r8, r9
         trn1            \r8\().8B,  \r0\().8B,  \r1\().8B
         trn2            \r9\().8B,  \r0\().8B,  \r1\().8B
-- 
2.39.2

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test
  2023-03-22  0:07     ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro J. Dekker
@ 2023-03-22  0:07       ` J. Dekker
  2023-03-22  9:04         ` Martin Storsjö
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
  2023-03-22  8:38       ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro Martin Storsjö
  2 siblings, 1 reply; 9+ messages in thread
From: J. Dekker @ 2023-03-22  0:07 UTC (permalink / raw)
  To: ffmpeg-devel

Signed-off-by: J. Dekker <jdek@itanimul.li>
---

 Added missing FATE target.

 tests/checkasm/Makefile       |  2 +-
 tests/checkasm/checkasm.c     |  1 +
 tests/checkasm/checkasm.h     |  1 +
 tests/checkasm/hevc_deblock.c | 87 +++++++++++++++++++++++++++++++++++
 tests/fate/checkasm.mak       |  1 +
 5 files changed, 91 insertions(+), 1 deletion(-)
 create mode 100644 tests/checkasm/hevc_deblock.c

diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index a6f06c7007..3e62a22bd6 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -28,7 +28,7 @@ AVCODECOBJS-$(CONFIG_HUFFYUV_DECODER)   += huffyuvdsp.o
 AVCODECOBJS-$(CONFIG_JPEG2000_DECODER)  += jpeg2000dsp.o
 AVCODECOBJS-$(CONFIG_OPUS_DECODER)      += opusdsp.o
 AVCODECOBJS-$(CONFIG_PIXBLOCKDSP)       += pixblockdsp.o
-AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_idct.o hevc_sao.o hevc_pel.o
+AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_deblock.o hevc_idct.o hevc_sao.o hevc_pel.o
 AVCODECOBJS-$(CONFIG_UTVIDEO_DECODER)   += utvideodsp.o
 AVCODECOBJS-$(CONFIG_V210_DECODER)      += v210dec.o
 AVCODECOBJS-$(CONFIG_V210_ENCODER)      += v210enc.o
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index e96d84a7da..c2184d260d 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -116,6 +116,7 @@ static const struct {
     #endif
     #if CONFIG_HEVC_DECODER
         { "hevc_add_res", checkasm_check_hevc_add_res },
+        { "hevc_deblock", checkasm_check_hevc_deblock },
         { "hevc_idct", checkasm_check_hevc_idct },
         { "hevc_pel", checkasm_check_hevc_pel },
         { "hevc_sao", checkasm_check_hevc_sao },
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index 8744a81218..89c643e6a0 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -60,6 +60,7 @@ void checkasm_check_h264dsp(void);
 void checkasm_check_h264pred(void);
 void checkasm_check_h264qpel(void);
 void checkasm_check_hevc_add_res(void);
+void checkasm_check_hevc_deblock(void);
 void checkasm_check_hevc_idct(void);
 void checkasm_check_hevc_pel(void);
 void checkasm_check_hevc_sao(void);
diff --git a/tests/checkasm/hevc_deblock.c b/tests/checkasm/hevc_deblock.c
new file mode 100644
index 0000000000..06c9c1969e
--- /dev/null
+++ b/tests/checkasm/hevc_deblock.c
@@ -0,0 +1,87 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <string.h>
+
+#include "libavutil/intreadwrite.h"
+#include "libavutil/mem_internal.h"
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/hevcdsp.h"
+
+#include "checkasm.h"
+
+static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
+
+#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
+#define BUF_STRIDE (8 * 2)
+#define BUF_LINES (8)
+#define BUF_OFFSET (BUF_STRIDE * BUF_LINES)
+#define BUF_SIZE (BUF_STRIDE * BUF_LINES + BUF_OFFSET * 2)
+
+#define randomize_buffers(buf0, buf1, size)                 \
+    do {                                                    \
+        uint32_t mask = pixel_mask[(bit_depth - 8) >> 1];   \
+        int k;                                              \
+        for (k = 0; k < size; k += 4) {                     \
+            uint32_t r = rnd() & mask;                      \
+            AV_WN32A(buf0 + k, r);                          \
+            AV_WN32A(buf1 + k, r);                          \
+        }                                                   \
+    } while (0)
+
+static void check_deblock_chroma(HEVCDSPContext h, int bit_depth)
+{
+    int32_t tc[2] = { 1, 1 };
+    uint8_t no_p[2] = { 0, 0 };
+    uint8_t no_q[2] = { 0, 0 };
+    LOCAL_ALIGNED_32(uint8_t, buf0, [BUF_SIZE]);
+    LOCAL_ALIGNED_32(uint8_t, buf1, [BUF_SIZE]);
+
+    declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, ptrdiff_t stride, int32_t *tc, uint8_t *no_p, uint8_t *no_q);
+
+    randomize_buffers(buf0, buf1, BUF_SIZE);
+    if (check_func(h.hevc_h_loop_filter_chroma, "hevc_h_loop_filter_chroma_%d", bit_depth)) {
+        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        if (memcmp(buf0, buf1, BUF_SIZE))
+            fail();
+        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+    }
+
+    randomize_buffers(buf0, buf1, BUF_SIZE);
+    if (check_func(h.hevc_v_loop_filter_chroma, "hevc_v_loop_filter_chroma_%d", bit_depth)) {
+        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+        if (memcmp(buf0, buf1, BUF_SIZE))
+            fail();
+        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
+    }
+}
+
+void checkasm_check_hevc_deblock(void)
+{
+    int bit_depth;
+
+    for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
+        HEVCDSPContext h;
+        ff_hevc_dsp_init(&h, bit_depth);
+        check_deblock_chroma(h, bit_depth);
+    }
+    report("chroma");
+}
diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak
index a4e95541f5..faac764388 100644
--- a/tests/fate/checkasm.mak
+++ b/tests/fate/checkasm.mak
@@ -15,6 +15,7 @@ FATE_CHECKASM = fate-checkasm-aacpsdsp                                  \
                 fate-checkasm-h264pred                                  \
                 fate-checkasm-h264qpel                                  \
                 fate-checkasm-hevc_add_res                              \
+                fate-checkasm-hevc_deblock                              \
                 fate-checkasm-hevc_idct                                 \
                 fate-checkasm-hevc_pel                                  \
                 fate-checkasm-hevc_sao                                  \
-- 
2.39.2

