coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 08:46:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Add SME implementation of ScaleRowDown2Box

Browse Source 
There is no benefit from an SVE version of this kernel for devices with
an SVE vector length of 128-bits, so skip directly to SME instead.  We
do not use the ZA tile here, so this is a purely streaming-SVE (SSVE)
implementation.

Change-Id: I5021aeda30f4c5f1aa4cc6326c8d7886851d2c09
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5913885
Reviewed-by: Justin Green <greenjustin@google.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
George Steed 2024-08-27 14:56:13 +01:00 committed by Frank Barchard
parent b0f72309c6
commit aec4b4e22e
3 changed files with 60 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
include/libyuv/scale_row.h
View File

@ -1424,6 +1424,10 @@ void ScaleRowDown2Box_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width);
void ScaleRowDown2Box_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width);
void ScaleRowDown4_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride,

7
source/scale.cc
View File

@ -76,10 +76,9 @@ static void ScalePlaneDown2(int src_width,
#endif
#if defined(HAS_SCALEROWDOWN2_SME)
if (TestCpuFlag(kCpuHasSME)) {
if (filtering == kFilterNone || filtering == kFilterLinear) {
ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_SME
: ScaleRowDown2Linear_SME;
}
ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_SME
: filtering == kFilterLinear ? ScaleRowDown2Linear_SME
: ScaleRowDown2Box_SME;
}
#endif
#if defined(HAS_SCALEROWDOWN2_SSSE3)

53
source/scale_sme.cc
View File

@ -96,6 +96,59 @@ __arm_locally_streaming void ScaleRowDown2Linear_SME(const uint8_t* src_ptr,
: "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
#endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
// defined(__aarch64__)