mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2025-12-06 16:56:55 +08:00
1c501a8f3f
- Renumber cpuid bits to use low byte to ID the type of CPU and upper 24 bits for features Intel CPUs starting at Icelake support FSMR adl:Has FSMR 0x8000 arl:Has FSMR 0x0 bdw:Has FSMR 0x0 clx:Has FSMR 0x0 cnl:Has FSMR 0x0 cpx:Has FSMR 0x0 emr:Has FSMR 0x8000 glm:Has FSMR 0x0 glp:Has FSMR 0x0 gnr:Has FSMR 0x8000 gnr256:Has FSMR 0x8000 hsw:Has FSMR 0x0 icl:Has FSMR 0x8000 icx:Has FSMR 0x8000 ivb:Has FSMR 0x0 knl:Has FSMR 0x0 knm:Has FSMR 0x0 lnl:Has FSMR 0x8000 mrm:Has FSMR 0x0 mtl:Has FSMR 0x8000 nhm:Has FSMR 0x0 pnr:Has FSMR 0x0 rpl:Has FSMR 0x8000 skl:Has FSMR 0x0 skx:Has FSMR 0x0 slm:Has FSMR 0x0 slt:Has FSMR 0x0 snb:Has FSMR 0x0 snr:Has FSMR 0x0 spr:Has FSMR 0x8000 srf:Has FSMR 0x0 tgl:Has FSMR 0x8000 tnt:Has FSMR 0x0 wsm:Has FSMR 0x0 Intel CPUs starting at Ivybridge support ERMS adl:Has ERMS 0x4000 arl:Has ERMS 0x4000 bdw:Has ERMS 0x4000 clx:Has ERMS 0x4000 cnl:Has ERMS 0x4000 cpx:Has ERMS 0x4000 emr:Has ERMS 0x4000 glm:Has ERMS 0x4000 glp:Has ERMS 0x4000 gnr:Has ERMS 0x4000 gnr256:Has ERMS 0x4000 hsw:Has ERMS 0x4000 icl:Has ERMS 0x4000 icx:Has ERMS 0x4000 ivb:Has ERMS 0x4000 knl:Has ERMS 0x4000 knm:Has ERMS 0x4000 lnl:Has ERMS 0x4000 mrm:Has ERMS 0x0 mtl:Has ERMS 0x4000 nhm:Has ERMS 0x0 pnr:Has ERMS 0x0 rpl:Has ERMS 0x4000 skl:Has ERMS 0x4000 skx:Has ERMS 0x4000 slm:Has ERMS 0x4000 slt:Has ERMS 0x0 snb:Has ERMS 0x0 snr:Has ERMS 0x4000 spr:Has ERMS 0x4000 srf:Has ERMS 0x4000 tgl:Has ERMS 0x4000 tnt:Has ERMS 0x4000 wsm:Has ERMS 0x0 Change-Id: I18e5a3905f2691ab66d4d0cb6f668c0a0ff72d37 Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/6027541 Reviewed-by: richard winterton <rrwinterton@gmail.com>
293 lines
13 KiB
C++
293 lines
13 KiB
C++
/*
|
|
* Copyright 2024 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/scale_row.h"
|
|
|
|
#ifdef __cplusplus
|
|
namespace libyuv {
|
|
extern "C" {
|
|
#endif
|
|
|
|
|
|
#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
|
|
defined(__aarch64__)
|
|
|
|
__arm_locally_streaming void ScaleRowDown2_SME(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
(void)src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cntb %x[vl] \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"1: \n"
|
|
"ptrue p0.b \n"
|
|
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
|
|
"incb %[src_ptr], all, mul #2 \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"st1b {z1.b}, p0, [%[dst_ptr]] \n"
|
|
"incb %[dst_ptr] \n"
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.b, wzr, %w[dst_width] \n"
|
|
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
|
|
"st1b {z1.b}, p0, [%[dst_ptr]] \n"
|
|
|
|
"99: \n"
|
|
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
|
|
[dst_ptr] "+r"(dst), // %[dst_ptr]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "memory", "cc", "z0", "z1", "p0");
|
|
}
|
|
|
|
__arm_locally_streaming void ScaleRowDown2Linear_SME(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
(void)src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cntb %x[vl] \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"1: \n"
|
|
"ptrue p0.b \n"
|
|
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
|
|
"incb %[src_ptr], all, mul #2 \n"
|
|
"urhadd z0.b, p0/m, z0.b, z1.b \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"st1b {z0.b}, p0, [%[dst_ptr]] \n"
|
|
"incb %[dst_ptr] \n"
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.b, wzr, %w[dst_width] \n"
|
|
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
|
|
"urhadd z0.b, p0/m, z0.b, z1.b \n"
|
|
"st1b {z0.b}, p0, [%[dst_ptr]] \n"
|
|
|
|
"99: \n"
|
|
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
|
|
[dst_ptr] "+r"(dst), // %[dst_ptr]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "memory", "cc", "z0", "z1", "p0");
|
|
}
|
|
|
|
#define SCALEROWDOWN2BOX_SVE \
|
|
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n" \
|
|
"ld2b {z2.b, z3.b}, p0/z, [%[src2_ptr]] \n" \
|
|
"incb %[src_ptr], all, mul #2 \n" \
|
|
"incb %[src2_ptr], all, mul #2 \n" \
|
|
"uaddlb z4.h, z0.b, z1.b \n" \
|
|
"uaddlt z5.h, z0.b, z1.b \n" \
|
|
"uaddlb z6.h, z2.b, z3.b \n" \
|
|
"uaddlt z7.h, z2.b, z3.b \n" \
|
|
"add z4.h, z4.h, z6.h \n" \
|
|
"add z5.h, z5.h, z7.h \n" \
|
|
"rshrnb z0.b, z4.h, #2 \n" \
|
|
"rshrnt z0.b, z5.h, #2 \n" \
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n" \
|
|
"st1b {z0.b}, p0, [%[dst_ptr]] \n" \
|
|
"incb %[dst_ptr] \n"
|
|
|