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit
  2023-03-22  0:07     ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro J. Dekker
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test J. Dekker
@ 2023-03-22  0:07       ` J. Dekker
  2023-03-22  9:26         ` Martin Storsjö
  2023-03-22  8:38       ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro Martin Storsjö
  2 siblings, 1 reply; 9+ messages in thread
From: J. Dekker @ 2023-03-22  0:07 UTC (permalink / raw)
  To: ffmpeg-devel

Signed-off-by: J. Dekker <jdek@itanimul.li>
---

 - Using clip macro
 - Avoided need for .elseif at all, just used better mvni

 libavcodec/aarch64/Makefile               |   3 +-
 libavcodec/aarch64/hevcdsp_deblock_neon.S | 180 ++++++++++++++++++++++
 libavcodec/aarch64/hevcdsp_init_aarch64.c |  18 +++
 3 files changed, 200 insertions(+), 1 deletion(-)
 create mode 100644 libavcodec/aarch64/hevcdsp_deblock_neon.S

diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
index 02fb51c3ab..216191640c 100644
--- a/libavcodec/aarch64/Makefile
+++ b/libavcodec/aarch64/Makefile
@@ -65,7 +65,8 @@ NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_16bpp_neon.o       \
                                            aarch64/vp9lpf_neon.o               \
                                            aarch64/vp9mc_16bpp_neon.o          \
                                            aarch64/vp9mc_neon.o
-NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_idct_neon.o         \
+NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_deblock_neon.o      \
+                                           aarch64/hevcdsp_idct_neon.o         \
                                            aarch64/hevcdsp_init_aarch64.o      \
                                            aarch64/hevcdsp_qpel_neon.o         \
                                            aarch64/hevcdsp_sao_neon.o
diff --git a/libavcodec/aarch64/hevcdsp_deblock_neon.S b/libavcodec/aarch64/hevcdsp_deblock_neon.S
new file mode 100644
index 0000000000..49b40f21c8
--- /dev/null
+++ b/libavcodec/aarch64/hevcdsp_deblock_neon.S
@@ -0,0 +1,180 @@
+/* -*-arm64-*-
+ * vim: syntax=arm64asm
+ *
+ * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
+ * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+
+#include "libavutil/aarch64/asm.S"
+#include "neon.S"
+
+.macro hevc_loop_filter_chroma_start bitdepth
+        mov             x4, x30
+        ldr             w14, [x2]
+        ldr             w15, [x2, #4]
+.if \bitdepth > 8
+        lsl             w14, w14, #(\bitdepth - 8)
+        lsl             w15, w15, #(\bitdepth - 8)
+.endif
+        adds            w2, w14, w15
+        b.eq            1f
+        dup             v16.4h, w14
+        dup             v17.4h, w15
+        trn1            v16.2d, v16.2d, v17.2d
+.if \bitdepth > 8
+        mvni            v19.8h, #((0xff << (\bitdepth - 8)) & 0xff), lsl #8
+        movi            v18.8h, #0
+.endif
+        neg             v17.8h, v16.8h
+.endm
+
+.macro hevc_loop_filter_chroma_body bitdepth
+.if \bitdepth <= 8
+        uxtl            v0.8h, v0.8b // p1
+        uxtl            v1.8h, v1.8b // p0
+        uxtl            v2.8h, v2.8b // q0
+        uxtl            v3.8h, v3.8b // q1
+.endif
+        sub             v5.8h, v2.8h, v1.8h // q0 - p0
+        sub             v6.8h, v0.8h, v3.8h // p1 - q1
+        shl             v5.8h, v5.8h, #2
+        add             v5.8h, v6.8h, v5.8h
+        srshr           v5.8h, v5.8h, #3
+        clip            v17.8h, v16.8h, v5.8h
+        sqadd           v1.8h, v1.8h, v5.8h // p0 + delta
+        sqsub           v2.8h, v2.8h, v5.8h // q0 - delta
+.if \bitdepth <= 8
+        sqxtun          v1.8b, v1.8h
+        sqxtun          v2.8b, v2.8h
+.else
+        smin            v1.8h, v1.8h, v19.8h
+        smin            v2.8h, v2.8h, v19.8h
+        clip            v18.8h, v19.8h, v1.8h, v2.8h
+.endif
+.endm
+
+function hevc_loop_filter_chroma_body_8_neon, export=0
+        hevc_loop_filter_chroma_body 8
+        ret
+endfunc
+
+function hevc_loop_filter_chroma_body_10_neon, export=0
