Quelle  macros_msa.h   Sprache: C

/*
 *  Copyright 2016 The LibYuv 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.
 */

#ifndef INCLUDE_LIBYUV_MACROS_MSA_H_
#define INCLUDE_LIBYUV_MACROS_MSA_H_

#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include <msa.h>
#include <stdint.h>

#if (__mips_isa_rev >= 6)
#define LW(psrc)                                       \
  ({                                                   \
    const uint8_t* psrc_lw_m = (const uint8_t*)(psrc); \
    uint32_t val_m;                                    \
    asm("lw %[val_m], %[psrc_lw_m] \n"              \
        : [val_m] "=r"(val_m)                          \
        : [psrc_lw_m] "m"(*psrc_lw_m));                \
    val_m;                                             \
  })

#if (__mips == 64)
#define LD(psrc)                                       \
  ({                                                   \
    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
    uint64_t val_m = 0;                                \
    asm("ld %[val_m], %[psrc_ld_m] \n"              \
        : [val_m] "=r"(val_m)                          \
        : [psrc_ld_m] "m"(*psrc_ld_m));                \
    val_m;                                             \
  })
#else  // !(__mips == 64)
#define LD(psrc)                                                         \
  ({                                                                     \
    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc);                   \
    uint32_t val0_m, val1_m;                                             \
    uint64_t val_m = 0;                                                  \
    val0_m = LW(psrc_ld_m);                                              \
    val1_m = LW(psrc_ld_m + 4);                                          \
    val_m = (uint64_t)(val1_m);                             /* NOLINT */ \
    val_m = (uint64_t)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
    val_m = (uint64_t)(val_m | (uint64_t)val0_m);           /* NOLINT */ \
    val_m;                                                               \
  })
#endif  // (__mips == 64)

#define SW(val, pdst)                                   \
  ({                                                    \
    uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
    uint32_t val_m = (val);                             \
    asm("sw %[val_m], %[pdst_sw_m] \n"               \
        : [pdst_sw_m] "=m"(*pdst_sw_m)                  \
        : [val_m] "r"(val_m));                          \
  })

#if (__mips == 64)
#define SD(val, pdst)                                   \
  ({                                                    \
    uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
    uint64_t val_m = (val);                             \
    asm("sd %[val_m], %[pdst_sd_m] \n"               \
        : [pdst_sd_m] "=m"(*pdst_sd_m)                  \
        : [val_m] "r"(val_m));                          \
  })
#else  // !(__mips == 64)
#define SD(val, pdst)                                        \
  ({                                                         \
    uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */      \
    uint32_t val0_m, val1_m;                                 \
    val0_m = (uint32_t)((val)&0x00000000FFFFFFFF);           \
    val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
    SW(val0_m, pdst_sd_m);                                   \
    SW(val1_m, pdst_sd_m + 4);                               \
  })
#endif  // !(__mips == 64)
#else   // !(__mips_isa_rev >= 6)
#define LW(psrc)                                \
  ({                                            \
    uint8_t* psrc_lw_m = (uint8_t*)(psrc);      \
    uint32_t val_lw_m;                          \
                                                \
    asm("lwr %[val_lw_m], 0(%[psrc_lw_m]) \n\t" \
        "lwl %[val_lw_m], 3(%[psrc_lw_m]) \n\t" \
                                                \
        : [val_lw_m] "=&r"(val_lw_m)            \
        : [psrc_lw_m] "r"(psrc_lw_m));          \
                                                \
    val_lw_m;                                   \
  })

#if (__mips == 64)
#define LD(psrc)                                \
  ({                                            \
    uint8_t* psrc_ld_m = (uint8_t*)(psrc);      \
    uint64_t val_ld_m = 0;                      \
                                                \
    asm("ldr %[val_ld_m], 0(%[psrc_ld_m]) \n\t" \
        "ldl %[val_ld_m], 7(%[psrc_ld_m]) \n\t" \
                                                \
        : [val_ld_m] "=&r"(val_ld_m)            \
        : [psrc_ld_m] "r"(psrc_ld_m));          \
                                                \
    val_ld_m;                                   \
  })
#else  // !(__mips == 64)
#define LD(psrc)                                                         \
  ({                                                                     \
    const uint8_t* psrc_ld_m = (const uint8_t*)(psrc);                   \
    uint32_t val0_m, val1_m;                                             \
    uint64_t val_m = 0;                                                  \
    val0_m = LW(psrc_ld_m);                                              \
    val1_m = LW(psrc_ld_m + 4);                                          \
    val_m = (uint64_t)(val1_m);                             /* NOLINT */ \
    val_m = (uint64_t)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
    val_m = (uint64_t)(val_m | (uint64_t)val0_m);           /* NOLINT */ \
    val_m;                                                               \
  })
#endif  // (__mips == 64)

