Quelle  rotate_gcc.cc   Sprache: C

/*
 *  Copyright 2015 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/rotate_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__) && !defined(_MSC_VER)))

// Transpose 8x8. 32 or 64 bit, but not NaCL for 64 bit.
#if defined(HAS_TRANSPOSEWX8_SSSE3)
void TransposeWx8_SSSE3(const uint8_t* src,
                        int src_stride,
                        uint8_t* dst,
                        int dst_stride,
                        int width) {
  asm volatile(
      // Read in the data from the source pointer.
      // First round of bit swap.
      LABELALIGN
      "1: \n"
      "movq (%0),%%xmm0 \n"
      "movq (%0,%3),%%xmm1 \n"
      "lea (%0,%3,2),%0 \n"
      "punpcklbw %%xmm1,%%xmm0 \n"
      "movq (%0),%%xmm2 \n"
      "movdqa %%xmm0,%%xmm1 \n"
      "palignr $0x8,%%xmm1,%%xmm1 \n"
      "movq (%0,%3),%%xmm3 \n"
      "lea (%0,%3,2),%0 \n"
      "punpcklbw %%xmm3,%%xmm2 \n"
      "movdqa %%xmm2,%%xmm3 \n"
      "movq (%0),%%xmm4 \n"
      "palignr $0x8,%%xmm3,%%xmm3 \n"
      "movq (%0,%3),%%xmm5 \n"
      "lea (%0,%3,2),%0 \n"
      "punpcklbw %%xmm5,%%xmm4 \n"
      "movdqa %%xmm4,%%xmm5 \n"
      "movq (%0),%%xmm6 \n"
      "palignr $0x8,%%xmm5,%%xmm5 \n"
      "movq (%0,%3),%%xmm7 \n"
      "lea (%0,%3,2),%0 \n"
      "punpcklbw %%xmm7,%%xmm6 \n"
      "neg %3 \n"
      "movdqa %%xmm6,%%xmm7 \n"
      "lea 0x8(%0,%3,8),%0 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      "neg %3 \n"
      // Second round of bit swap.
      "punpcklwd %%xmm2,%%xmm0 \n"
      "punpcklwd %%xmm3,%%xmm1 \n"
      "movdqa %%xmm0,%%xmm2 \n"
      "movdqa %%xmm1,%%xmm3 \n"
      "palignr $0x8,%%xmm2,%%xmm2 \n"
      "palignr $0x8,%%xmm3,%%xmm3 \n"
      "punpcklwd %%xmm6,%%xmm4 \n"
      "punpcklwd %%xmm7,%%xmm5 \n"
      "movdqa %%xmm4,%%xmm6 \n"
      "movdqa %%xmm5,%%xmm7 \n"
      "palignr $0x8,%%xmm6,%%xmm6 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      // Third round of bit swap.
      // Write to the destination pointer.
      "punpckldq %%xmm4,%%xmm0 \n"
      "movq %%xmm0,(%1) \n"
      "movdqa %%xmm0,%%xmm4 \n"
      "palignr $0x8,%%xmm4,%%xmm4 \n"
      "movq %%xmm4,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm6,%%xmm2 \n"
      "movdqa %%xmm2,%%xmm6 \n"
      "movq %%xmm2,(%1) \n"
      "palignr $0x8,%%xmm6,%%xmm6 \n"
      "punpckldq %%xmm5,%%xmm1 \n"
      "movq %%xmm6,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "movdqa %%xmm1,%%xmm5 \n"
      "movq %%xmm1,(%1) \n"
      "palignr $0x8,%%xmm5,%%xmm5 \n"
      "movq %%xmm5,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm7,%%xmm3 \n"
      "movq %%xmm3,(%1) \n"
      "movdqa %%xmm3,%%xmm7 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      "sub $0x8,%2 \n"
      "movq %%xmm7,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "jg 1b \n"
      : "+r"(src),                    // %0
        "+r"(dst),                    // %1
        "+r"(width)                   // %2
      : "r"((intptr_t)(src_stride)),  // %3
        "r"((intptr_t)(dst_stride))   // %4
      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");
}
#endif  // defined(HAS_TRANSPOSEWX8_SSSE3)

// Transpose 16x8. 64 bit
#if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
void TransposeWx8_Fast_SSSE3(const uint8_t* src,
                             int src_stride,
                             uint8_t* dst,
                             int dst_stride,
                             int width) {
  asm volatile(
      // Read in the data from the source pointer.