+hevc_loop_filter_chroma_body_12_neon:
+        hevc_loop_filter_chroma_body 10
+        ret
+endfunc
+
+// void ff_hevc_h_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
+
+.macro hevc_h_loop_filter_chroma bitdepth
+function ff_hevc_h_loop_filter_chroma_\bitdepth\()_neon, export=1
+        hevc_loop_filter_chroma_start \bitdepth
+        sub             x0, x0, x1, lsl #1
+.if \bitdepth > 8
+        ld1             {v0.8h}, [x0], x1
+        ld1             {v1.8h}, [x0], x1
+        ld1             {v2.8h}, [x0], x1
+        ld1             {v3.8h}, [x0]
+.else
+        ld1             {v0.8b}, [x0], x1
+        ld1             {v1.8b}, [x0], x1
+        ld1             {v2.8b}, [x0], x1
+        ld1             {v3.8b}, [x0]
+.endif
+        sub             x0, x0, x1, lsl #1
+        bl              hevc_loop_filter_chroma_body_\bitdepth\()_neon
+.if \bitdepth > 8
+        st1             {v1.8h}, [x0], x1
+        st1             {v2.8h}, [x0]
+.else
+        st1             {v1.8b}, [x0], x1
+        st1             {v2.8b}, [x0]
+.endif
+1:      ret             x4
+endfunc
+.endm
+
+.macro hevc_v_loop_filter_chroma bitdepth
+function ff_hevc_v_loop_filter_chroma_\bitdepth\()_neon, export=1
+        hevc_loop_filter_chroma_start \bitdepth
+.if \bitdepth > 8
+        sub             x0, x0, #8
+        add             x3, x0, x1
+        lsl             x1, x1, #1
+        ld1             {v20.8h}, [x0], x1
+        ld1             {v21.8h}, [x3], x1
+        ld1             {v0.8h},  [x0], x1
+        ld1             {v1.8h},  [x3], x1
+        ld1             {v2.8h},  [x0], x1
+        ld1             {v3.8h},  [x3], x1
+        ld1             {v22.8h}, [x0], x1
+        ld1             {v23.8h}, [x3], x1
+        transpose_8x8H  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+.else
+        sub             x0, x0, #4
+        add             x3, x0, x1
+        lsl             x1, x1, #1
+        ld1             {v20.8b}, [x0], x1
+        ld1             {v21.8b}, [x3], x1
+        ld1             {v0.8b},  [x0], x1
+        ld1             {v1.8b},  [x3], x1
+        ld1             {v2.8b},  [x0], x1
+        ld1             {v3.8b},  [x3], x1
+        ld1             {v22.8b}, [x0], x1
+        ld1             {v23.8b}, [x3], x1
+        transpose_8x8B  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+.endif
+        sub             x0, x0, x1, lsl #2
+        sub             x3, x3, x1, lsl #2
+        bl              hevc_loop_filter_chroma_body_\bitdepth\()_neon
+.if \bitdepth > 8
+        transpose_8x8H  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+        st1             {v20.8h}, [x0], x1
+        st1             {v21.8h}, [x3], x1
+        st1             {v0.8h},  [x0], x1
+        st1             {v1.8h},  [x3], x1
+        st1             {v2.8h},  [x0], x1
+        st1             {v3.8h},  [x3], x1
+        st1             {v22.8h}, [x0], x1
+        st1             {v23.8h}, [x3]
+.else
+        xtn             v0.8b, v0.8h // restore
+        xtn             v3.8b, v3.8h
+        transpose_8x8B  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25
+        st1             {v20.8b}, [x0], x1
+        st1             {v21.8b}, [x3], x1
+        st1             {v0.8b},  [x0], x1
+        st1             {v1.8b},  [x3], x1
+        st1             {v2.8b},  [x0], x1
+        st1             {v3.8b},  [x3], x1
+        st1             {v22.8b}, [x0], x1
+        st1             {v23.8b}, [x3]
+.endif
+1:      ret             x4
+endfunc
+.endm
+
+hevc_h_loop_filter_chroma 8
+hevc_h_loop_filter_chroma 10
+hevc_h_loop_filter_chroma 12
+
+hevc_v_loop_filter_chroma 8
+hevc_v_loop_filter_chroma 10
+hevc_v_loop_filter_chroma 12
diff --git a/libavcodec/aarch64/hevcdsp_init_aarch64.c b/libavcodec/aarch64/hevcdsp_init_aarch64.c
index 1deefca0a2..a923bae35c 100644
--- a/libavcodec/aarch64/hevcdsp_init_aarch64.c
+++ b/libavcodec/aarch64/hevcdsp_init_aarch64.c
@@ -25,6 +25,18 @@
 #include "libavutil/aarch64/cpu.h"
 #include "libavcodec/hevcdsp.h"
 
