michael@0: // Copyright (c) 2010 The Chromium Authors. All rights reserved.
michael@0: // Use of this source code is governed by a BSD-style license that can be
michael@0: // found in the LICENSE file.
michael@0: 
michael@0: #include "yuv_row.h"
michael@0: #include "mozilla/SSE.h"
michael@0: 
michael@0: #define DCHECK(a)
michael@0: 
michael@0: extern "C" {
michael@0: 
michael@0: #if defined(ARCH_CPU_X86_64)
michael@0: 
michael@0: // We don't need CPUID guards here, since x86-64 implies SSE2.
michael@0: 
michael@0: // AMD64 ABI uses register paremters.
michael@0: void FastConvertYUVToRGB32Row(const uint8* y_buf,  // rdi
michael@0:                               const uint8* u_buf,  // rsi
michael@0:                               const uint8* v_buf,  // rdx
michael@0:                               uint8* rgb_buf,      // rcx
michael@0:                               int width) {         // r8
michael@0:   asm(
michael@0:   "jmp    1f\n"
michael@0: "0:"
michael@0:   "movzb  (%1),%%r10\n"
michael@0:   "add    $0x1,%1\n"
michael@0:   "movzb  (%2),%%r11\n"
michael@0:   "add    $0x1,%2\n"
michael@0:   "movq   2048(%5,%%r10,8),%%xmm0\n"
michael@0:   "movzb  (%0),%%r10\n"
michael@0:   "movq   4096(%5,%%r11,8),%%xmm1\n"
michael@0:   "movzb  0x1(%0),%%r11\n"
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0:   "movq   (%5,%%r10,8),%%xmm2\n"
michael@0:   "add    $0x2,%0\n"
michael@0:   "movq   (%5,%%r11,8),%%xmm3\n"
michael@0:   "paddsw %%xmm0,%%xmm2\n"
michael@0:   "paddsw %%xmm0,%%xmm3\n"
michael@0:   "shufps $0x44,%%xmm3,%%xmm2\n"
michael@0:   "psraw  $0x6,%%xmm2\n"
michael@0:   "packuswb %%xmm2,%%xmm2\n"
michael@0:   "movq   %%xmm2,0x0(%3)\n"
michael@0:   "add    $0x8,%3\n"
michael@0: "1:"
michael@0:   "sub    $0x2,%4\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0: "2:"
michael@0:   "add    $0x1,%4\n"
michael@0:   "js     3f\n"
michael@0: 
michael@0:   "movzb  (%1),%%r10\n"
michael@0:   "movq   2048(%5,%%r10,8),%%xmm0\n"
michael@0:   "movzb  (%2),%%r10\n"
michael@0:   "movq   4096(%5,%%r10,8),%%xmm1\n"
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0:   "movzb  (%0),%%r10\n"
michael@0:   "movq   (%5,%%r10,8),%%xmm1\n"
michael@0:   "paddsw %%xmm0,%%xmm1\n"
michael@0:   "psraw  $0x6,%%xmm1\n"
michael@0:   "packuswb %%xmm1,%%xmm1\n"
michael@0:   "movd   %%xmm1,0x0(%3)\n"
michael@0: "3:"
michael@0:   :
michael@0:   : "r"(y_buf),  // %0
michael@0:     "r"(u_buf),  // %1
michael@0:     "r"(v_buf),  // %2
michael@0:     "r"(rgb_buf),  // %3
michael@0:     "r"(width),  // %4
michael@0:     "r" (kCoefficientsRgbY)  // %5
michael@0:   : "memory", "r10", "r11", "xmm0", "xmm1", "xmm2", "xmm3"
michael@0: );
michael@0: }
michael@0: 
michael@0: void ScaleYUVToRGB32Row(const uint8* y_buf,  // rdi
michael@0:                         const uint8* u_buf,  // rsi
michael@0:                         const uint8* v_buf,  // rdx
michael@0:                         uint8* rgb_buf,      // rcx
michael@0:                         int width,           // r8
michael@0:                         int source_dx) {     // r9
michael@0:   asm(
michael@0:   "xor    %%r11,%%r11\n"
michael@0:   "sub    $0x2,%4\n"
