mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2026-05-01 03:19:18 +08:00
62c19d062d
Removed all SSE functions, macros, dispatching logic, and related unit tests across the repository to reduce code size and complexity. Left cpuid detection intact. Supported architectures like AVX2, NEON, SVE, etc. are unaffected. R=rrwinterton@gmail.com Bug: None Test: Build and run libyuv_unittest Change-Id: Id19608dba35b79c4c8fc31f920a6a968883d300f
131 lines
4.1 KiB
C++
131 lines
4.1 KiB
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__)) && \
|
|
!defined(LIBYUV_ENABLE_ROWWIN)
|
|
|
|
// "memory" clobber prevents the reads from being removed
|
|
|
|
#if defined(__x86_64__)
|
|
|
|
#else
|
|
|
|
#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};
|
|
|
|
|
|
|
|
#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
|
|
|
|
|
|
|
|
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
|
|
};
|
|
|
|
|
|
#endif // defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|