      // First round of bit swap.
      LABELALIGN
      "1: \n"
      "movdqu (%0),%%xmm0 \n"
      "movdqu (%0,%3),%%xmm1 \n"
      "lea (%0,%3,2),%0 \n"
      "movdqa %%xmm0,%%xmm8 \n"
      "punpcklbw %%xmm1,%%xmm0 \n"
      "punpckhbw %%xmm1,%%xmm8 \n"
      "movdqu (%0),%%xmm2 \n"
      "movdqa %%xmm0,%%xmm1 \n"
      "movdqa %%xmm8,%%xmm9 \n"
      "palignr $0x8,%%xmm1,%%xmm1 \n"
      "palignr $0x8,%%xmm9,%%xmm9 \n"
      "movdqu (%0,%3),%%xmm3 \n"
      "lea (%0,%3,2),%0 \n"
      "movdqa %%xmm2,%%xmm10 \n"
      "punpcklbw %%xmm3,%%xmm2 \n"
      "punpckhbw %%xmm3,%%xmm10 \n"
      "movdqa %%xmm2,%%xmm3 \n"
      "movdqa %%xmm10,%%xmm11 \n"
      "movdqu (%0),%%xmm4 \n"
      "palignr $0x8,%%xmm3,%%xmm3 \n"
      "palignr $0x8,%%xmm11,%%xmm11 \n"
      "movdqu (%0,%3),%%xmm5 \n"
      "lea (%0,%3,2),%0 \n"
      "movdqa %%xmm4,%%xmm12 \n"
      "punpcklbw %%xmm5,%%xmm4 \n"
      "punpckhbw %%xmm5,%%xmm12 \n"
      "movdqa %%xmm4,%%xmm5 \n"
      "movdqa %%xmm12,%%xmm13 \n"
      "movdqu (%0),%%xmm6 \n"
      "palignr $0x8,%%xmm5,%%xmm5 \n"
      "palignr $0x8,%%xmm13,%%xmm13 \n"
      "movdqu (%0,%3),%%xmm7 \n"
      "lea (%0,%3,2),%0 \n"
      "movdqa %%xmm6,%%xmm14 \n"
      "punpcklbw %%xmm7,%%xmm6 \n"
      "punpckhbw %%xmm7,%%xmm14 \n"
      "neg %3 \n"
      "movdqa %%xmm6,%%xmm7 \n"
      "movdqa %%xmm14,%%xmm15 \n"
      "lea 0x10(%0,%3,8),%0 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      "palignr $0x8,%%xmm15,%%xmm15 \n"
      "neg %3 \n"
      // Second round of bit swap.
      "punpcklwd %%xmm2,%%xmm0 \n"
      "punpcklwd %%xmm3,%%xmm1 \n"
      "movdqa %%xmm0,%%xmm2 \n"
      "movdqa %%xmm1,%%xmm3 \n"
      "palignr $0x8,%%xmm2,%%xmm2 \n"
      "palignr $0x8,%%xmm3,%%xmm3 \n"
      "punpcklwd %%xmm6,%%xmm4 \n"
      "punpcklwd %%xmm7,%%xmm5 \n"
      "movdqa %%xmm4,%%xmm6 \n"
      "movdqa %%xmm5,%%xmm7 \n"
      "palignr $0x8,%%xmm6,%%xmm6 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      "punpcklwd %%xmm10,%%xmm8 \n"
      "punpcklwd %%xmm11,%%xmm9 \n"
      "movdqa %%xmm8,%%xmm10 \n"
      "movdqa %%xmm9,%%xmm11 \n"
      "palignr $0x8,%%xmm10,%%xmm10 \n"
      "palignr $0x8,%%xmm11,%%xmm11 \n"
      "punpcklwd %%xmm14,%%xmm12 \n"
      "punpcklwd %%xmm15,%%xmm13 \n"
      "movdqa %%xmm12,%%xmm14 \n"
      "movdqa %%xmm13,%%xmm15 \n"
      "palignr $0x8,%%xmm14,%%xmm14 \n"
      "palignr $0x8,%%xmm15,%%xmm15 \n"
      // Third round of bit swap.
      // Write to the destination pointer.
