mips: Optimization of AC3 FP encoder and EAC3 FP decoder
authorBojan Zivkovic <bojan@mips.com>
Tue, 6 Nov 2012 13:06:21 +0000 (14:06 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Mon, 19 Nov 2012 21:24:44 +0000 (22:24 +0100)
Signed-off-by: Bojan Zivkovic <bojan@mips.com>
Reveiwed-by: Vitor Sessak <vitor1001@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
doc/mips.txt
libavcodec/ac3dsp.c
libavcodec/ac3dsp.h
libavcodec/mips/Makefile
libavcodec/mips/ac3dsp_mips.c [new file with mode: 0644]
libavutil/float_dsp.c
libavutil/float_dsp.h
libavutil/mips/Makefile [new file with mode: 0644]
libavutil/mips/float_dsp_mips.c [new file with mode: 0644]

index 6fa6fb4..aabdef0 100644 (file)
@@ -45,8 +45,10 @@ Example of copyright notice:
 Files that have MIPS copyright notice in them:
 ===============================================
 * libavutil/mips/
+      float_dsp_mips.c
       libm_mips.h
 * libavcodec/mips/
+      ac3dsp_mips.c
       acelp_filters_mips.c
       acelp_vectors_mips.c
       amrwbdec_mips.c
index 49866eb..9277411 100644 (file)
@@ -258,4 +258,6 @@ av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact)
         ff_ac3dsp_init_arm(c, bit_exact);
     if (ARCH_X86)
         ff_ac3dsp_init_x86(c, bit_exact);
+    if (ARCH_MIPS)
+        ff_ac3dsp_init_mips(c, bit_exact);
 }
index fbc63f6..7269c57 100644 (file)
@@ -139,5 +139,6 @@ typedef struct AC3DSPContext {
 void ff_ac3dsp_init    (AC3DSPContext *c, int bit_exact);
 void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact);
 void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact);
+void ff_ac3dsp_init_mips(AC3DSPContext *c, int bit_exact);
 
 #endif /* AVCODEC_AC3DSP_H */
index f4c6ac8..feeec99 100644 (file)
@@ -13,3 +13,4 @@ OBJS-$(CONFIG_FFT)                        += mips/fft_init_table.o
 MIPSFPU-OBJS-$(CONFIG_FFT)                += mips/fft_mips.o
 MIPSFPU-OBJS                              += mips/dsputil_mips.o           \
                                              mips/fmtconvert_mips.o
+OBJS-$(CONFIG_AC3DSP)                     += mips/ac3dsp_mips.o
diff --git a/libavcodec/mips/ac3dsp_mips.c b/libavcodec/mips/ac3dsp_mips.c
new file mode 100644 (file)
index 0000000..e503645
--- /dev/null
@@ -0,0 +1,372 @@
+/*
+ * Copyright (c) 2012
+ *      MIPS Technologies, Inc., California.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
+ *    contributors may be used to endorse or promote products derived from
+ *    this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * Author:  Branimir Vasic (bvasic@mips.com)
+ *
+ * Various AC-3 DSP Utils optimized for MIPS
+ *
+ * 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
+ */
+
+/**
+ * @file
+ * Reference: libavcodec/ac3dsp.c
+ */
+
+#include "config.h"
+#include "libavcodec/ac3dsp.h"
+#include "libavcodec/ac3.h"
+
+
+#if HAVE_INLINE_ASM
+#if HAVE_MIPSDSPR1
+static void ac3_bit_alloc_calc_bap_mips(int16_t *mask, int16_t *psd,
+                                        int start, int end,
+                                        int snr_offset, int floor,
+                                        const uint8_t *bap_tab, uint8_t *bap)
+{
+    int band, band_end, cond;
+    int m, address1, address2;
+    int16_t *psd1, *psd_end;
+    uint8_t *bap1;
+
+    if (snr_offset == -960) {
+        memset(bap, 0, AC3_MAX_COEFS);
+        return;
+    }
+
+    psd1 = &psd[start];
+    bap1 = &bap[start];
+    band = ff_ac3_bin_to_band_tab[start];
+
+    do {
+        m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor;
+        band_end = ff_ac3_band_start_tab[++band];
+        band_end = FFMIN(band_end, end);
+        psd_end = psd + band_end - 1;
+
+        __asm__ volatile (
+            "slt        %[cond],        %[psd1],        %[psd_end]  \n\t"
+            "beqz       %[cond],        1f                          \n\t"
+            "2:                                                     \n\t"
