arm64: int32_to_float_fmul neon asm
authorJanne Grunau <janne-libav@jannau.net>
Thu, 3 Dec 2015 10:04:29 +0000 (11:04 +0100)
committerJanne Grunau <janne-libav@jannau.net>
Mon, 14 Dec 2015 15:45:02 +0000 (16:45 +0100)
3% faster dts decoding on a cortex-a57.

                                 cortex-a57   cortex-a53
int32_to_float_fmul_array8_c:    1270.9       4475.6
int32_to_float_fmul_array8_neon:  328.6        569.2
int32_to_float_fmul_scalar_c:     928.5       4119.6
int32_to_float_fmul_scalar_neon:  309.1        524.1

libavcodec/aarch64/Makefile
libavcodec/aarch64/fmtconvert_init.c [new file with mode: 0644]
libavcodec/aarch64/fmtconvert_neon.S [new file with mode: 0644]
libavcodec/fmtconvert.c
libavcodec/fmtconvert.h

index 2175578..022ed84 100644 (file)
@@ -1,5 +1,6 @@
 OBJS-$(CONFIG_DCA_DECODER)              += aarch64/dcadsp_init.o
 OBJS-$(CONFIG_FFT)                      += aarch64/fft_init_aarch64.o
+OBJS-$(CONFIG_FMTCONVERT)               += aarch64/fmtconvert_init.o
 OBJS-$(CONFIG_H264CHROMA)               += aarch64/h264chroma_init_aarch64.o
 OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o
 OBJS-$(CONFIG_H264PRED)                 += aarch64/h264pred_init.o
@@ -19,6 +20,7 @@ ARMV8-OBJS-$(CONFIG_VIDEODSP)           += aarch64/videodsp.o
 NEON-OBJS-$(CONFIG_DCA_DECODER)         += aarch64/dcadsp_neon.o               \
                                            aarch64/synth_filter_neon.o
 NEON-OBJS-$(CONFIG_FFT)                 += aarch64/fft_neon.o
+NEON-OBJS-$(CONFIG_FMTCONVERT)          += aarch64/fmtconvert_neon.o
 NEON-OBJS-$(CONFIG_H264CHROMA)          += aarch64/h264cmc_neon.o
 NEON-OBJS-$(CONFIG_H264DSP)             += aarch64/h264dsp_neon.o              \
                                            aarch64/h264idct_neon.o
diff --git a/libavcodec/aarch64/fmtconvert_init.c b/libavcodec/aarch64/fmtconvert_init.c
new file mode 100644 (file)
index 0000000..0a55a1b
--- /dev/null
@@ -0,0 +1,43 @@
+/*
+ * ARM optimized Format Conversion Utils
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/attributes.h"
+#include "libavutil/aarch64/cpu.h"
+#include "libavcodec/avcodec.h"
+#include "libavcodec/fmtconvert.h"
+
+void ff_int32_to_float_fmul_array8_neon(FmtConvertContext *c, float *dst,
+                                        const int32_t *src, const float *mul,
+                                        int len);
+void ff_int32_to_float_fmul_scalar_neon(float *dst, const int32_t *src,
+                                        float mul, int len);
+
+av_cold void ff_fmt_convert_init_aarch64(FmtConvertContext *c,
+                                         AVCodecContext *avctx)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+        c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_neon;
+        c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_neon;
+    }
+}
diff --git a/libavcodec/aarch64/fmtconvert_neon.S b/libavcodec/aarch64/fmtconvert_neon.S
new file mode 100644 (file)
index 0000000..3b33c87
--- /dev/null
@@ -0,0 +1,76 @@
+/*
+ * ARM NEON optimised Format Conversion Utils
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ * Copyright (c) 2015 Janne Grunau  <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "libavutil/aarch64/asm.S"
+
+function ff_int32_to_float_fmul_scalar_neon, export=1
+        ld1             {v1.4s,v2.4s}, [x1], #32
+        scvtf           v1.4s,  v1.4s
+        scvtf           v2.4s,  v2.4s
+1:
+        subs            w2,  w2,  #8
+        fmul            v3.4s,  v1.4s,  v0.s[0]
+        fmul            v4.4s,  v2.4s,  v0.s[0]
+        b.le            2f
+        ld1             {v1.4s,v2.4s}, [x1], #32
+        st1             {v3.4s,v4.4s}, [x0], #32
+        scvtf           v1.4s,  v1.4s
+        scvtf           v2.4s,  v2.4s
+        b               1b
+2:
+        st1             {v3.4s,v4.4s}, [x0]
+        ret
+endfunc
+
+function ff_int32_to_float_fmul_array8_neon, export=1
+        lsr             w4,  w4,  #3
+        subs            w5,  w4,  #1
+        b.eq            1f
+2:
+        ld1             {v0.4s,v1.4s}, [x2], #32
+        ld1             {v2.4s,v3.4s}, [x2], #32
+        scvtf           v0.4s,  v0.4s
+        scvtf           v1.4s,  v1.4s
+        ld1             {v16.2s},  [x3], #8
+        scvtf           v2.4s,  v2.4s
+        scvtf           v3.4s,  v3.4s
+        fmul            v4.4s,  v0.4s,  v16.s[0]
+        fmul            v5.4s,  v1.4s,  v16.s[0]
+        fmul            v6.4s,  v2.4s,  v16.s[1]
+        fmul            v7.4s,  v3.4s,  v16.s[1]
+        st1             {v4.4s,v5.4s}, [x1], #32
+        st1             {v6.4s,v7.4s}, [x1], #32
+        subs            w5,  w5,  #2
+        b.gt            2b
+        b.eq            1f
+        ret
+1:
+        ld1             {v0.4s,v1.4s}, [x2]
+        ld1             {v16.s}[0],  [x3]
+        scvtf           v0.4s,  v0.4s
+        scvtf           v1.4s,  v1.4s
+        fmul            v4.4s,  v0.4s,  v16.s[0]
+        fmul            v5.4s,  v1.4s,  v16.s[0]
+        st1             {v4.4s,v5.4s}, [x1]
+        ret
+endfunc
index 2dff704..5e29bfa 100644 (file)
@@ -46,7 +46,12 @@ av_cold void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx)
     c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_c;
     c->int32_to_float_fmul_array8 = int32_to_float_fmul_array8_c;
 
-    if (ARCH_ARM) ff_fmt_convert_init_arm(c, avctx);
-    if (ARCH_PPC) ff_fmt_convert_init_ppc(c, avctx);
-    if (ARCH_X86) ff_fmt_convert_init_x86(c, avctx);
+    if (ARCH_AARCH64)
+        ff_fmt_convert_init_aarch64(c, avctx);
+    if (ARCH_ARM)
+        ff_fmt_convert_init_arm(c, avctx);
+    if (ARCH_PPC)
+        ff_fmt_convert_init_ppc(c, avctx);
+    if (ARCH_X86)
+        ff_fmt_convert_init_x86(c, avctx);
 }
index 7de890b..dd603f9 100644 (file)
@@ -58,6 +58,7 @@ typedef struct FmtConvertContext {
 
 void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx);
 
+void ff_fmt_convert_init_aarch64(FmtConvertContext *c, AVCodecContext *avctx);
 void ff_fmt_convert_init_arm(FmtConvertContext *c, AVCodecContext *avctx);
 void ff_fmt_convert_init_ppc(FmtConvertContext *c, AVCodecContext *avctx);
 void ff_fmt_convert_init_x86(FmtConvertContext *c, AVCodecContext *avctx);