michael@0:   "js     1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "mov    %%r11,%%r10\n"
michael@0:   "sar    $0x11,%%r10\n"
michael@0:   "movzb  (%1,%%r10,1),%%rax\n"
michael@0:   "movq   2048(%5,%%rax,8),%%xmm0\n"
michael@0:   "movzb  (%2,%%r10,1),%%rax\n"
michael@0:   "movq   4096(%5,%%rax,8),%%xmm1\n"
michael@0:   "lea    (%%r11,%6),%%r10\n"
michael@0:   "sar    $0x10,%%r11\n"
michael@0:   "movzb  (%0,%%r11,1),%%rax\n"
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0:   "movq   (%5,%%rax,8),%%xmm1\n"
michael@0:   "lea    (%%r10,%6),%%r11\n"
michael@0:   "sar    $0x10,%%r10\n"
michael@0:   "movzb  (%0,%%r10,1),%%rax\n"
michael@0:   "movq   (%5,%%rax,8),%%xmm2\n"
michael@0:   "paddsw %%xmm0,%%xmm1\n"
michael@0:   "paddsw %%xmm0,%%xmm2\n"
michael@0:   "shufps $0x44,%%xmm2,%%xmm1\n"
michael@0:   "psraw  $0x6,%%xmm1\n"
michael@0:   "packuswb %%xmm1,%%xmm1\n"
michael@0:   "movq   %%xmm1,0x0(%3)\n"
michael@0:   "add    $0x8,%3\n"
michael@0:   "sub    $0x2,%4\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0: "1:"
michael@0:   "add    $0x1,%4\n"
michael@0:   "js     2f\n"
michael@0: 
michael@0:   "mov    %%r11,%%r10\n"
michael@0:   "sar    $0x11,%%r10\n"
michael@0:   "movzb  (%1,%%r10,1),%%rax\n"
michael@0:   "movq   2048(%5,%%rax,8),%%xmm0\n"
michael@0:   "movzb  (%2,%%r10,1),%%rax\n"
michael@0:   "movq   4096(%5,%%rax,8),%%xmm1\n"
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0:   "sar    $0x10,%%r11\n"
michael@0:   "movzb  (%0,%%r11,1),%%rax\n"
michael@0:   "movq   (%5,%%rax,8),%%xmm1\n"
michael@0:   "paddsw %%xmm0,%%xmm1\n"
michael@0:   "psraw  $0x6,%%xmm1\n"
michael@0:   "packuswb %%xmm1,%%xmm1\n"
michael@0:   "movd   %%xmm1,0x0(%3)\n"
michael@0: 
michael@0: "2:"
michael@0:   :
michael@0:   : "r"(y_buf),  // %0
michael@0:     "r"(u_buf),  // %1
michael@0:     "r"(v_buf),  // %2
michael@0:     "r"(rgb_buf),  // %3
michael@0:     "r"(width),  // %4
michael@0:     "r" (kCoefficientsRgbY),  // %5
michael@0:     "r"(static_cast<long>(source_dx))  // %6
michael@0:   : "memory", "r10", "r11", "rax", "xmm0", "xmm1", "xmm2"
michael@0: );
michael@0: }
michael@0: 
michael@0: void LinearScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width,
michael@0:                               int source_dx) {
michael@0:   asm(
michael@0:   "xor    %%r11,%%r11\n"   // x = 0
michael@0:   "sub    $0x2,%4\n"
michael@0:   "js     2f\n"
michael@0:   "cmp    $0x20000,%6\n"   // if source_dx >= 2.0
michael@0:   "jl     0f\n"
michael@0:   "mov    $0x8000,%%r11\n" // x = 0.5 for 1/2 or less
michael@0: "0:"
michael@0: 
michael@0: "1:"
michael@0:   "mov    %%r11,%%r10\n"
michael@0:   "sar    $0x11,%%r10\n"
michael@0: 
michael@0:   "movzb  (%1, %%r10, 1), %%r13 \n"
michael@0:   "movzb  1(%1, %%r10, 1), %%r14 \n"
michael@0:   "mov    %%r11, %%rax \n"
michael@0:   "and    $0x1fffe, %%rax \n"
michael@0:   "imul   %%rax, %%r14 \n"
michael@0:   "xor    $0x1fffe, %%rax \n"
michael@0:   "imul   %%rax, %%r13 \n"
michael@0:   "add    %%r14, %%r13 \n"
michael@0:   "shr    $17, %%r13 \n"
michael@0:   "movq   2048(%5,%%r13,8), %%xmm0\n"
michael@0: 
michael@0:   "movzb  (%2, %%r10, 1), %%r13 \n"
michael@0:   "movzb  1(%2, %%r10, 1), %%r14 \n"