+            "lh         %[address1],    0(%[psd1])                  \n\t"
+            "lh         %[address2],    2(%[psd1])                  \n\t"
+            "addiu      %[psd1],        %[psd1],        4           \n\t"
+            "subu       %[address1],    %[address1],    %[m]        \n\t"
+            "sra        %[address1],    %[address1],    5           \n\t"
+            "addiu      %[address1],    %[address1],    -32         \n\t"
+            "shll_s.w   %[address1],    %[address1],    26          \n\t"
+            "subu       %[address2],    %[address2],    %[m]        \n\t"
+            "sra        %[address2],    %[address2],    5           \n\t"
+            "sra        %[address1],    %[address1],    26          \n\t"
+            "addiu      %[address1],    %[address1],    32          \n\t"
+            "lbux       %[address1],    %[address1](%[bap_tab])     \n\t"
+            "addiu      %[address2],    %[address2],    -32         \n\t"
+            "shll_s.w   %[address2],    %[address2],    26          \n\t"
+            "sb         %[address1],    0(%[bap1])                  \n\t"
+            "slt        %[cond],        %[psd1],        %[psd_end]  \n\t"
+            "sra        %[address2],    %[address2],    26          \n\t"
+            "addiu      %[address2],    %[address2],    32          \n\t"
+            "lbux       %[address2],    %[address2](%[bap_tab])     \n\t"
+            "sb         %[address2],    1(%[bap1])                  \n\t"
+            "addiu      %[bap1],        %[bap1],        2           \n\t"
+            "bnez       %[cond],        2b                          \n\t"
+            "addiu      %[psd_end],     %[psd_end],     2           \n\t"
+            "slt        %[cond],        %[psd1],        %[psd_end]  \n\t"
+            "beqz       %[cond],        3f                          \n\t"
+            "1:                                                     \n\t"
+            "lh         %[address1],    0(%[psd1])                  \n\t"
+            "addiu      %[psd1],        %[psd1],        2           \n\t"
+            "subu       %[address1],    %[address1],    %[m]        \n\t"
+            "sra        %[address1],    %[address1],    5           \n\t"
+            "addiu      %[address1],    %[address1],    -32         \n\t"
+            "shll_s.w   %[address1],    %[address1],    26          \n\t"
+            "sra        %[address1],    %[address1],    26          \n\t"
+            "addiu      %[address1],    %[address1],    32          \n\t"
+            "lbux       %[address1],    %[address1](%[bap_tab])     \n\t"
+            "sb         %[address1],    0(%[bap1])                  \n\t"
+            "addiu      %[bap1],        %[bap1],        1           \n\t"
+            "3:                                                     \n\t"
+
+            : [address1]"=&r"(address1), [address2]"=&r"(address2),
+              [cond]"=&r"(cond), [bap1]"+r"(bap1),
+              [psd1]"+r"(psd1), [psd_end]"+r"(psd_end)
+            : [m]"r"(m), [bap_tab]"r"(bap_tab)
+            : "memory"
+        );
+    } while (end > band_end);
+}
+
+static void ac3_update_bap_counts_mips(uint16_t mant_cnt[16], uint8_t *bap,
+                                       int len)
+{
+    int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+    __asm__ volatile (
+        "andi   %[temp3],   %[len],         3               \n\t"
+        "addu   %[temp2],   %[bap],         %[len]          \n\t"
+        "addu   %[temp4],   %[bap],         %[temp3]        \n\t"
+        "beq    %[temp2],   %[temp4],       4f              \n\t"
+        "1:                                                 \n\t"
+        "lbu    %[temp0],   -1(%[temp2])                    \n\t"
+        "lbu    %[temp5],   -2(%[temp2])                    \n\t"
+        "lbu    %[temp6],   -3(%[temp2])                    \n\t"
+        "sll    %[temp0],   %[temp0],       1               \n\t"
