Quelle  compare_gcc.cc   Sprache: C

/*
 *  Copyright 2012 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.
 */

#include "libyuv/basic_types.h"

#include "libyuv/compare_row.h"
#include "libyuv/row.h"

#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif

// This module is for GCC x86 and x64.
#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__))

// "memory" clobber prevents the reads from being removed

#if defined(__x86_64__)
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
                               const uint8_t* src_b,
                               int count) {
  uint64_t diff;

  asm volatile(
      "xor %3,%3 \n"
      "xor %%r8,%%r8 \n"
      "xor %%r9,%%r9 \n"
      "xor %%r10,%%r10 \n"

      // Process 32 bytes per loop.
      LABELALIGN
      "1: \n"
      "mov (%0),%%rcx \n"
      "mov 0x8(%0),%%rdx \n"
      "xor (%1),%%rcx \n"
      "xor 0x8(%1),%%rdx \n"
      "popcnt %%rcx,%%rcx \n"
      "popcnt %%rdx,%%rdx \n"
      "mov 0x10(%0),%%rsi \n"
      "mov 0x18(%0),%%rdi \n"
      "xor 0x10(%1),%%rsi \n"
      "xor 0x18(%1),%%rdi \n"
      "popcnt %%rsi,%%rsi \n"
      "popcnt %%rdi,%%rdi \n"
      "add $0x20,%0 \n"
      "add $0x20,%1 \n"
      "add %%rcx,%3 \n"
      "add %%rdx,%%r8 \n"
      "add %%rsi,%%r9 \n"
      "add %%rdi,%%r10 \n"
      "sub $0x20,%2 \n"
      "jg 1b \n"

      "add %%r8, %3 \n"
      "add %%r9, %3 \n"
      "add %%r10, %3 \n"
      : "+r"(src_a),  // %0
        "+r"(src_b),  // %1
        "+r"(count),  // %2
        "=&r"(diff)   // %3
      :
      : "cc", "memory", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10");

  return (uint32_t)(diff);
}
#else
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
                               const uint8_t* src_b,
                               int count) {
  uint32_t diff = 0u;

  asm volatile(
      // Process 16 bytes per loop.
      LABELALIGN
      "1: \n"
      "mov (%0),%%ecx \n"
      "mov 0x4(%0),%%edx \n"
      "xor (%1),%%ecx \n"
      "xor 0x4(%1),%%edx \n"
      "popcnt %%ecx,%%ecx \n"
      "add %%ecx,%3 \n"
      "popcnt %%edx,%%edx \n"
      "add %%edx,%3 \n"
      "mov 0x8(%0),%%ecx \n"
      "mov 0xc(%0),%%edx \n"
      "xor 0x8(%1),%%ecx \n"
      "xor 0xc(%1),%%edx \n"
      "popcnt %%ecx,%%ecx \n"
      "add %%ecx,%3 \n"
      "popcnt %%edx,%%edx \n"
      "add %%edx,%3 \n"
      "add $0x10,%0 \n"
      "add $0x10,%1 \n"
      "sub $0x10,%2 \n"
      "jg 1b \n"
      : "+r"(src_a),  // %0
        "+r"(src_b),  // %1
        "+r"(count),  // %2
        "+r"(diff)    // %3
      :
      : "cc", "memory", "ecx", "edx");

  return diff;
}
#endif

static const vec8 kNibbleMask = {15, 15, 15, 15, 15, 15, 15, 15,
                                 15, 15, 15, 15, 15, 15, 15, 15};
static const vec8 kBitCount = {0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};

uint32_t HammingDistance_SSSE3(const uint8_t* src_a,
                               const uint8_t* src_b,
                               int count) {
  uint32_t diff;

  asm volatile(
      "movdqa %4,%%xmm2 \n"
      "movdqa %5,%%xmm3 \n"
      "pxor %%xmm0,%%xmm0 \n"
      "pxor %%xmm1,%%xmm1 \n"
      "sub %0,%1 \n"