michael@0:   "mov    %%r11, %%rax \n"
michael@0:   "and    $0x1fffe, %%rax \n"
michael@0:   "imul   %%rax, %%r14 \n"
michael@0:   "xor    $0x1fffe, %%rax \n"
michael@0:   "imul   %%rax, %%r13 \n"
michael@0:   "add    %%r14, %%r13 \n"
michael@0:   "shr    $17, %%r13 \n"
michael@0:   "movq   4096(%5,%%r13,8), %%xmm1\n"
michael@0: 
michael@0:   "mov    %%r11, %%rax \n"
michael@0:   "lea    (%%r11,%6),%%r10\n"
michael@0:   "sar    $0x10,%%r11\n"
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0: 
michael@0:   "movzb  (%0, %%r11, 1), %%r13 \n"
michael@0:   "movzb  1(%0, %%r11, 1), %%r14 \n"
michael@0:   "and    $0xffff, %%rax \n"
michael@0:   "imul   %%rax, %%r14 \n"
michael@0:   "xor    $0xffff, %%rax \n"
michael@0:   "imul   %%rax, %%r13 \n"
michael@0:   "add    %%r14, %%r13 \n"
michael@0:   "shr    $16, %%r13 \n"
michael@0:   "movq   (%5,%%r13,8),%%xmm1\n"
michael@0: 
michael@0:   "mov    %%r10, %%rax \n"
michael@0:   "lea    (%%r10,%6),%%r11\n"
michael@0:   "sar    $0x10,%%r10\n"
michael@0: 
michael@0:   "movzb  (%0,%%r10,1), %%r13 \n"
michael@0:   "movzb  1(%0,%%r10,1), %%r14 \n"
michael@0:   "and    $0xffff, %%rax \n"
michael@0:   "imul   %%rax, %%r14 \n"
michael@0:   "xor    $0xffff, %%rax \n"
michael@0:   "imul   %%rax, %%r13 \n"
michael@0:   "add    %%r14, %%r13 \n"
michael@0:   "shr    $16, %%r13 \n"
michael@0:   "movq   (%5,%%r13,8),%%xmm2\n"
michael@0: 
michael@0:   "paddsw %%xmm0,%%xmm1\n"
michael@0:   "paddsw %%xmm0,%%xmm2\n"
michael@0:   "shufps $0x44,%%xmm2,%%xmm1\n"
michael@0:   "psraw  $0x6,%%xmm1\n"
michael@0:   "packuswb %%xmm1,%%xmm1\n"
michael@0:   "movq   %%xmm1,0x0(%3)\n"
michael@0:   "add    $0x8,%3\n"
michael@0:   "sub    $0x2,%4\n"
michael@0:   "jns    1b\n"
michael@0: 
michael@0: "2:"
michael@0:   "add    $0x1,%4\n"
michael@0:   "js     3f\n"
michael@0: 
michael@0:   "mov    %%r11,%%r10\n"
michael@0:   "sar    $0x11,%%r10\n"
michael@0: 
michael@0:   "movzb  (%1,%%r10,1), %%r13 \n"
michael@0:   "movq   2048(%5,%%r13,8),%%xmm0\n"
michael@0: 
michael@0:   "movzb  (%2,%%r10,1), %%r13 \n"
michael@0:   "movq   4096(%5,%%r13,8),%%xmm1\n"
michael@0: 
michael@0:   "paddsw %%xmm1,%%xmm0\n"
michael@0:   "sar    $0x10,%%r11\n"
michael@0: 
michael@0:   "movzb  (%0,%%r11,1), %%r13 \n"
michael@0:   "movq   (%5,%%r13,8),%%xmm1\n"
michael@0: 
michael@0:   "paddsw %%xmm0,%%xmm1\n"
michael@0:   "psraw  $0x6,%%xmm1\n"
michael@0:   "packuswb %%xmm1,%%xmm1\n"
michael@0:   "movd   %%xmm1,0x0(%3)\n"
michael@0: 
michael@0: "3:"
michael@0:   :
michael@0:   : "r"(y_buf),  // %0
michael@0:     "r"(u_buf),  // %1
michael@0:     "r"(v_buf),  // %2
michael@0:     "r"(rgb_buf),  // %3
michael@0:     "r"(width),  // %4
michael@0:     "r" (kCoefficientsRgbY),  // %5
michael@0:     "r"(static_cast<long>(source_dx))  // %6
michael@0:   : "memory", "r10", "r11", "r13", "r14", "rax", "xmm0", "xmm1", "xmm2"
michael@0: );
michael@0: }
michael@0: 
michael@0: #elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && !defined(__PIC__)
michael@0: 
michael@0: // PIC version is slower because less registers are available, so
michael@0: // non-PIC is used on platforms where it is possible.