+        "addu   %[temp0],   %[mant_cnt],    %[temp0]        \n\t"
+        "sll    %[temp5],   %[temp5],       1               \n\t"
+        "addu   %[temp5],   %[mant_cnt],    %[temp5]        \n\t"
+        "lhu    %[temp1],   0(%[temp0])                     \n\t"
+        "sll    %[temp6],   %[temp6],       1               \n\t"
+        "addu   %[temp6],   %[mant_cnt],    %[temp6]        \n\t"
+        "addiu  %[temp1],   %[temp1],       1               \n\t"
+        "sh     %[temp1],   0(%[temp0])                     \n\t"
+        "lhu    %[temp1],   0(%[temp5])                     \n\t"
+        "lbu    %[temp7],   -4(%[temp2])                    \n\t"
+        "addiu  %[temp2],   %[temp2],       -4              \n\t"
+        "addiu  %[temp1],   %[temp1],       1               \n\t"
+        "sh     %[temp1],   0(%[temp5])                     \n\t"
+        "lhu    %[temp1],   0(%[temp6])                     \n\t"
+        "sll    %[temp7],   %[temp7],       1               \n\t"
+        "addu   %[temp7],   %[mant_cnt],    %[temp7]        \n\t"
+        "addiu  %[temp1],   %[temp1],1                      \n\t"
+        "sh     %[temp1],   0(%[temp6])                     \n\t"
+        "lhu    %[temp1],   0(%[temp7])                     \n\t"
+        "addiu  %[temp1],   %[temp1],       1               \n\t"
+        "sh     %[temp1],   0(%[temp7])                     \n\t"
+        "bne    %[temp2],   %[temp4],       1b              \n\t"
+        "4:                                                 \n\t"
+        "beqz   %[temp3],   2f                              \n\t"
+        "3:                                                 \n\t"
+        "addiu  %[temp3],   %[temp3],       -1              \n\t"
+        "lbu    %[temp0],   -1(%[temp2])                    \n\t"
+        "addiu  %[temp2],   %[temp2],       -1              \n\t"
+        "sll    %[temp0],   %[temp0],       1               \n\t"
+        "addu   %[temp0],   %[mant_cnt],    %[temp0]        \n\t"
+        "lhu    %[temp1],   0(%[temp0])                     \n\t"
+        "addiu  %[temp1],   %[temp1],       1               \n\t"
+        "sh     %[temp1],   0(%[temp0])                     \n\t"
+        "bgtz   %[temp3],   3b                              \n\t"
+        "2:                                                 \n\t"
+
+        : [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
+          [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
+          [temp4] "=&r" (temp4), [temp5] "=&r" (temp5),
+          [temp6] "=&r" (temp6), [temp7] "=&r" (temp7)
+        : [len] "r" (len), [bap] "r" (bap),
+          [mant_cnt] "r" (mant_cnt)
+        : "memory"
+    );
+}
+#endif
+
+#if HAVE_MIPSFPU
+static void float_to_fixed24_mips(int32_t *dst, const float *src, unsigned int len)
+{
+    const float scale = 1 << 24;
+    float src0, src1, src2, src3, src4, src5, src6, src7;
+    int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
+
+    do {
+        __asm__ volatile (
+            "lwc1       %[src0],    0(%[src])               \n\t"
+            "lwc1       %[src1],    4(%[src])               \n\t"
+            "lwc1       %[src2],    8(%[src])               \n\t"
+            "lwc1       %[src3],    12(%[src])              \n\t"
+            "lwc1       %[src4],    16(%[src])              \n\t"
+            "lwc1       %[src5],    20(%[src])              \n\t"
+            "lwc1       %[src6],    24(%[src])              \n\t"
+            "lwc1       %[src7],    28(%[src])              \n\t"
+            "mul.s      %[src0],    %[src0],    %[scale]    \n\t"
+            "mul.s      %[src1],    %[src1],    %[scale]    \n\t"
+            "mul.s      %[src2],    %[src2],    %[scale]    \n\t"
+            "mul.s      %[src3],    %[src3],    %[scale]    \n\t"
+            "mul.s      %[src4],    %[src4],    %[scale]    \n\t"