      LABELALIGN
      "1: \n"
      "movdqa (%0),%%xmm4 \n"
      "movdqa 0x10(%0), %%xmm5 \n"
      "pxor (%0,%1), %%xmm4 \n"
      "movdqa %%xmm4,%%xmm6 \n"
      "pand %%xmm2,%%xmm6 \n"
      "psrlw $0x4,%%xmm4 \n"
      "movdqa %%xmm3,%%xmm7 \n"
      "pshufb %%xmm6,%%xmm7 \n"
      "pand %%xmm2,%%xmm4 \n"
      "movdqa %%xmm3,%%xmm6 \n"
      "pshufb %%xmm4,%%xmm6 \n"
      "paddb %%xmm7,%%xmm6 \n"
      "pxor 0x10(%0,%1),%%xmm5 \n"
      "add $0x20,%0 \n"
      "movdqa %%xmm5,%%xmm4 \n"
      "pand %%xmm2,%%xmm5 \n"
      "psrlw $0x4,%%xmm4 \n"
      "movdqa %%xmm3,%%xmm7 \n"
      "pshufb %%xmm5,%%xmm7 \n"
      "pand %%xmm2,%%xmm4 \n"
      "movdqa %%xmm3,%%xmm5 \n"
      "pshufb %%xmm4,%%xmm5 \n"
      "paddb %%xmm7,%%xmm5 \n"
      "paddb %%xmm5,%%xmm6 \n"
      "psadbw %%xmm1,%%xmm6 \n"
      "paddd %%xmm6,%%xmm0 \n"
      "sub $0x20,%2 \n"
      "jg 1b \n"

      "pshufd $0xaa,%%xmm0,%%xmm1 \n"
      "paddd %%xmm1,%%xmm0 \n"
      "movd %%xmm0, %3 \n"
      : "+r"(src_a),       // %0
        "+r"(src_b),       // %1
        "+r"(count),       // %2
        "=r"(diff)         // %3
      : "m"(kNibbleMask),  // %4
        "m"(kBitCount)     // %5
      : "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");

  return diff;
}

#ifdef HAS_HAMMINGDISTANCE_AVX2
uint32_t HammingDistance_AVX2(const uint8_t* src_a,
                              const uint8_t* src_b,
                              int count) {
  uint32_t diff;

  asm volatile(
      "vbroadcastf128 %4,%%ymm2 \n"
      "vbroadcastf128 %5,%%ymm3 \n"
      "vpxor %%ymm0,%%ymm0,%%ymm0 \n"
      "vpxor %%ymm1,%%ymm1,%%ymm1 \n"
      "sub %0,%1 \n"

      LABELALIGN
      "1: \n"
      "vmovdqa (%0),%%ymm4 \n"
      "vmovdqa 0x20(%0), %%ymm5 \n"
      "vpxor (%0,%1), %%ymm4, %%ymm4 \n"
      "vpand %%ymm2,%%ymm4,%%ymm6 \n"
      "vpsrlw $0x4,%%ymm4,%%ymm4 \n"
      "vpshufb %%ymm6,%%ymm3,%%ymm6 \n"
      "vpand %%ymm2,%%ymm4,%%ymm4 \n"
      "vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
      "vpaddb %%ymm4,%%ymm6,%%ymm6 \n"
      "vpxor 0x20(%0,%1),%%ymm5,%%ymm4 \n"
      "add $0x40,%0 \n"
      "vpand %%ymm2,%%ymm4,%%ymm5 \n"
      "vpsrlw $0x4,%%ymm4,%%ymm4 \n"
      "vpshufb %%ymm5,%%ymm3,%%ymm5 \n"
      "vpand %%ymm2,%%ymm4,%%ymm4 \n"
      "vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
      "vpaddb %%ymm5,%%ymm4,%%ymm4 \n"
      "vpaddb %%ymm6,%%ymm4,%%ymm4 \n"
      "vpsadbw %%ymm1,%%ymm4,%%ymm4 \n"
      "vpaddd %%ymm0,%%ymm4,%%ymm0 \n"
      "sub $0x40,%2 \n"
      "jg 1b \n"

      "vpermq $0xb1,%%ymm0,%%ymm1 \n"
      "vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
      "vpermq $0xaa,%%ymm0,%%ymm1 \n"
      "vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
      "vmovd %%xmm0,%3 \n"
      "vzeroupper \n"
      : "+r"(src_a),       // %0
        "+r"(src_b),       // %1
        "+r"(count),       // %2
        "=r"(diff)         // %3
      : "m"(kNibbleMask),  // %4
        "m"(kBitCount)     // %5
      : "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6");

  return diff;
}
#endif  // HAS_HAMMINGDISTANCE_AVX2

uint32_t SumSquareError_SSE2(const uint8_t* src_a,
                             const uint8_t* src_b,
                             int count) {
  uint32_t sse;
  asm volatile(
      "pxor %%xmm0,%%xmm0 \n"
      "pxor %%xmm5,%%xmm5 \n"