michael@0: void FastConvertYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                                   const uint8* u_buf,
michael@0:                                   const uint8* v_buf,
michael@0:                                   uint8* rgb_buf,
michael@0:                                   int width);
michael@0:   asm(
michael@0:   ".text\n"
michael@0:   ".global FastConvertYUVToRGB32Row_SSE\n"
michael@0:   ".type FastConvertYUVToRGB32Row_SSE, @function\n"
michael@0: "FastConvertYUVToRGB32Row_SSE:\n"
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x28(%esp),%edi\n"
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0:   "mov    0x34(%esp),%ecx\n"
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "movzbl (%edi),%eax\n"
michael@0:   "add    $0x1,%edi\n"
michael@0:   "movzbl (%esi),%ebx\n"
michael@0:   "add    $0x1,%esi\n"
michael@0:   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
michael@0:   "movzbl (%edx),%eax\n"
michael@0:   "paddsw kCoefficientsRgbY+4096(,%ebx,8),%mm0\n"
michael@0:   "movzbl 0x1(%edx),%ebx\n"
michael@0:   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
michael@0:   "add    $0x2,%edx\n"
michael@0:   "movq   kCoefficientsRgbY(,%ebx,8),%mm2\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: "1:"
michael@0:   "sub    $0x2,%ecx\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0:   "and    $0x1,%ecx\n"
michael@0:   "je     2f\n"
michael@0: 
michael@0:   "movzbl (%edi),%eax\n"
michael@0:   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
michael@0:   "movzbl (%esi),%eax\n"
michael@0:   "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
michael@0:   "movzbl (%edx),%eax\n"
michael@0:   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "packuswb %mm1,%mm1\n"
michael@0:   "movd   %mm1,0x0(%ebp)\n"
michael@0: "2:"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: void FastConvertYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     FastConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width);
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
michael@0: }
michael@0: 
michael@0: 
michael@0: void ScaleYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                             const uint8* u_buf,
michael@0:                             const uint8* v_buf,
michael@0:                             uint8* rgb_buf,
michael@0:                             int width,
michael@0:                             int source_dx);
michael@0:   asm(
michael@0:   ".text\n"
michael@0:   ".global ScaleYUVToRGB32Row_SSE\n"
michael@0:   ".type ScaleYUVToRGB32Row_SSE, @function\n"
michael@0: "ScaleYUVToRGB32Row_SSE:\n"
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x28(%esp),%edi\n"
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0:   "mov    0x34(%esp),%ecx\n"
michael@0:   "xor    %ebx,%ebx\n"
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%edi,%eax,1),%eax\n"
michael@0:   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%esi,%eax,1),%eax\n"
michael@0:   "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   kCoefficientsRgbY(,%eax,8),%mm2\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: "1:"
michael@0:   "sub    $0x2,%ecx\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0:   "and    $0x1,%ecx\n"
michael@0:   "je     2f\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%edi,%eax,1),%eax\n"
michael@0:   "movq   kCoefficientsRgbY+2048(,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%esi,%eax,1),%eax\n"
michael@0:   "paddsw kCoefficientsRgbY+4096(,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   kCoefficientsRgbY(,%eax,8),%mm1\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "packuswb %mm1,%mm1\n"
michael@0:   "movd   %mm1,0x0(%ebp)\n"
michael@0: 
michael@0: "2:"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: void ScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                         const uint8* u_buf,
michael@0:                         const uint8* v_buf,
michael@0:                         uint8* rgb_buf,
michael@0:                         int width,
michael@0:                         int source_dx)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     ScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
michael@0:                            width, source_dx);
michael@0:   }
michael@0: 
michael@0:   ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
michael@0:                        width, source_dx);
michael@0: }