      "punpckldq %%xmm4,%%xmm0 \n"
      "movq %%xmm0,(%1) \n"
      "movdqa %%xmm0,%%xmm4 \n"
      "palignr $0x8,%%xmm4,%%xmm4 \n"
      "movq %%xmm4,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm6,%%xmm2 \n"
      "movdqa %%xmm2,%%xmm6 \n"
      "movq %%xmm2,(%1) \n"
      "palignr $0x8,%%xmm6,%%xmm6 \n"
      "punpckldq %%xmm5,%%xmm1 \n"
      "movq %%xmm6,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "movdqa %%xmm1,%%xmm5 \n"
      "movq %%xmm1,(%1) \n"
      "palignr $0x8,%%xmm5,%%xmm5 \n"
      "movq %%xmm5,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm7,%%xmm3 \n"
      "movq %%xmm3,(%1) \n"
      "movdqa %%xmm3,%%xmm7 \n"
      "palignr $0x8,%%xmm7,%%xmm7 \n"
      "movq %%xmm7,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm12,%%xmm8 \n"
      "movq %%xmm8,(%1) \n"
      "movdqa %%xmm8,%%xmm12 \n"
      "palignr $0x8,%%xmm12,%%xmm12 \n"
      "movq %%xmm12,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm14,%%xmm10 \n"
      "movdqa %%xmm10,%%xmm14 \n"
      "movq %%xmm10,(%1) \n"
      "palignr $0x8,%%xmm14,%%xmm14 \n"
      "punpckldq %%xmm13,%%xmm9 \n"
      "movq %%xmm14,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "movdqa %%xmm9,%%xmm13 \n"
      "movq %%xmm9,(%1) \n"
      "palignr $0x8,%%xmm13,%%xmm13 \n"
      "movq %%xmm13,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "punpckldq %%xmm15,%%xmm11 \n"
      "movq %%xmm11,(%1) \n"
      "movdqa %%xmm11,%%xmm15 \n"
      "palignr $0x8,%%xmm15,%%xmm15 \n"
      "sub $0x10,%2 \n"
      "movq %%xmm15,(%1,%4) \n"
      "lea (%1,%4,2),%1 \n"
      "jg 1b \n"
      : "+r"(src),                    // %0
        "+r"(dst),                    // %1
        "+r"(width)                   // %2
      : "r"((intptr_t)(src_stride)),  // %3
        "r"((intptr_t)(dst_stride))   // %4
      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7", "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
        "xmm15");
}
#endif  // defined(HAS_TRANSPOSEWX8_FAST_SSSE3)

// Transpose UV 8x8.  64 bit.
#if defined(HAS_TRANSPOSEUVWX8_SSE2)
void TransposeUVWx8_SSE2(const uint8_t* src,
                         int src_stride,
                         uint8_t* dst_a,
                         int dst_stride_a,
                         uint8_t* dst_b,
                         int dst_stride_b,
                         int width) {
  asm volatile(
      // Read in the data from the source pointer.
      // First round of bit swap.
      LABELALIGN
      "1: \n"
      "movdqu (%0),%%xmm0 \n"
      "movdqu (%0,%4),%%xmm1 \n"
      "lea (%0,%4,2),%0 \n"
      "movdqa %%xmm0,%%xmm8 \n"
      "punpcklbw %%xmm1,%%xmm0 \n"
      "punpckhbw %%xmm1,%%xmm8 \n"
      "movdqa %%xmm8,%%xmm1 \n"
      "movdqu (%0),%%xmm2 \n"
      "movdqu (%0,%4),%%xmm3 \n"
      "lea (%0,%4,2),%0 \n"
      "movdqa %%xmm2,%%xmm8 \n"
      "punpcklbw %%xmm3,%%xmm2 \n"
      "punpckhbw %%xmm3,%%xmm8 \n"
      "movdqa %%xmm8,%%xmm3 \n"
      "movdqu (%0),%%xmm4 \n"
      "movdqu (%0,%4),%%xmm5 \n"
      "lea (%0,%4,2),%0 \n"
      "movdqa %%xmm4,%%xmm8 \n"
      "punpcklbw %%xmm5,%%xmm4 \n"
      "punpckhbw %%xmm5,%%xmm8 \n"
      "movdqa %%xmm8,%%xmm5 \n"
      "movdqu (%0),%%xmm6 \n"
      "movdqu (%0,%4),%%xmm7 \n"
      "lea (%0,%4,2),%0 \n"
      "movdqa %%xmm6,%%xmm8 \n"
      "punpcklbw %%xmm7,%%xmm6 \n"
      "neg %4 \n"
      "lea 0x10(%0,%4,8),%0 \n"
      "punpckhbw %%xmm7,%%xmm8 \n"
      "movdqa %%xmm8,%%xmm7 \n"
      "neg %4 \n"
      // Second round of bit swap.