+            "mul.s      %[src5],    %[src5],    %[scale]    \n\t"
+            "mul.s      %[src6],    %[src6],    %[scale]    \n\t"
+            "mul.s      %[src7],    %[src7],    %[scale]    \n\t"
+            "cvt.w.s    %[src0],    %[src0]                 \n\t"
+            "cvt.w.s    %[src1],    %[src1]                 \n\t"
+            "cvt.w.s    %[src2],    %[src2]                 \n\t"
+            "cvt.w.s    %[src3],    %[src3]                 \n\t"
+            "cvt.w.s    %[src4],    %[src4]                 \n\t"
+            "cvt.w.s    %[src5],    %[src5]                 \n\t"
+            "cvt.w.s    %[src6],    %[src6]                 \n\t"
+            "cvt.w.s    %[src7],    %[src7]                 \n\t"
+            "mfc1       %[temp0],   %[src0]                 \n\t"
+            "mfc1       %[temp1],   %[src1]                 \n\t"
+            "mfc1       %[temp2],   %[src2]                 \n\t"
+            "mfc1       %[temp3],   %[src3]                 \n\t"
+            "mfc1       %[temp4],   %[src4]                 \n\t"
+            "mfc1       %[temp5],   %[src5]                 \n\t"
+            "mfc1       %[temp6],   %[src6]                 \n\t"
+            "mfc1       %[temp7],   %[src7]                 \n\t"
+            "sw         %[temp0],   0(%[dst])               \n\t"
+            "sw         %[temp1],   4(%[dst])               \n\t"
+            "sw         %[temp2],   8(%[dst])               \n\t"
+            "sw         %[temp3],   12(%[dst])              \n\t"
+            "sw         %[temp4],   16(%[dst])              \n\t"
+            "sw         %[temp5],   20(%[dst])              \n\t"
+            "sw         %[temp6],   24(%[dst])              \n\t"
+            "sw         %[temp7],   28(%[dst])              \n\t"
+
+            : [dst] "+r" (dst), [src] "+r" (src),
+              [src0] "=&f" (src0), [src1] "=&f" (src1),
+              [src2] "=&f" (src2), [src3] "=&f" (src3),
+              [src4] "=&f" (src4), [src5] "=&f" (src5),
+              [src6] "=&f" (src6), [src7] "=&f" (src7),
+              [temp0] "=r" (temp0), [temp1] "=r" (temp1),
+              [temp2] "=r" (temp2), [temp3] "=r" (temp3),
+              [temp4] "=r" (temp4), [temp5] "=r" (temp5),
+              [temp6] "=r" (temp6), [temp7] "=r" (temp7)
+            : [scale] "f" (scale)
+            : "memory"
+        );
+        src = src + 8;
+        dst = dst + 8;
+        len -= 8;
+    } while (len > 0);
+}
+
+static void ac3_downmix_mips(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
+{
+    int i, j;
+    float v0, v1, v2, v3;
+    float v4, v5, v6, v7;
+    float samples0, samples1, samples2, samples3, matrix_j, matrix_j2;
+    float *samples_p,*matrix_p;
+    if (out_ch == 2) {
+        for (i = 0; i < len; i += 4) {
+            v0 = v1 = v2 = v3 = 0.0f;
+            v4 = v5 = v6 = v7 = 0.0f;
+            samples_p = &samples[0][i];
+            matrix_p = &matrix[0][0];
+            __asm__ volatile (
+                "move   %[j],           $zero                               \n\t"
+                "1:                                                         \n\t"
+                "lwc1   %[matrix_j],    0(%[matrix_p])                      \n\t"
+                "lwc1   %[matrix_j2],   4(%[matrix_p])                      \n\t"
+                "lwc1   %[samples0],    0(%[samples_p])                     \n\t"
+                "lwc1   %[samples1],    4(%[samples_p])                     \n\t"
+                "lwc1   %[samples2],    8(%[samples_p])                     \n\t"
+                "lwc1   %[samples3],    12(%[samples_p])                    \n\t"
+                "addiu  %[matrix_p],    8                                   \n\t"
+                "madd.s %[v0],          %[v0],  %[samples0],    %[matrix_j] \n\t"
+                "madd.s %[v1],          %[v1],  %[samples1],    %[matrix_j] \n\t"
