From 97f074f134048276db2f2c552e6e6b24fe0a6894 Mon Sep 17 00:00:00 2001 From: Shivraj Patil Date: Wed, 22 Apr 2015 14:34:44 +0530 Subject: avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for HEVC copy and hv mc functions Incorporated review comment. Removed "__" from volatile. Signed-off-by: Shivraj Patil Reviewed-by: Nedeljko Babic Signed-off-by: Michael Niedermayer --- libavutil/mips/generic_macros_msa.h | 133 ++++++++++++++++++++++++++++++++++++ 1 file changed, 133 insertions(+) (limited to 'libavutil/mips/generic_macros_msa.h') diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h index 374382ff7e..8b945ec5dd 100644 --- a/libavutil/mips/generic_macros_msa.h +++ b/libavutil/mips/generic_macros_msa.h @@ -50,7 +50,25 @@ *((v8i16 *) (pdest)) = (vec); \ } +#define STORE_SW(vec, pdest) \ +{ \ + *((v4i32 *) (pdest)) = (vec); \ +} + #if (__mips_isa_rev >= 6) + #define STORE_WORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint32_t val_m = (val); \ + \ + __asm__ volatile ( \ + "sw %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #define STORE_DWORD(pdst, val) \ { \ uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ @@ -64,6 +82,19 @@ ); \ } #else + #define STORE_WORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint32_t val_m = (val); \ + \ + __asm__ volatile ( \ + "usw %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #define STORE_DWORD(pdst, val) \ { \ uint8_t *dst1_m = (uint8_t *) (pdst); \ @@ -83,6 +114,13 @@ } #endif +#define LOAD_2VECS_SB(psrc, stride, \ + val0, val1) \ +{ \ + val0 = LOAD_SB(psrc + 0 * stride); \ + val1 = LOAD_SB(psrc + 1 * stride); \ +} + #define LOAD_4VECS_SB(psrc, stride, \ val0, val1, val2, val3) \ { \ @@ -92,6 +130,15 @@ val3 = LOAD_SB(psrc + 3 * stride); \ } +#define LOAD_6VECS_SB(psrc, stride, \ + out0, out1, out2, out3, out4, out5) \ +{ \ + LOAD_4VECS_SB((psrc), (stride), \ + (out0), (out1), (out2), (out3)); \ + LOAD_2VECS_SB((psrc + 4 * stride), (stride), \ + (out4), (out5)); \ +} + #define LOAD_7VECS_SB(psrc, stride, \ val0, val1, val2, val3, \ val4, val5, val6) \ @@ -115,6 +162,48 @@ (out4), (out5), (out6), (out7)); \ } +#define STORE_2VECS_SH(ptr, stride, \ + in0, in1) \ +{ \ + STORE_SH(in0, ((ptr) + 0 * stride)); \ + STORE_SH(in1, ((ptr) + 1 * stride)); \ +} + +#define STORE_4VECS_SH(ptr, stride, \ + in0, in1, in2, in3) \ +{ \ + STORE_SH(in0, ((ptr) + 0 * stride)); \ + STORE_SH(in1, ((ptr) + 1 * stride)); \ + STORE_SH(in2, ((ptr) + 2 * stride)); \ + STORE_SH(in3, ((ptr) + 3 * stride)); \ +} + +#define STORE_6VECS_SH(ptr, stride, \ + in0, in1, in2, in3, \ + in4, in5) \ +{ \ + STORE_SH(in0, ((ptr) + 0 * stride)); \ + STORE_SH(in1, ((ptr) + 1 * stride)); \ + STORE_SH(in2, ((ptr) + 2 * stride)); \ + STORE_SH(in3, ((ptr) + 3 * stride)); \ + STORE_SH(in4, ((ptr) + 4 * stride)); \ + STORE_SH(in5, ((ptr) + 5 * stride)); \ +} + +#define STORE_8VECS_SH(ptr, stride, \ + in0, in1, in2, in3, \ + in4, in5, in6, in7) \ +{ \ + STORE_SH(in0, ((ptr) + 0 * stride)); \ + STORE_SH(in1, ((ptr) + 1 * stride)); \ + STORE_SH(in2, ((ptr) + 2 * stride)); \ + STORE_SH(in3, ((ptr) + 3 * stride)); \ + STORE_SH(in4, ((ptr) + 4 * stride)); \ + STORE_SH(in5, ((ptr) + 5 * stride)); \ + STORE_SH(in6, ((ptr) + 6 * stride)); \ + STORE_SH(in7, ((ptr) + 7 * stride)); \ +} + #define ILVR_B_2VECS_SB(in0_r, in1_r, in0_l, in1_l, \ out0, out1) \ { \ @@ -164,6 +253,28 @@ out6, out7); \ } +#define ILVR_H_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1) \ +{ \ + out0 = __msa_ilvr_h((v8i16) (in0_l), (v8i16) (in0_r)); \ + out1 = __msa_ilvr_h((v8i16) (in1_l), (v8i16) (in1_r)); \ +} + +#define ILVR_H_6VECS_SH(in0_r, in1_r, in2_r, \ + in3_r, in4_r, in5_r, \ + in0_l, in1_l, in2_l, \ + in3_l, in4_l, in5_l, \ + out0, out1, out2, \ + out3, out4, out5) \ +{ \ + ILVR_H_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1); \ + ILVR_H_2VECS_SH(in2_r, in3_r, in2_l, in3_l, \ + out2, out3); \ + ILVR_H_2VECS_SH(in4_r, in5_r, in4_l, in5_l, \ + out4, out5); \ +} + #define ILVL_B_2VECS_SB(in0_r, in1_r, in0_l, in1_l, \ out0, out1) \ { \ @@ -196,6 +307,28 @@ out4, out5); \ } +#define ILVL_H_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1) \ +{ \ + out0 = __msa_ilvl_h((v8i16) (in0_l), (v8i16) (in0_r)); \ + out1 = __msa_ilvl_h((v8i16) (in1_l), (v8i16) (in1_r)); \ +} + +#define ILVL_H_6VECS_SH(in0_r, in1_r, in2_r, \ + in3_r, in4_r, in5_r, \ + in0_l, in1_l, in2_l, \ + in3_l, in4_l, in5_l, \ + out0, out1, out2, \ + out3, out4, out5) \ +{ \ + ILVL_H_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1); \ + ILVL_H_2VECS_SH(in2_r, in3_r, in2_l, in3_l, \ + out2, out3); \ + ILVL_H_2VECS_SH(in4_r, in5_r, in4_l, in5_l, \ + out4, out5); \ +} + #define ILVR_D_2VECS_SB(out0, in0_l, in0_r, \ out1, in1_l, in1_r) \ { \ -- cgit v1.2.3