+void ff_hevc_v_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                         const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_v_loop_filter_chroma_10_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_v_loop_filter_chroma_12_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                         const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_10_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
+void ff_hevc_h_loop_filter_chroma_12_neon(uint8_t *_pix, ptrdiff_t _stride,
+                                          const int *_tc, const uint8_t *_no_p, const uint8_t *_no_q);
 void ff_hevc_add_residual_4x4_8_neon(uint8_t *_dst, const int16_t *coeffs,
                                      ptrdiff_t stride);
 void ff_hevc_add_residual_4x4_10_neon(uint8_t *_dst, const int16_t *coeffs,
@@ -117,6 +129,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
     if (!have_neon(av_get_cpu_flags())) return;
 
     if (bit_depth == 8) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_8_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_8_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_8_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_8_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_8_neon;
@@ -167,6 +181,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->put_hevc_qpel_bi[9][0][1]   = ff_hevc_put_hevc_qpel_bi_h16_8_neon;
     }
     if (bit_depth == 10) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_10_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_10_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_10_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_10_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_10_neon;
@@ -180,6 +196,8 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->idct_dc[3]                  = ff_hevc_idct_32x32_dc_10_neon;
     }
     if (bit_depth == 12) {
+        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_12_neon;
+        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_12_neon;
         c->add_residual[0]             = ff_hevc_add_residual_4x4_12_neon;
         c->add_residual[1]             = ff_hevc_add_residual_8x8_12_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_12_neon;
-- 
2.39.2

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro
  2023-03-22  0:07     ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro J. Dekker
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test J. Dekker
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
@ 2023-03-22  8:38       ` Martin Storsjö
  2 siblings, 0 replies; 9+ messages in thread
From: Martin Storsjö @ 2023-03-22  8:38 UTC (permalink / raw)
  To: FFmpeg development discussions and patches

On Wed, 22 Mar 2023, J. Dekker wrote:

> Signed-off-by: J. Dekker <jdek@itanimul.li>
> ---
> libavcodec/aarch64/hevcdsp_idct_neon.S | 19 +++++--------------
> libavcodec/aarch64/neon.S              | 11 +++++++++++
> 2 files changed, 16 insertions(+), 14 deletions(-)
>
> diff --git a/libavcodec/aarch64/hevcdsp_idct_neon.S b/libavcodec/aarch64/hevcdsp_idct_neon.S
> index 467cb0f48a..3e59dd20bb 100644
> --- a/libavcodec/aarch64/hevcdsp_idct_neon.S
> +++ b/libavcodec/aarch64/hevcdsp_idct_neon.S
> @@ -5,7 +5,7 @@
>  *
>  * Ported from arm/hevcdsp_idct_neon.S by
>  * Copyright (c) 2020 Reimar Döffinger
> - * Copyright (c) 2020 J. Dekker
> + * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
>  *
>  * This file is part of FFmpeg.
>  *
> @@ -38,13 +38,6 @@ const trans, align=4
>         .short          31, 22, 13, 4
> endconst
> 
> -.macro clip2 in1, in2, min, max
> -        smax            \in1, \in1, \min
> -        smax            \in2, \in2, \min
> -        smin            \in1, \in1, \max
> -        smin            \in2, \in2, \max
> -.endm
> -
> function ff_hevc_add_residual_4x4_8_neon, export=1
>         ld1             {v0.8h-v1.8h}, [x1]
>         ld1             {v2.s}[0], [x0], x2
> @@ -182,7 +175,7 @@ function hevc_add_residual_4x4_16_neon, export=0
>         ld1             {v3.d}[1], [x12], x2
>         movi            v4.8h, #0
>         sqadd           v1.8h, v1.8h, v3.8h
> -        clip2           v0.8h, v1.8h, v4.8h, v21.8h
> +        clip            v4.8h, v21.8h, v0.8h, v1.8h
>         st1             {v0.d}[0], [x0],  x2
>         st1             {v0.d}[1], [x0],  x2
>         st1             {v1.d}[0], [x0],  x2
> @@ -201,7 +194,7 @@ function hevc_add_residual_8x8_16_neon, export=0
>         sqadd           v0.8h, v0.8h, v2.8h
>         ld1             {v3.8h}, [x12]
>         sqadd           v1.8h, v1.8h, v3.8h
> -        clip2           v0.8h, v1.8h, v4.8h, v21.8h
> +        clip            v4.8h, v21.8h, v0.8h, v1.8h
>         st1             {v0.8h}, [x0],  x2
>         st1             {v1.8h}, [x12], x2
>         bne             1b
> @@ -221,8 +214,7 @@ function hevc_add_residual_16x16_16_neon, export=0
>         sqadd           v1.8h, v1.8h, v17.8h
>         sqadd           v2.8h, v2.8h, v18.8h
>         sqadd           v3.8h, v3.8h, v19.8h
> -        clip2           v0.8h, v1.8h, v20.8h, v21.8h
> -        clip2           v2.8h, v3.8h, v20.8h, v21.8h
> +        clip            v20.8h, v21.8h, v0.8h, v1.8h, v2.8h, v3.8h
>         st1             {v0.8h-v1.8h}, [x0],  x2
>         st1             {v2.8h-v3.8h}, [x12], x2
>         bne             1b
> @@ -239,8 +231,7 @@ function hevc_add_residual_32x32_16_neon, export=0
>         sqadd           v1.8h, v1.8h, v17.8h
>         sqadd           v2.8h, v2.8h, v18.8h
>         sqadd           v3.8h, v3.8h, v19.8h
> -        clip2           v0.8h, v1.8h, v20.8h, v21.8h
> -        clip2           v2.8h, v3.8h, v20.8h, v21.8h
> +        clip            v20.8h, v21.8h, v0.8h, v1.8h, v2.8h, v3.8h
>         st1             {v0.8h-v3.8h}, [x0], x2
>         bne             1b
>         ret
> diff --git a/libavcodec/aarch64/neon.S b/libavcodec/aarch64/neon.S
> index 1ad32c359d..bc105e4861 100644
> --- a/libavcodec/aarch64/neon.S
> +++ b/libavcodec/aarch64/neon.S
> @@ -1,6 +1,8 @@
> /*
>  * This file is part of FFmpeg.
>  *
> + * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
> + *
>  * FFmpeg is free software; you can redistribute it and/or
>  * modify it under the terms of the GNU Lesser General Public
>  * License as published by the Free Software Foundation; either
> @@ -16,6 +18,15 @@
>  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
>  */
> 
> +.macro clip min, max, regs:vararg
> +.irp x, \regs
> +        smax            \x, \x, \min
> +.endr
> +.irp x, \regs
> +        smin            \x, \x, \max
> +.endr
> +.endm
> +

LGTM, the vararg argument handling looks neat here.

// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test J. Dekker
@ 2023-03-22  9:04         ` Martin Storsjö
  0 siblings, 0 replies; 9+ messages in thread
From: Martin Storsjö @ 2023-03-22  9:04 UTC (permalink / raw)
  To: FFmpeg development discussions and patches

On Wed, 22 Mar 2023, J. Dekker wrote:

> Signed-off-by: J. Dekker <jdek@itanimul.li>
> ---
>
> Added missing FATE target.
>
> tests/checkasm/Makefile       |  2 +-
> tests/checkasm/checkasm.c     |  1 +
> tests/checkasm/checkasm.h     |  1 +
> tests/checkasm/hevc_deblock.c | 87 +++++++++++++++++++++++++++++++++++
> tests/fate/checkasm.mak       |  1 +
> 5 files changed, 91 insertions(+), 1 deletion(-)
> create mode 100644 tests/checkasm/hevc_deblock.c
>
> diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> index a6f06c7007..3e62a22bd6 100644
> --- a/tests/checkasm/Makefile
> +++ b/tests/checkasm/Makefile
> @@ -28,7 +28,7 @@ AVCODECOBJS-$(CONFIG_HUFFYUV_DECODER)   += huffyuvdsp.o
> AVCODECOBJS-$(CONFIG_JPEG2000_DECODER)  += jpeg2000dsp.o
> AVCODECOBJS-$(CONFIG_OPUS_DECODER)      += opusdsp.o
> AVCODECOBJS-$(CONFIG_PIXBLOCKDSP)       += pixblockdsp.o
> -AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_idct.o hevc_sao.o hevc_pel.o
> +AVCODECOBJS-$(CONFIG_HEVC_DECODER)      += hevc_add_res.o hevc_deblock.o hevc_idct.o hevc_sao.o hevc_pel.o
> AVCODECOBJS-$(CONFIG_UTVIDEO_DECODER)   += utvideodsp.o
> AVCODECOBJS-$(CONFIG_V210_DECODER)      += v210dec.o
> AVCODECOBJS-$(CONFIG_V210_ENCODER)      += v210enc.o
> diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
> index e96d84a7da..c2184d260d 100644
> --- a/tests/checkasm/checkasm.c
> +++ b/tests/checkasm/checkasm.c
> @@ -116,6 +116,7 @@ static const struct {
>     #endif
>     #if CONFIG_HEVC_DECODER
>         { "hevc_add_res", checkasm_check_hevc_add_res },
> +        { "hevc_deblock", checkasm_check_hevc_deblock },
>         { "hevc_idct", checkasm_check_hevc_idct },
>         { "hevc_pel", checkasm_check_hevc_pel },
>         { "hevc_sao", checkasm_check_hevc_sao },
> diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
> index 8744a81218..89c643e6a0 100644
> --- a/tests/checkasm/checkasm.h
> +++ b/tests/checkasm/checkasm.h
> @@ -60,6 +60,7 @@ void checkasm_check_h264dsp(void);
> void checkasm_check_h264pred(void);
> void checkasm_check_h264qpel(void);
> void checkasm_check_hevc_add_res(void);
> +void checkasm_check_hevc_deblock(void);
> void checkasm_check_hevc_idct(void);
> void checkasm_check_hevc_pel(void);
> void checkasm_check_hevc_sao(void);
> diff --git a/tests/checkasm/hevc_deblock.c b/tests/checkasm/hevc_deblock.c
> new file mode 100644
> index 0000000000..06c9c1969e
> --- /dev/null
> +++ b/tests/checkasm/hevc_deblock.c
> @@ -0,0 +1,87 @@
> +/*
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License along
> + * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
> + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
> + */
> +
> +#include <string.h>
> +
> +#include "libavutil/intreadwrite.h"
> +#include "libavutil/mem_internal.h"
> +
> +#include "libavcodec/avcodec.h"
> +#include "libavcodec/hevcdsp.h"
> +
> +#include "checkasm.h"
> +
> +static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
> +
> +#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
> +#define BUF_STRIDE (8 * 2)
> +#define BUF_LINES (8)
> +#define BUF_OFFSET (BUF_STRIDE * BUF_LINES)
> +#define BUF_SIZE (BUF_STRIDE * BUF_LINES + BUF_OFFSET * 2)

This looks like it's at least plenty of space, although it leaves me a bit 
confused about the intended layout of the test buffer. The way I think 
about it is that we'd have our buffer to be e.g. a 8x8 square (or 16x16); 
then when we're deblocking vertically, we'd have BUF_OFFSET be 
(height/2)*stride, i.e. starting at the middle left side of the square. 
When deblocking horizontally, the start position can be (width/2), i.e at 
the center of the top edge of the square.

The current version is acceptable though, but I don't quite get the 
intended layout/sizing here, other than "big enough".

> +
> +#define randomize_buffers(buf0, buf1, size)                 \
> +    do {                                                    \
> +        uint32_t mask = pixel_mask[(bit_depth - 8) >> 1];   \
> +        int k;                                              \
> +        for (k = 0; k < size; k += 4) {                     \
> +            uint32_t r = rnd() & mask;                      \
> +            AV_WN32A(buf0 + k, r);                          \
> +            AV_WN32A(buf1 + k, r);                          \
> +        }                                                   \
> +    } while (0)
> +
> +static void check_deblock_chroma(HEVCDSPContext h, int bit_depth)
> +{
> +    int32_t tc[2] = { 1, 1 };
> +    uint8_t no_p[2] = { 0, 0 };
> +    uint8_t no_q[2] = { 0, 0 };

I think it might be good to check what we honor these flags too; either an 
exhaustive loop testing all combinations of 0/1 for the no_p/no_q flags, 
plus possibly some different values for tc - or at least maybe a couple 
tests with these flags/coefficients set to random combinations?

> +    LOCAL_ALIGNED_32(uint8_t, buf0, [BUF_SIZE]);
> +    LOCAL_ALIGNED_32(uint8_t, buf1, [BUF_SIZE]);
> +
> +    declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *pix, ptrdiff_t stride, int32_t *tc, uint8_t *no_p, uint8_t *no_q);
> +
> +    randomize_buffers(buf0, buf1, BUF_SIZE);
> +    if (check_func(h.hevc_h_loop_filter_chroma, "hevc_h_loop_filter_chroma_%d", bit_depth)) {
> +        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +        if (memcmp(buf0, buf1, BUF_SIZE))
> +            fail();
> +        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +    }
> +
> +    randomize_buffers(buf0, buf1, BUF_SIZE);
> +    if (check_func(h.hevc_v_loop_filter_chroma, "hevc_v_loop_filter_chroma_%d", bit_depth)) {
> +        call_ref(buf0 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +        call_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +        if (memcmp(buf0, buf1, BUF_SIZE))
> +            fail();
> +        bench_new(buf1 + BUF_OFFSET, BUF_STRIDE, tc, no_p, no_q);
> +    }
> +}

I mentioned last time that pure random data isn't a very good input for 
deblocking tests, but I see that the hevc chroma deblocker doesn't compare 
pixel differences for deciding whether to deblock or not, so with that in 
mind, I guess this is sufficient.