michael@0: 
michael@0: void LinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                                   const uint8* u_buf,
michael@0:                                   const uint8* v_buf,
michael@0:                                   uint8* rgb_buf,
michael@0:                                   int width,
michael@0:                                   int source_dx);
michael@0:   asm(
michael@0:   ".text\n"
michael@0:   ".global LinearScaleYUVToRGB32Row_SSE\n"
michael@0:   ".type LinearScaleYUVToRGB32Row_SSE, @function\n"
michael@0: "LinearScaleYUVToRGB32Row_SSE:\n"
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x28(%esp),%edi\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0: 
michael@0:   // source_width = width * source_dx + ebx
michael@0:   "mov    0x34(%esp), %ecx\n"
michael@0:   "imull  0x38(%esp), %ecx\n"
michael@0:   "mov    %ecx, 0x34(%esp)\n"
michael@0: 
michael@0:   "mov    0x38(%esp), %ecx\n"
michael@0:   "xor    %ebx,%ebx\n"     // x = 0
michael@0:   "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
michael@0:   "jl     1f\n"
michael@0:   "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0: 
michael@0:   "movzbl (%edi,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%edi,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "andl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $17, %ecx \n"
michael@0:   "movq   kCoefficientsRgbY+2048(,%ecx,8),%mm0\n"
michael@0: 
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0: 
michael@0:   "movzbl (%esi,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%esi,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "andl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $17, %ecx \n"
michael@0:   "paddsw kCoefficientsRgbY+4096(,%ecx,8),%mm0\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%edx,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "andl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $16, %ecx \n"
michael@0:   "movq   kCoefficientsRgbY(,%ecx,8),%mm1\n"
michael@0: 
michael@0:   "cmp    0x34(%esp), %ebx\n"
michael@0:   "jge    2f\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%edx,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "andl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $16, %ecx \n"
michael@0:   "movq   kCoefficientsRgbY(,%ecx,8),%mm2\n"
michael@0: 
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: 
michael@0: "1:"
michael@0:   "cmp    0x34(%esp), %ebx\n"
michael@0:   "jl     0b\n"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: 
michael@0: "2:"
michael@0:   "paddsw %mm0, %mm1\n"
michael@0:   "psraw $6, %mm1\n"
michael@0:   "packuswb %mm1, %mm1\n"
michael@0:   "movd %mm1, (%ebp)\n"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: void LinearScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width,
michael@0:                               int source_dx)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     LinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf,
michael@0:                                  width, source_dx);
michael@0:   }
michael@0: 
michael@0:   LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf,
michael@0:                              width, source_dx);
michael@0: }
michael@0: 
michael@0: #elif defined(MOZILLA_MAY_SUPPORT_SSE) && defined(ARCH_CPU_X86_32) && defined(__PIC__)
michael@0: 
michael@0: void PICConvertYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                                  const uint8* u_buf,
michael@0:                                  const uint8* v_buf,
michael@0:                                  uint8* rgb_buf,
michael@0:                                  int width,
michael@0:                                  int16 *kCoefficientsRgbY);
michael@0: 
michael@0:   asm(
michael@0:   ".text\n"
michael@0: #if defined(XP_MACOSX)
michael@0: "_PICConvertYUVToRGB32Row_SSE:\n"
michael@0: #else
michael@0: "PICConvertYUVToRGB32Row_SSE:\n"
michael@0: #endif
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x28(%esp),%edi\n"
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0:   "mov    0x38(%esp),%ecx\n"
michael@0: 
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "movzbl (%edi),%eax\n"
michael@0:   "add    $0x1,%edi\n"
michael@0:   "movzbl (%esi),%ebx\n"
michael@0:   "add    $0x1,%esi\n"
michael@0:   "movq   2048(%ecx,%eax,8),%mm0\n"
michael@0:   "movzbl (%edx),%eax\n"
michael@0:   "paddsw 4096(%ecx,%ebx,8),%mm0\n"
michael@0:   "movzbl 0x1(%edx),%ebx\n"