      "movdqa %%xmm0,%%xmm8 \n"
      "movdqa %%xmm1,%%xmm9 \n"
      "punpckhwd %%xmm2,%%xmm8 \n"
      "punpckhwd %%xmm3,%%xmm9 \n"
      "punpcklwd %%xmm2,%%xmm0 \n"
      "punpcklwd %%xmm3,%%xmm1 \n"
      "movdqa %%xmm8,%%xmm2 \n"
      "movdqa %%xmm9,%%xmm3 \n"
      "movdqa %%xmm4,%%xmm8 \n"
      "movdqa %%xmm5,%%xmm9 \n"
      "punpckhwd %%xmm6,%%xmm8 \n"
      "punpckhwd %%xmm7,%%xmm9 \n"
      "punpcklwd %%xmm6,%%xmm4 \n"
      "punpcklwd %%xmm7,%%xmm5 \n"
      "movdqa %%xmm8,%%xmm6 \n"
      "movdqa %%xmm9,%%xmm7 \n"
      // Third round of bit swap.
      // Write to the destination pointer.
      "movdqa %%xmm0,%%xmm8 \n"
      "punpckldq %%xmm4,%%xmm0 \n"
      "movlpd %%xmm0,(%1) \n"  // Write back U channel
      "movhpd %%xmm0,(%2) \n"  // Write back V channel
      "punpckhdq %%xmm4,%%xmm8 \n"
      "movlpd %%xmm8,(%1,%5) \n"
      "lea (%1,%5,2),%1 \n"
      "movhpd %%xmm8,(%2,%6) \n"
      "lea (%2,%6,2),%2 \n"
      "movdqa %%xmm2,%%xmm8 \n"
      "punpckldq %%xmm6,%%xmm2 \n"
      "movlpd %%xmm2,(%1) \n"
      "movhpd %%xmm2,(%2) \n"
      "punpckhdq %%xmm6,%%xmm8 \n"
      "movlpd %%xmm8,(%1,%5) \n"
      "lea (%1,%5,2),%1 \n"
      "movhpd %%xmm8,(%2,%6) \n"
      "lea (%2,%6,2),%2 \n"
      "movdqa %%xmm1,%%xmm8 \n"
      "punpckldq %%xmm5,%%xmm1 \n"
      "movlpd %%xmm1,(%1) \n"
      "movhpd %%xmm1,(%2) \n"
      "punpckhdq %%xmm5,%%xmm8 \n"
      "movlpd %%xmm8,(%1,%5) \n"
      "lea (%1,%5,2),%1 \n"
      "movhpd %%xmm8,(%2,%6) \n"
      "lea (%2,%6,2),%2 \n"
      "movdqa %%xmm3,%%xmm8 \n"
      "punpckldq %%xmm7,%%xmm3 \n"
      "movlpd %%xmm3,(%1) \n"
      "movhpd %%xmm3,(%2) \n"
      "punpckhdq %%xmm7,%%xmm8 \n"
      "sub $0x8,%3 \n"
      "movlpd %%xmm8,(%1,%5) \n"
      "lea (%1,%5,2),%1 \n"
      "movhpd %%xmm8,(%2,%6) \n"
      "lea (%2,%6,2),%2 \n"
      "jg 1b \n"
      : "+r"(src),                      // %0
        "+r"(dst_a),                    // %1
        "+r"(dst_b),                    // %2
        "+r"(width)                     // %3
      : "r"((intptr_t)(src_stride)),    // %4
        "r"((intptr_t)(dst_stride_a)),  // %5
        "r"((intptr_t)(dst_stride_b))   // %6
      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7", "xmm8", "xmm9");
}
#endif  // defined(HAS_TRANSPOSEUVWX8_SSE2)
#endif  // defined(__x86_64__) || defined(__i386__)

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