|
__arm_locally_streaming void ScaleRowDown2Box_SME(const uint8_t* src_ptr,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
const uint8_t* src2_ptr = src_ptr + src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cntb %x[vl] \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"ptrue p0.b \n"
|
|
"1: \n" //
|
|
SCALEROWDOWN2BOX_SVE
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.b, wzr, %w[dst_width] \n" //
|
|
SCALEROWDOWN2BOX_SVE
|
|
|
|
"99: \n"
|
|
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
|
|
[src2_ptr] "+r"(src2_ptr), // %[src2_ptr]
|
|
[dst_ptr] "+r"(dst), // %[dst_ptr]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "memory", "cc", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0");
|
|
}
|
|
|
|
#undef SCALEROWDOWN2BOX_SVE
|
|
|
|
__arm_locally_streaming void ScaleUVRowDown2_SME(const uint8_t* src_uv,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_uv,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
(void)src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cnth %x[vl] \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"1: \n"
|
|
"ptrue p0.b \n"
|
|
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
|
|
"incb %[src_uv], all, mul #2 \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"st1h {z1.h}, p0, [%[dst_uv]] \n"
|
|
"incb %[dst_uv] \n"
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.h, wzr, %w[dst_width] \n"
|
|
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
|
|
"st1h {z1.h}, p0, [%[dst_uv]] \n"
|
|
|
|
"99: \n"
|
|
: [src_uv] "+r"(src_uv), // %[src_uv]
|
|
[dst_uv] "+r"(dst_uv), // %[dst_uv]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "memory", "cc", "z0", "z1", "p0");
|
|
}
|
|
|
|
__arm_locally_streaming void ScaleUVRowDown2Linear_SME(const uint8_t* src_uv,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_uv,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
(void)src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cnth %x[vl] \n"
|
|
"ptrue p1.b \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"ptrue p0.h \n"
|
|
"1: \n"
|
|
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
|
|
"incb %[src_uv], all, mul #2 \n"
|
|
"urhadd z0.b, p1/m, z0.b, z1.b \n"
|
|
"st1h {z0.h}, p0, [%[dst_uv]] \n"
|
|
"incb %[dst_uv], all, mul #1 \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.h, wzr, %w[dst_width] \n"
|
|
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
|
|
"urhadd z0.b, p1/m, z0.b, z1.b \n"
|
|
"st1h {z0.h}, p0, [%[dst_uv]] \n"
|
|
|
|
"99: \n"
|
|
: [src_uv] "+r"(src_uv), // %[src_uv]
|
|
[dst_uv] "+r"(dst_uv), // %[dst_uv]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "z0", "z1", "p0", "p1");
|
|
}
|
|
|
|
#define SCALEUVROWDOWN2BOX_SVE \
|
|
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n" \
|
|
"ld2h {z2.h, z3.h}, p0/z, [%[src2_uv]] \n" \
|
|
"incb %[src_uv], all, mul #2 \n" \
|
|
"incb %[src2_uv], all, mul #2 \n" \
|
|
"uaddlb z4.h, z0.b, z1.b \n" \
|
|
"uaddlt z5.h, z0.b, z1.b \n" \
|
|
"uaddlb z6.h, z2.b, z3.b \n" \
|
|
"uaddlt z7.h, z2.b, z3.b \n" \
|
|
"add z4.h, z4.h, z6.h \n" \
|
|
"add z5.h, z5.h, z7.h \n" \
|
|
"rshrnb z0.b, z4.h, #2 \n" \
|
|
"rshrnt z0.b, z5.h, #2 \n" \
|
|
"st1h {z0.h}, p0, [%[dst_uv]] \n" \
|
|
"incb %[dst_uv], all, mul #1 \n"
|
|
|
|
__arm_locally_streaming void ScaleUVRowDown2Box_SME(const uint8_t* src_uv,
|
|
ptrdiff_t src_stride,
|
|
uint8_t* dst_uv,
|
|
int dst_width) {
|
|
// Streaming-SVE only, no use of ZA tile.
|
|
const uint8_t* src2_uv = src_uv + src_stride;
|
|
int vl;
|
|
asm volatile(
|
|
"cnth %x[vl] \n"
|
|
"ptrue p1.b \n"
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.lt 2f \n"
|
|
|
|
"ptrue p0.h \n"
|
|
"1: \n" //
|
|
SCALEUVROWDOWN2BOX_SVE
|
|
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.ge 1b \n"
|
|
|
|
"2: \n"
|
|
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
|
|
"b.eq 99f \n"
|
|
|
|
"whilelt p0.h, wzr, %w[dst_width] \n" //
|
|
SCALEUVROWDOWN2BOX_SVE
|
|
|
|
"99: \n"
|
|
: [src_uv] "+r"(src_uv), // %[src_uv]
|
|
[src2_uv] "+r"(src2_uv), // %[src2_uv]
|
|
[dst_uv] "+r"(dst_uv), // %[dst_uv]
|
|
[dst_width] "+r"(dst_width), // %[dst_width]
|
|
[vl] "=r"(vl) // %[vl]
|
|
:
|
|
: "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0", "p1");
|
|
}
|
|
|
|
#undef SCALEUVROWDOWN2BOX_SVE
|
|
|
|
#endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
|
|
// defined(__aarch64__)
|
|
|
|
#ifdef __cplusplus
|
|
} // extern "C"
|
|
} // namespace libyuv
|
|
#endif
|