+                "madd.s %[v2],          %[v2],  %[samples2],    %[matrix_j] \n\t"
+                "madd.s %[v3],          %[v3],  %[samples3],    %[matrix_j] \n\t"
+                "madd.s %[v4],          %[v4],  %[samples0],    %[matrix_j2]\n\t"
+                "madd.s %[v5],          %[v5],  %[samples1],    %[matrix_j2]\n\t"
+                "madd.s %[v6],          %[v6],  %[samples2],    %[matrix_j2]\n\t"
+                "madd.s %[v7],          %[v7],  %[samples3],    %[matrix_j2]\n\t"
+                "addiu  %[j],           1                                   \n\t"
+                "addiu  %[samples_p],   1024                                \n\t"
+                "bne    %[j],           %[in_ch],   1b                      \n\t"
+                :[samples0]"=&f"(samples0), [samples1]"=&f"(samples1),
+                 [samples2]"=&f"(samples2), [samples3]"=&f"(samples3),
+                 [samples_p]"+r"(samples_p), [matrix_j]"=&f"(matrix_j),
+                 [matrix_p]"+r"(matrix_p), [v0]"+f"(v0), [v1]"+f"(v1),
+                 [v2]"+f"(v2), [v3]"+f"(v3), [v4]"+f"(v4), [v5]"+f"(v5),
+                 [v6]"+f"(v6), [v7]"+f"(v7),[j]"=&r"(j), [matrix_j2]"=&f"(matrix_j2)
+                :[in_ch]"r"(in_ch)
+                :"memory"
+            );
+            samples[0][i  ] = v0;
+            samples[0][i+1] = v1;
+            samples[0][i+2] = v2;
+            samples[0][i+3] = v3;
+            samples[1][i  ] = v4;
+            samples[1][i+1] = v5;
+            samples[1][i+2] = v6;
+            samples[1][i+3] = v7;
+        }
+    } else if (out_ch == 1) {
+        for (i = 0; i < len; i += 4) {
+            v0 = v1 = v2 = v3 = 0.0f;
+            samples_p = &samples[0][i];
+            matrix_p = &matrix[0][0];
+            __asm__ volatile (
+                "move   %[j],           $zero                               \n\t"
+                "1:                                                         \n\t"
+                "lwc1   %[matrix_j],    0(%[matrix_p])                      \n\t"
+                "lwc1   %[samples0],    0(%[samples_p])                     \n\t"
+                "lwc1   %[samples1],    4(%[samples_p])                     \n\t"
+                "lwc1   %[samples2],    8(%[samples_p])                     \n\t"
+                "lwc1   %[samples3],    12(%[samples_p])                    \n\t"
+                "addiu  %[matrix_p],    8                                   \n\t"
+                "madd.s %[v0],          %[v0],  %[samples0],    %[matrix_j] \n\t"
+                "madd.s %[v1],          %[v1],  %[samples1],    %[matrix_j] \n\t"
+                "madd.s %[v2],          %[v2],  %[samples2],    %[matrix_j] \n\t"
+                "madd.s %[v3],          %[v3],  %[samples3],    %[matrix_j] \n\t"
+                "addiu  %[j],           1                                   \n\t"
+                "addiu  %[samples_p],   1024                                \n\t"
+                "bne    %[j],           %[in_ch],   1b                      \n\t"
+                :[samples0]"=&f"(samples0), [samples1]"=&f"(samples1),
+                 [samples2]"=&f"(samples2), [samples3]"=&f"(samples3),
+                 [samples_p]"+r"(samples_p), [matrix_j]"=&f"(matrix_j),
+                 [matrix_p]"+r"(matrix_p), [v0]"+f"(v0), [v1]"+f"(v1),
+                 [v2]"+f"(v2), [v3]"+f"(v3), [j]"=&r"(j)
+                :[in_ch]"r"(in_ch)
+                :"memory"
+            );
+            samples[0][i  ] = v0;
+            samples[0][i+1] = v1;
+            samples[0][i+2] = v2;
+            samples[0][i+3] = v3;
+        }
+    }
+}
+#endif
+#endif /* HAVE_INLINE_ASM */
+
+void ff_ac3dsp_init_mips(AC3DSPContext *c, int bit_exact) {
+#if HAVE_INLINE_ASM
+#if HAVE_MIPSDSPR1
+    c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_mips;
+    c->update_bap_counts  = ac3_update_bap_counts_mips;
+#endif
+#if HAVE_MIPSFPU
+    c->float_to_fixed24 = float_to_fixed24_mips;
