mips msa vp8 copy mem optimization

average improvement ~2x-4x

Change-Id: I3af3ecced96c5b8e0cb811256e5089e28fe013a2

4 files changed, 100 insertions, 3 deletions

diff --git a/vp8/common/mips/msa/copymem_msa.c b/vp8/common/mips/msa/copymem_msa.c
new file mode 100644
index 000000000..002a5ed91
--- /dev/null
+++ b/vp8/common/mips/msa/copymem_msa.c
@@ -0,0 +1,70 @@
+/*
+ *  Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "./vp8_rtcd.h"
+#include "vp8/common/mips/msa/vp8_macros_msa.h"
+
+static void copy_8x4_msa(uint8_t *src, int32_t src_stride,
+                         uint8_t *dst, int32_t dst_stride)
+{
+    uint64_t src0, src1, src2, src3;
+
+    LD4(src, src_stride, src0, src1, src2, src3);
+    SD4(src0, src1, src2, src3, dst, dst_stride);
+}
+
+static void copy_8x8_msa(uint8_t *src, int32_t src_stride,
+                         uint8_t *dst, int32_t dst_stride)
+{
+    uint64_t src0, src1, src2, src3;
+
+    LD4(src, src_stride, src0, src1, src2, src3);
+    src += (4 * src_stride);
+    SD4(src0, src1, src2, src3, dst, dst_stride);
+    dst += (4 * dst_stride);
+
+    LD4(src, src_stride, src0, src1, src2, src3);
+    SD4(src0, src1, src2, src3, dst, dst_stride);
+}
+
+static void copy_16x16_msa(uint8_t *src, int32_t src_stride,
+                           uint8_t *dst, int32_t dst_stride)
+{
+    v16u8 src0, src1, src2, src3, src4, src5, src6, src7;
+    v16u8 src8, src9, src10, src11, src12, src13, src14, src15;
+
+    LD_UB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7);
+    src += (8 * src_stride);
+    LD_UB8(src, src_stride, src8, src9, src10, src11, src12, src13, src14,
+           src15);
+
+    ST_UB8(src0, src1, src2, src3, src4, src5, src6, src7, dst, dst_stride);
+    dst += (8 * dst_stride);
+    ST_UB8(src8, src9, src10, src11, src12, src13, src14, src15, dst,
+           dst_stride);
+}
+
+void vp8_copy_mem16x16_msa(uint8_t *src, int32_t src_stride,
+                           uint8_t *dst, int32_t dst_stride)
+{
+    copy_16x16_msa(src, src_stride, dst, dst_stride);
+}
+
+void vp8_copy_mem8x8_msa(uint8_t *src, int32_t src_stride,
+                         uint8_t *dst, int32_t dst_stride)
+{
+    copy_8x8_msa(src, src_stride, dst, dst_stride);
+}
+
+void vp8_copy_mem8x4_msa(uint8_t *src, int32_t src_stride,
+                         uint8_t *dst, int32_t dst_stride)
+{
+    copy_8x4_msa(src, src_stride, dst, dst_stride);
+}
diff --git a/vp8/common/mips/msa/vp8_macros_msa.h b/vp8/common/mips/msa/vp8_macros_msa.h
index 8170ecf1d..46980900a 100644
--- a/vp8/common/mips/msa/vp8_macros_msa.h
+++ b/vp8/common/mips/msa/vp8_macros_msa.h
@@ -222,6 +222,23 @@
     out3 = LW((psrc) + 3 * stride);                \
 }
 
+/* Description : Load double words with stride
+   Arguments   : Inputs  - psrc, stride
+                 Outputs - out0, out1
+   Details     : Load double word in 'out0' from (psrc)
+                 Load double word in 'out1' from (psrc + stride)
+*/
+#define LD2(psrc, stride, out0, out1)  \
+{                                      \
+    out0 = LD((psrc));                 \
+    out1 = LD((psrc) + stride);        \
+}
+#define LD4(psrc, stride, out0, out1, out2, out3)  \
+{                                                  \
+    LD2((psrc), stride, out0, out1);               \
+    LD2((psrc) + 2 * stride, stride, out2, out3);  \
+}
+
 /* Description : Store 4 words with stride
    Arguments   : Inputs - in0, in1, in2, in3, pdst, stride
    Details     : Store word from 'in0' to (pdst)
@@ -298,6 +315,7 @@
     LD_B4(RTYPE, (psrc) + 4 * stride, stride, out4, out5, out6, out7);  \
 }
 #define LD_UB8(...) LD_B8(v16u8, __VA_ARGS__)
+#define LD_SB8(...) LD_B8(v16i8, __VA_ARGS__)
 
 /* Description : Load vectors with 8 halfword elements with stride
    Arguments   : Inputs  - psrc, stride
@@ -339,6 +357,14 @@
 #define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
 #define ST_SB4(...) ST_B4(v16i8, __VA_ARGS__)
 
+#define ST_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,        \
+              pdst, stride)                                         \
+{                                                                   \
+    ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride);                 \
+    ST_B4(RTYPE, in4, in5, in6, in7, (pdst) + 4 * stride, stride);  \
+}
+#define ST_UB8(...) ST_B8(v16u8, __VA_ARGS__)
+
 /* Description : Store vectors of 8 halfword elements with stride
    Arguments   : Inputs - in0, in1, pdst, stride
    Details     : Store 8 halfword elements from 'in0' to (pdst)
diff --git a/vp8/common/rtcd_defs.pl b/vp8/common/rtcd_defs.pl
index 3a278f7a8..aa47c0fa9 100644
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -138,17 +138,17 @@ $vp8_dc_only_idct_add_dspr2=vp8_dc_only_idct_add_dspr2;
 # RECON
 #
 add_proto qw/void vp8_copy_mem16x16/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_copy_mem16x16 mmx sse2 media neon dspr2/;
+specialize qw/vp8_copy_mem16x16 mmx sse2 media neon dspr2 msa/;
 $vp8_copy_mem16x16_media=vp8_copy_mem16x16_v6;
 $vp8_copy_mem16x16_dspr2=vp8_copy_mem16x16_dspr2;
 
 add_proto qw/void vp8_copy_mem8x8/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_copy_mem8x8 mmx media neon dspr2/;
+specialize qw/vp8_copy_mem8x8 mmx media neon dspr2 msa/;
 $vp8_copy_mem8x8_media=vp8_copy_mem8x8_v6;
 $vp8_copy_mem8x8_dspr2=vp8_copy_mem8x8_dspr2;
 
 add_proto qw/void vp8_copy_mem8x4/, "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch";
-specialize qw/vp8_copy_mem8x4 mmx media neon dspr2/;
+specialize qw/vp8_copy_mem8x4 mmx media neon dspr2 msa/;
 $vp8_copy_mem8x4_media=vp8_copy_mem8x4_v6;
 $vp8_copy_mem8x4_dspr2=vp8_copy_mem8x4_dspr2;
 
diff --git a/vp8/vp8_common.mk b/vp8/vp8_common.mk
index 8248cb70c..3e390953c 100644
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -114,6 +114,7 @@ VP8_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/idct_blk_dspr2.c
 VP8_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/dequantize_dspr2.c
 
 # common (c)
+VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/copymem_msa.c
 VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/idct_msa.c
 VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/loopfilter_filters_msa.c
 VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/sixtap_filter_msa.c