michael@0:   "movq   0(%ecx,%eax,8),%mm1\n"
michael@0:   "add    $0x2,%edx\n"
michael@0:   "movq   0(%ecx,%ebx,8),%mm2\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: "1:"
michael@0:   "subl   $0x2,0x34(%esp)\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0:   "andl   $0x1,0x34(%esp)\n"
michael@0:   "je     2f\n"
michael@0: 
michael@0:   "movzbl (%edi),%eax\n"
michael@0:   "movq   2048(%ecx,%eax,8),%mm0\n"
michael@0:   "movzbl (%esi),%eax\n"
michael@0:   "paddsw 4096(%ecx,%eax,8),%mm0\n"
michael@0:   "movzbl (%edx),%eax\n"
michael@0:   "movq   0(%ecx,%eax,8),%mm1\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "packuswb %mm1,%mm1\n"
michael@0:   "movd   %mm1,0x0(%ebp)\n"
michael@0: "2:"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: void FastConvertYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     PICConvertYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
michael@0:                                 &kCoefficientsRgbY[0][0]);
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
michael@0: }
michael@0: 
michael@0: void PICScaleYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                                const uint8* u_buf,
michael@0:                                const uint8* v_buf,
michael@0:                                uint8* rgb_buf,
michael@0:                                int width,
michael@0:                                int source_dx,
michael@0:                                int16 *kCoefficientsRgbY);
michael@0: 
michael@0:   asm(
michael@0:   ".text\n"
michael@0: #if defined(XP_MACOSX)
michael@0: "_PICScaleYUVToRGB32Row_SSE:\n"
michael@0: #else
michael@0: "PICScaleYUVToRGB32Row_SSE:\n"
michael@0: #endif
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x28(%esp),%edi\n"
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0:   "mov    0x3c(%esp),%ecx\n"
michael@0:   "xor    %ebx,%ebx\n"
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%edi,%eax,1),%eax\n"
michael@0:   "movq   2048(%ecx,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%esi,%eax,1),%eax\n"
michael@0:   "paddsw 4096(%ecx,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   0(%ecx,%eax,8),%mm1\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   0(%ecx,%eax,8),%mm2\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: "1:"
michael@0:   "subl   $0x2,0x34(%esp)\n"
michael@0:   "jns    0b\n"
michael@0: 
michael@0:   "andl   $0x1,0x34(%esp)\n"
michael@0:   "je     2f\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%edi,%eax,1),%eax\n"
michael@0:   "movq   2048(%ecx,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0:   "movzbl (%esi,%eax,1),%eax\n"
michael@0:   "paddsw 4096(%ecx,%eax,8),%mm0\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%eax\n"
michael@0:   "movq   0(%ecx,%eax,8),%mm1\n"
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "packuswb %mm1,%mm1\n"
michael@0:   "movd   %mm1,0x0(%ebp)\n"
michael@0: 
michael@0: "2:"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: void ScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                         const uint8* u_buf,
michael@0:                         const uint8* v_buf,
michael@0:                         uint8* rgb_buf,
michael@0:                         int width,
michael@0:                         int source_dx)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     PICScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width, source_dx,
michael@0:                               &kCoefficientsRgbY[0][0]);
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
michael@0: }
michael@0: 
michael@0: void PICLinearScaleYUVToRGB32Row_SSE(const uint8* y_buf,
michael@0:                                      const uint8* u_buf,
michael@0:                                      const uint8* v_buf,
michael@0:                                      uint8* rgb_buf,
michael@0:                                      int width,
michael@0:                                      int source_dx,
michael@0:                                      int16 *kCoefficientsRgbY);
michael@0: 
michael@0:   asm(
michael@0:   ".text\n"
michael@0: #if defined(XP_MACOSX)
michael@0: "_PICLinearScaleYUVToRGB32Row_SSE:\n"
michael@0: #else
michael@0: "PICLinearScaleYUVToRGB32Row_SSE:\n"
michael@0: #endif
michael@0:   "pusha\n"
michael@0:   "mov    0x24(%esp),%edx\n"
michael@0:   "mov    0x30(%esp),%ebp\n"
michael@0:   "mov    0x34(%esp),%ecx\n"
michael@0:   "mov    0x3c(%esp),%edi\n"