The current test is at least good enough for making sure we run through 
the code (for the non-functional test aspects, register clobbering etc) 
and makes sure we do some work so that we can do proper benchmarking.

So I guess this is acceptable in this form, but with a suggestion to check 
more combinations of tc/no_p/no_c.

// Martin

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit
  2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
@ 2023-03-22  9:26         ` Martin Storsjö
  0 siblings, 0 replies; 9+ messages in thread
From: Martin Storsjö @ 2023-03-22  9:26 UTC (permalink / raw)
  To: FFmpeg development discussions and patches

On Wed, 22 Mar 2023, J. Dekker wrote:

> Signed-off-by: J. Dekker <jdek@itanimul.li>
> ---
>
> - Using clip macro
> - Avoided need for .elseif at all, just used better mvni

Can you provide some benchmark numbers for it in the commit message, to 
get the ballpark figures?

> libavcodec/aarch64/Makefile               |   3 +-
> libavcodec/aarch64/hevcdsp_deblock_neon.S | 180 ++++++++++++++++++++++
> libavcodec/aarch64/hevcdsp_init_aarch64.c |  18 +++
> 3 files changed, 200 insertions(+), 1 deletion(-)
> create mode 100644 libavcodec/aarch64/hevcdsp_deblock_neon.S
>
> diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
> index 02fb51c3ab..216191640c 100644
> --- a/libavcodec/aarch64/Makefile
> +++ b/libavcodec/aarch64/Makefile
> @@ -65,7 +65,8 @@ NEON-OBJS-$(CONFIG_VP9_DECODER)         += aarch64/vp9itxfm_16bpp_neon.o       \
>                                            aarch64/vp9lpf_neon.o               \
>                                            aarch64/vp9mc_16bpp_neon.o          \
>                                            aarch64/vp9mc_neon.o
> -NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_idct_neon.o         \
> +NEON-OBJS-$(CONFIG_HEVC_DECODER)        += aarch64/hevcdsp_deblock_neon.o      \
> +                                           aarch64/hevcdsp_idct_neon.o         \
>                                            aarch64/hevcdsp_init_aarch64.o      \
>                                            aarch64/hevcdsp_qpel_neon.o         \
>                                            aarch64/hevcdsp_sao_neon.o
> diff --git a/libavcodec/aarch64/hevcdsp_deblock_neon.S b/libavcodec/aarch64/hevcdsp_deblock_neon.S
> new file mode 100644
> index 0000000000..49b40f21c8
> --- /dev/null
> +++ b/libavcodec/aarch64/hevcdsp_deblock_neon.S
> @@ -0,0 +1,180 @@
> +/* -*-arm64-*-
> + * vim: syntax=arm64asm
> + *
> + * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
> + * Copyright (c) 2023 J. Dekker <jdek@itanimul.li>
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +
> +#include "libavutil/aarch64/asm.S"
> +#include "neon.S"
> +
> +.macro hevc_loop_filter_chroma_start bitdepth
> +        mov             x4, x30
> +        ldr             w14, [x2]
> +        ldr             w15, [x2, #4]
> +.if \bitdepth > 8
> +        lsl             w14, w14, #(\bitdepth - 8)
> +        lsl             w15, w15, #(\bitdepth - 8)
> +.endif
> +        adds            w2, w14, w15
> +        b.eq            1f
> +        dup             v16.4h, w14
> +        dup             v17.4h, w15
> +        trn1            v16.2d, v16.2d, v17.2d
> +.if \bitdepth > 8
> +        mvni            v19.8h, #((0xff << (\bitdepth - 8)) & 0xff), lsl #8
> +        movi            v18.8h, #0
> +.endif
> +        neg             v17.8h, v16.8h
> +.endm
> +
> +.macro hevc_loop_filter_chroma_body bitdepth
> +.if \bitdepth <= 8
> +        uxtl            v0.8h, v0.8b // p1
> +        uxtl            v1.8h, v1.8b // p0
> +        uxtl            v2.8h, v2.8b // q0
> +        uxtl            v3.8h, v3.8b // q1
> +.endif
> +        sub             v5.8h, v2.8h, v1.8h // q0 - p0
> +        sub             v6.8h, v0.8h, v3.8h // p1 - q1
> +        shl             v5.8h, v5.8h, #2
> +        add             v5.8h, v6.8h, v5.8h
> +        srshr           v5.8h, v5.8h, #3
> +        clip            v17.8h, v16.8h, v5.8h
> +        sqadd           v1.8h, v1.8h, v5.8h // p0 + delta
> +        sqsub           v2.8h, v2.8h, v5.8h // q0 - delta

For the bitdepth==8 case, I'm wondering if it'd be more straightforward to 
do like the arm code does, i.e. don't do uxtl at the start, but widen with 
usubl at the "q0-p0" stage, and then add with uaddw at the end. That makes 
the function less templateable though. Would you mind giving that a try 
just to get benchmark numbers for it? If there's not a big difference I 
guess this more templated form is fine too.