+    c->downmix          = ac3_downmix_mips;
+#endif
+#endif
+
+}
index f5a8360..48f67c6 100644 (file)
@@ -50,5 +50,7 @@ void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
     ff_float_dsp_init_ppc(fdsp, bit_exact);
 #elif ARCH_X86
     ff_float_dsp_init_x86(fdsp);
+#elif ARCH_MIPS
+    ff_float_dsp_init_mips(fdsp);
 #endif
 }
index 735eb34..48a7847 100644 (file)
@@ -65,5 +65,6 @@ void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int strict);
 void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp);
 void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int strict);
 void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp);
+void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp);
 
 #endif /* AVUTIL_FLOAT_DSP_H */
diff --git a/libavutil/mips/Makefile b/libavutil/mips/Makefile
new file mode 100644 (file)
index 0000000..d8abef4
--- /dev/null
@@ -0,0 +1 @@
+MIPSFPU-OBJS-$(HAVE_INLINE_ASM) += mips/float_dsp_mips.o
diff --git a/libavutil/mips/float_dsp_mips.c b/libavutil/mips/float_dsp_mips.c
new file mode 100644 (file)
index 0000000..30ed20f
--- /dev/null
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2012
+ *      MIPS Technologies, Inc., California.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
+ *    contributors may be used to endorse or promote products derived from
+ *    this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * Author:  Branimir Vasic (bvasic@mips.com)
+ *
+ * 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
+ */
+
+/**
+ * @file
+ * Reference: libavutil/float_dsp.c
+ */
+
+#include "libavutil/float_dsp.h"
+
+static void vector_fmul_mips(float *dst, const float *src0, const float *src1,
+                             int len)
+{
+    int i;
+
+    if (len & 3) {
+        for (i = 0; i < len; i++)
+            dst[i] = src0[i] * src1[i];
+    } else {
+        float *d     = (float *)dst;
+        float *d_end = d + len;
+        float *s0    = (float *)src0;
+        float *s1    = (float *)src1;
+
+        float src0_0, src0_1, src0_2, src0_3;
+        float src1_0, src1_1, src1_2, src1_3;
+
+        __asm__ volatile (
+            "1:                                         \n\t"
+            "lwc1   %[src0_0],  0(%[s0])                \n\t"
+            "lwc1   %[src1_0],  0(%[s1])                \n\t"
+            "lwc1   %[src0_1],  4(%[s0])                \n\t"
+            "lwc1   %[src1_1],  4(%[s1])                \n\t"
+            "lwc1   %[src0_2],  8(%[s0])                \n\t"
+            "lwc1   %[src1_2],  8(%[s1])                \n\t"
+            "lwc1   %[src0_3],  12(%[s0])               \n\t"
+            "lwc1   %[src1_3],  12(%[s1])               \n\t"
+            "mul.s  %[src0_0],  %[src0_0],  %[src1_0]   \n\t"
+            "mul.s  %[src0_1],  %[src0_1],  %[src1_1]   \n\t"
+            "mul.s  %[src0_2],  %[src0_2],  %[src1_2]   \n\t"
+            "mul.s  %[src0_3],  %[src0_3],  %[src1_3]   \n\t"
+            "swc1   %[src0_0],  0(%[d])                 \n\t"
+            "swc1   %[src0_1],  4(%[d])                 \n\t"
+            "swc1   %[src0_2],  8(%[d])                 \n\t"
+            "swc1   %[src0_3],  12(%[d])                \n\t"
+            "addiu  %[s0],      %[s0],      16          \n\t"
+            "addiu  %[s1],      %[s1],      16          \n\t"
+            "addiu  %[d],       %[d],       16          \n\t"
+            "bne    %[d],       %[d_end],   1b          \n\t"
+
+            : [src0_0]"=&f"(src0_0), [src0_1]"=&f"(src0_1),
+              [src0_2]"=&f"(src0_2), [src0_3]"=&f"(src0_3),
+              [src1_0]"=&f"(src1_0), [src1_1]"=&f"(src1_1),
+              [src1_2]"=&f"(src1_2), [src1_3]"=&f"(src1_3),
+              [d]"+r"(d), [s0]"+r"(s0), [s1]"+r"(s1)
+            : [d_end]"r"(d_end)
+            : "memory"
+        );
+    }
+}
+
+void ff_float_dsp_init_mips(AVFloatDSPContext *fdsp) {
+    fdsp->vector_fmul = vector_fmul_mips;
+}