coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Add SME impls of ScaleARGBRowDown2{,Linear,Box}

Browse Source 
Mostly just straightforward copies of the Neon code ported to
Streaming-SVE, these follow the same pattern as the prior ScaleRowDown2
and ScaleUVRowDown2 SME kernels, but operating on 32-bit ARGB tuples
rather than 8-bit data or 16-bit UV tuples.

These is no benefit from this kernel when the SVE vector length is only
128 bits, so skip writing a non-streaming SVE implementation.

Change-Id: I15600c2498cc592f5ea1d97b78fafec327de7947
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/6070783
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Justin Green <greenjustin@google.com>
This commit is contained in:
George Steed 2024-08-27 17:43:25 +01:00 committed by Frank Barchard
parent 7391559cb4
commit 15f2ae7d70
3 changed files with 159 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
include/libyuv/scale_row.h
View File

@ -128,6 +128,7 @@ extern "C" {
#define HAS_SCALEUVROWDOWN2_SME #define HAS_SCALEUVROWDOWN2_SME
#define HAS_SCALEUVROWDOWN2LINEAR_SME #define HAS_SCALEUVROWDOWN2LINEAR_SME
#define HAS_SCALEUVROWDOWN2BOX_SME #define HAS_SCALEUVROWDOWN2BOX_SME
#define HAS_SCALEARGBROWDOWN2_SME
#endif #endif
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) #if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
@ -958,14 +959,26 @@ void ScaleARGBRowDown2_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst, uint8_t* dst,
int dst_width); int dst_width);
void ScaleARGBRowDown2_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width);
void ScaleARGBRowDown2Linear_NEON(const uint8_t* src_argb, void ScaleARGBRowDown2Linear_NEON(const uint8_t* src_argb,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_argb, uint8_t* dst_argb,
int dst_width); int dst_width);
void ScaleARGBRowDown2Linear_SME(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width);
void ScaleARGBRowDown2Box_NEON(const uint8_t* src_ptr, void ScaleARGBRowDown2Box_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst, uint8_t* dst,
int dst_width); int dst_width);
void ScaleARGBRowDown2Box_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width);
void ScaleARGBRowDown2_RVV(const uint8_t* src_argb, void ScaleARGBRowDown2_RVV(const uint8_t* src_argb,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_argb, uint8_t* dst_argb,

13
source/scale_argb.cc
View File

@ -99,6 +99,14 @@ static void ScaleARGBDown2(int src_width,
} }
} }
#endif #endif
#if defined(HAS_SCALEARGBROWDOWN2_SME)
if (TestCpuFlag(kCpuHasSME)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_SME
: filtering == kFilterLinear
? ScaleARGBRowDown2Linear_SME
: ScaleARGBRowDown2Box_SME;
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_MSA) #if defined(HAS_SCALEARGBROWDOWN2_MSA)
if (TestCpuFlag(kCpuHasMSA)) { if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBRowDown2 = ScaleARGBRowDown2 =
@ -202,6 +210,11 @@ static int ScaleARGBDown4Box(int src_width,
} }
} }
#endif #endif
#if defined(HAS_SCALEARGBROWDOWN2_SME)
if (TestCpuFlag(kCpuHasSME)) {
ScaleARGBRowDown2 = ScaleARGBRowDown2Box_SME;
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_RVV) #if defined(HAS_SCALEARGBROWDOWN2_RVV)
if (TestCpuFlag(kCpuHasRVV)) { if (TestCpuFlag(kCpuHasRVV)) {
ScaleARGBRowDown2 = ScaleARGBRowDown2Box_RVV; ScaleARGBRowDown2 = ScaleARGBRowDown2Box_RVV;

133
source/scale_sme.cc
View File

@ -283,6 +283,139 @@ __arm_locally_streaming void ScaleUVRowDown2Box_SME(const uint8_t* src_uv,
#undef SCALEUVROWDOWN2BOX_SVE #undef SCALEUVROWDOWN2BOX_SVE
__arm_locally_streaming void ScaleARGBRowDown2_SME(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.b \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"incb %[src_argb], all, mul #2 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1w {z1.s}, p0, [%[dst_argb]] \n"
"incb %[dst_argb] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.s, wzr, %w[dst_width] \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"st1w {z1.s}, p0, [%[dst_argb]] \n"
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleARGBRowDown2Linear_SME(
const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.s \n"
"1: \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"incb %[src_argb], all, mul #2 \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1w {z0.s}, p0, [%[dst_argb]] \n"
"incb %[dst_argb], 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.s, wzr, %w[dst_width] \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1w {z0.s}, p0, [%[dst_argb]] \n"
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0", "p1");
}
#define SCALEARGBROWDOWN2BOX_SVE \
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n" \
"ld2w {z2.s, z3.s}, p0/z, [%[src2_argb]] \n" \
"incb %[src_argb], all, mul #2 \n" \
"incb %[src2_argb], 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" \
"st1w {z0.s}, p0, [%[dst_argb]] \n" \
"incb %[dst_argb], all, mul #1 \n"
__arm_locally_streaming void ScaleARGBRowDown2Box_SME(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
const uint8_t* src2_argb = src_argb + src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.s \n"
"1: \n" //
SCALEARGBROWDOWN2BOX_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.s, wzr, %w[dst_width] \n" //
SCALEARGBROWDOWN2BOX_SVE
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[src2_argb] "+r"(src2_argb), // %[src2_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0",
"p1");
}
#endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && #endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
// defined(__aarch64__) // defined(__aarch64__)