If you try this, just comment out the more complex function that does 
transposes. (We can simplify that one a little if we don't need to reverse 
the uxtl.)

> +.if \bitdepth <= 8
> +        sqxtun          v1.8b, v1.8h
> +        sqxtun          v2.8b, v2.8h
> +.else
> +        smin            v1.8h, v1.8h, v19.8h
> +        smin            v2.8h, v2.8h, v19.8h
> +        clip            v18.8h, v19.8h, v1.8h, v2.8h

Don't you do duplicate clipping here - the smin instructions are redundant 
with the clip macro?

> +.endif
> +.endm
> +
> +function hevc_loop_filter_chroma_body_8_neon, export=0
> +        hevc_loop_filter_chroma_body 8
> +        ret
> +endfunc
> +
> +function hevc_loop_filter_chroma_body_10_neon, export=0
> +hevc_loop_filter_chroma_body_12_neon:
> +        hevc_loop_filter_chroma_body 10
> +        ret
> +endfunc
> +
> +// void ff_hevc_h_loop_filter_chroma_8_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
> +
> +.macro hevc_h_loop_filter_chroma bitdepth
> +function ff_hevc_h_loop_filter_chroma_\bitdepth\()_neon, export=1
> +        hevc_loop_filter_chroma_start \bitdepth
> +        sub             x0, x0, x1, lsl #1
> +.if \bitdepth > 8
> +        ld1             {v0.8h}, [x0], x1
> +        ld1             {v1.8h}, [x0], x1
> +        ld1             {v2.8h}, [x0], x1
> +        ld1             {v3.8h}, [x0]
> +.else
> +        ld1             {v0.8b}, [x0], x1
> +        ld1             {v1.8b}, [x0], x1
> +        ld1             {v2.8b}, [x0], x1
> +        ld1             {v3.8b}, [x0]
> +.endif
> +        sub             x0, x0, x1, lsl #1
> +        bl              hevc_loop_filter_chroma_body_\bitdepth\()_neon
> +.if \bitdepth > 8
> +        st1             {v1.8h}, [x0], x1
> +        st1             {v2.8h}, [x0]
> +.else
> +        st1             {v1.8b}, [x0], x1
> +        st1             {v2.8b}, [x0]
> +.endif
> +1:      ret             x4
> +endfunc
> +.endm
> +
> +.macro hevc_v_loop_filter_chroma bitdepth
> +function ff_hevc_v_loop_filter_chroma_\bitdepth\()_neon, export=1
> +        hevc_loop_filter_chroma_start \bitdepth
> +.if \bitdepth > 8
> +        sub             x0, x0, #8
> +        add             x3, x0, x1
> +        lsl             x1, x1, #1
> +        ld1             {v20.8h}, [x0], x1
> +        ld1             {v21.8h}, [x3], x1
> +        ld1             {v0.8h},  [x0], x1
> +        ld1             {v1.8h},  [x3], x1
> +        ld1             {v2.8h},  [x0], x1
> +        ld1             {v3.8h},  [x3], x1
> +        ld1             {v22.8h}, [x0], x1
> +        ld1             {v23.8h}, [x3], x1
> +        transpose_8x8H  v20, v21, v0, v1, v2, v3, v22, v23, v24, v25

Reading, transposing and writing back all of 8x8 pixels here, when we 
really just want a 4x8 slice of pixels, is a bit excessive.

I see that the existing 32 bit assembly does it like that though, so I 
guess it's acceptable - but I'd like to leave a remark that it can be done 
more optimized.

I believe this should be doable by loading the first 4 rows into v0-v3 
.d[0] for 10/12 bpp and .s[0] for 8bpp, the last 4 rows into .d[1] or 
.s[1], then do a transpose_4x8H or 4x8B on it, then the same in reverse at 
the end.

// Martin

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

^ permalink raw reply	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2023-03-22  9:27 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-21 18:35 [FFmpeg-devel] [PATCH 1/2] checkasm: add hevc_deblock chroma test J. Dekker
2023-03-21 18:35 ` [FFmpeg-devel] [PATCH v2 2/2] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
2023-03-21 20:30   ` Martin Storsjö
2023-03-22  0:07     ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro J. Dekker
2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 2/3] checkasm: add hevc_deblock chroma test J. Dekker
2023-03-22  9:04         ` Martin Storsjö
2023-03-22  0:07       ` [FFmpeg-devel] [PATCH v3 3/3] lavc/aarch64: add hevc deblock chroma 8-12bit J. Dekker
2023-03-22  9:26         ` Martin Storsjö
2023-03-22  8:38       ` [FFmpeg-devel] [PATCH 1/3] lavc/aarch64: add clip N macro Martin Storsjö

Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
		ffmpegdev@gitmailbox.com
	public-inbox-index ffmpegdev

Example config snippet for mirrors.


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git