michael@0:   "xor    %ebx,%ebx\n"
michael@0: 
michael@0:   // source_width = width * source_dx + ebx
michael@0:   "mov    0x34(%esp), %ecx\n"
michael@0:   "imull  0x38(%esp), %ecx\n"
michael@0:   "mov    %ecx, 0x34(%esp)\n"
michael@0: 
michael@0:   "mov    0x38(%esp), %ecx\n"
michael@0:   "xor    %ebx,%ebx\n"     // x = 0
michael@0:   "cmp    $0x20000,%ecx\n" // if source_dx >= 2.0
michael@0:   "jl     1f\n"
michael@0:   "mov    $0x8000,%ebx\n"  // x = 0.5 for 1/2 or less
michael@0:   "jmp    1f\n"
michael@0: 
michael@0: "0:"
michael@0:   "mov    0x28(%esp),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0: 
michael@0:   "movzbl (%esi,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%esi,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "andl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $17, %ecx \n"
michael@0:   "movq   2048(%edi,%ecx,8),%mm0\n"
michael@0: 
michael@0:   "mov    0x2c(%esp),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x11,%eax\n"
michael@0: 
michael@0:   "movzbl (%esi,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%esi,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "andl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0x1fffe, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $17, %ecx \n"
michael@0:   "paddsw 4096(%edi,%ecx,8),%mm0\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%edx,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "andl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $16, %ecx \n"
michael@0:   "movq   (%edi,%ecx,8),%mm1\n"
michael@0: 
michael@0:   "cmp    0x34(%esp), %ebx\n"
michael@0:   "jge    2f\n"
michael@0: 
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "sar    $0x10,%eax\n"
michael@0:   "movzbl (%edx,%eax,1),%ecx\n"
michael@0:   "movzbl 1(%edx,%eax,1),%esi\n"
michael@0:   "mov    %ebx,%eax\n"
michael@0:   "add    0x38(%esp),%ebx\n"
michael@0:   "andl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %esi \n"
michael@0:   "xorl   $0xffff, %eax \n"
michael@0:   "imul   %eax, %ecx \n"
michael@0:   "addl   %esi, %ecx \n"
michael@0:   "shrl   $16, %ecx \n"
michael@0:   "movq   (%edi,%ecx,8),%mm2\n"
michael@0: 
michael@0:   "paddsw %mm0,%mm1\n"
michael@0:   "paddsw %mm0,%mm2\n"
michael@0:   "psraw  $0x6,%mm1\n"
michael@0:   "psraw  $0x6,%mm2\n"
michael@0:   "packuswb %mm2,%mm1\n"
michael@0:   "movntq %mm1,0x0(%ebp)\n"
michael@0:   "add    $0x8,%ebp\n"
michael@0: 
michael@0: "1:"
michael@0:   "cmp    %ebx, 0x34(%esp)\n"
michael@0:   "jg     0b\n"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: 
michael@0: "2:"
michael@0:   "paddsw %mm0, %mm1\n"
michael@0:   "psraw $6, %mm1\n"
michael@0:   "packuswb %mm1, %mm1\n"
michael@0:   "movd %mm1, (%ebp)\n"
michael@0:   "popa\n"
michael@0:   "ret\n"
michael@0: #if !defined(XP_MACOSX)
michael@0:   ".previous\n"
michael@0: #endif
michael@0: );
michael@0: 
michael@0: 
michael@0: void LinearScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width,
michael@0:                               int source_dx)
michael@0: {
michael@0:   if (mozilla::supports_sse()) {
michael@0:     PICLinearScaleYUVToRGB32Row_SSE(y_buf, u_buf, v_buf, rgb_buf, width,
michael@0:                                     source_dx, &kCoefficientsRgbY[0][0]);
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
michael@0: }
michael@0: #else
michael@0: void FastConvertYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width) {
michael@0:   FastConvertYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, 1);
michael@0: }
michael@0: 
michael@0: void ScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                         const uint8* u_buf,
michael@0:                         const uint8* v_buf,
michael@0:                         uint8* rgb_buf,
michael@0:                         int width,
michael@0:                         int source_dx) {
michael@0:   ScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
michael@0: }
michael@0: 
michael@0: void LinearScaleYUVToRGB32Row(const uint8* y_buf,
michael@0:                               const uint8* u_buf,
michael@0:                               const uint8* v_buf,
michael@0:                               uint8* rgb_buf,
michael@0:                               int width,
michael@0:                               int source_dx) {
michael@0:   LinearScaleYUVToRGB32Row_C(y_buf, u_buf, v_buf, rgb_buf, width, source_dx);
michael@0: }
michael@0: #endif
michael@0: 
michael@0: }