#define SW(val, pdst)                                   \
  ({                                                    \
    uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
    uint32_t val_m = (val);                             \
    asm("usw %[val_m], %[pdst_sw_m] \n"              \
        : [pdst_sw_m] "=m"(*pdst_sw_m)                  \
        : [val_m] "r"(val_m));                          \
  })

#define SD(val, pdst)                                        \
  ({                                                         \
    uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */      \
    uint32_t val0_m, val1_m;                                 \
    val0_m = (uint32_t)((val)&0x00000000FFFFFFFF);           \
    val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
    SW(val0_m, pdst_sd_m);                                   \
    SW(val1_m, pdst_sd_m + 4);                               \
  })
#endif  // (__mips_isa_rev >= 6)

// TODO(fbarchard): Consider removing __VAR_ARGS versions.
#define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
#define LD_UB(...) LD_B(const v16u8, __VA_ARGS__)

#define LD_H(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
#define LD_UH(...) LD_H(const v8u16, __VA_ARGS__)

#define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)

#define ST_H(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UH(...) ST_H(v8u16, __VA_ARGS__)

/* Description : Load two vectors with 16 'byte' sized elements
   Arguments   : Inputs  - psrc, stride
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Load 16 byte elements in 'out0' from (psrc)
                 Load 16 byte elements in 'out1' from (psrc + stride)
*/
#define LD_B2(RTYPE, psrc, stride, out0, out1) \
  {                                            \
    out0 = LD_B(RTYPE, (psrc));                \
    out1 = LD_B(RTYPE, (psrc) + stride);       \
  }
#define LD_UB2(...) LD_B2(const v16u8, __VA_ARGS__)

#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) \
  {                                                        \
    LD_B2(RTYPE, (psrc), stride, out0, out1);              \
    LD_B2(RTYPE, (psrc) + 2 * stride, stride, out2, out3); \
  }
#define LD_UB4(...) LD_B4(const v16u8, __VA_ARGS__)

/* Description : Store two vectors with stride each having 16 'byte' sized
                 elements
   Arguments   : Inputs - in0, in1, pdst, stride
   Details     : Store 16 byte elements from 'in0' to (pdst)
                 Store 16 byte elements from 'in1' to (pdst + stride)
*/
#define ST_B2(RTYPE, in0, in1, pdst, stride) \
  {                                          \
    ST_B(RTYPE, in0, (pdst));                \
    ST_B(RTYPE, in1, (pdst) + stride);       \
  }
#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)

#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride)   \
  {                                                      \
    ST_B2(RTYPE, in0, in1, (pdst), stride);              \
    ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
  }
#define ST_UB4(...) ST_B4(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)
                 Store 8 halfword elements from 'in1' to (pdst + stride)
*/
#define ST_H2(RTYPE, in0, in1, pdst, stride) \
  {                                          \
    ST_H(RTYPE, in0, (pdst));                \
    ST_H(RTYPE, in1, (pdst) + stride);       \
  }
#define ST_UH2(...) ST_H2(v8u16, __VA_ARGS__)

// TODO(fbarchard): Consider using __msa_vshf_b and __msa_ilvr_b directly.
/* Description : Shuffle byte vector elements as per mask vector
   Arguments   : Inputs  - in0, in1, in2, in3, mask0, mask1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Byte elements from 'in0' & 'in1' are copied selectively to
                 'out0' as per control vector 'mask0'
*/
#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1)  \
  {                                                                   \
    out0 = (RTYPE)__msa_vshf_b((v16i8)mask0, (v16i8)in1, (v16i8)in0); \
    out1 = (RTYPE)__msa_vshf_b((v16i8)mask1, (v16i8)in3, (v16i8)in2); \
  }
#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)

/* Description : Interleave both left and right half of input vectors
   Arguments   : Inputs  - in0, in1
                 Outputs - out0, out1
                 Return Type - as per RTYPE
   Details     : Right half of byte elements from 'in0' and 'in1' are
                 interleaved and written to 'out0'
*/
#define ILVRL_B2(RTYPE, in0, in1, out0, out1)           \
  {                                                     \
    out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
    out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
  }
#define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)

#endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */

#endif  // INCLUDE_LIBYUV_MACROS_MSA_H_