      LABELALIGN
      "1: \n"
      "movdqu (%0),%%xmm1 \n"
      "lea 0x10(%0),%0 \n"
      "movdqu (%1),%%xmm2 \n"
      "lea 0x10(%1),%1 \n"
      "movdqa %%xmm1,%%xmm3 \n"
      "psubusb %%xmm2,%%xmm1 \n"
      "psubusb %%xmm3,%%xmm2 \n"
      "por %%xmm2,%%xmm1 \n"
      "movdqa %%xmm1,%%xmm2 \n"
      "punpcklbw %%xmm5,%%xmm1 \n"
      "punpckhbw %%xmm5,%%xmm2 \n"
      "pmaddwd %%xmm1,%%xmm1 \n"
      "pmaddwd %%xmm2,%%xmm2 \n"
      "paddd %%xmm1,%%xmm0 \n"
      "paddd %%xmm2,%%xmm0 \n"
      "sub $0x10,%2 \n"
      "jg 1b \n"

      "pshufd $0xee,%%xmm0,%%xmm1 \n"
      "paddd %%xmm1,%%xmm0 \n"
      "pshufd $0x1,%%xmm0,%%xmm1 \n"
      "paddd %%xmm1,%%xmm0 \n"
      "movd %%xmm0,%3 \n"
      : "+r"(src_a),  // %0
        "+r"(src_b),  // %1
        "+r"(count),  // %2
        "=r"(sse)     // %3
      :
      : "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
  return sse;
}

static const uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0};  // 33 ^ 16
static const uvec32 kHashMul0 = {
    0x0c3525e1,  // 33 ^ 15
    0xa3476dc1,  // 33 ^ 14
    0x3b4039a1,  // 33 ^ 13
    0x4f5f0981,  // 33 ^ 12
};
static const uvec32 kHashMul1 = {
    0x30f35d61,  // 33 ^ 11
    0x855cb541,  // 33 ^ 10
    0x040a9121,  // 33 ^ 9
    0x747c7101,  // 33 ^ 8
};
static const uvec32 kHashMul2 = {
    0xec41d4e1,  // 33 ^ 7
    0x4cfa3cc1,  // 33 ^ 6
    0x025528a1,  // 33 ^ 5
    0x00121881,  // 33 ^ 4
};
static const uvec32 kHashMul3 = {
    0x00008c61,  // 33 ^ 3
    0x00000441,  // 33 ^ 2
    0x00000021,  // 33 ^ 1
    0x00000001,  // 33 ^ 0
};

uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) {
  uint32_t hash;
  asm volatile(
      "movd %2,%%xmm0 \n"
      "pxor %%xmm7,%%xmm7 \n"
      "movdqa %4,%%xmm6 \n"

      LABELALIGN
      "1: \n"
      "movdqu (%0),%%xmm1 \n"
      "lea 0x10(%0),%0 \n"
      "pmulld %%xmm6,%%xmm0 \n"
      "movdqa %5,%%xmm5 \n"
      "movdqa %%xmm1,%%xmm2 \n"
      "punpcklbw %%xmm7,%%xmm2 \n"
      "movdqa %%xmm2,%%xmm3 \n"
      "punpcklwd %%xmm7,%%xmm3 \n"
      "pmulld %%xmm5,%%xmm3 \n"
      "movdqa %6,%%xmm5 \n"
      "movdqa %%xmm2,%%xmm4 \n"
      "punpckhwd %%xmm7,%%xmm4 \n"
      "pmulld %%xmm5,%%xmm4 \n"
      "movdqa %7,%%xmm5 \n"
      "punpckhbw %%xmm7,%%xmm1 \n"
      "movdqa %%xmm1,%%xmm2 \n"
      "punpcklwd %%xmm7,%%xmm2 \n"
      "pmulld %%xmm5,%%xmm2 \n"
      "movdqa %8,%%xmm5 \n"
      "punpckhwd %%xmm7,%%xmm1 \n"
      "pmulld %%xmm5,%%xmm1 \n"
      "paddd %%xmm4,%%xmm3 \n"
      "paddd %%xmm2,%%xmm1 \n"
      "paddd %%xmm3,%%xmm1 \n"
      "pshufd $0xe,%%xmm1,%%xmm2 \n"
      "paddd %%xmm2,%%xmm1 \n"
      "pshufd $0x1,%%xmm1,%%xmm2 \n"
      "paddd %%xmm2,%%xmm1 \n"
      "paddd %%xmm1,%%xmm0 \n"
      "sub $0x10,%1 \n"
      "jg 1b \n"
      "movd %%xmm0,%3 \n"
      : "+r"(src),        // %0
        "+r"(count),      // %1
        "+rm"(seed),      // %2
        "=r"(hash)        // %3
      : "m"(kHash16x33),  // %4
        "m"(kHashMul0),   // %5
        "m"(kHashMul1),   // %6
        "m"(kHashMul2),   // %7
        "m"(kHashMul3)    // %8
      : "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");
  return hash;
}
#endif  // defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))

#ifdef __cplusplus
}  // extern "C"
}  // namespace libyuv
#endif