diff options
| author | WANG Xuerui <git@xen0n.name> | 2023-09-06 22:53:55 +0800 |
|---|---|---|
| committer | Huacai Chen <chenhuacai@loongson.cn> | 2023-09-06 22:53:55 +0800 |
| commit | 75ded18a5e8e51ca2d26d55f010d60ae9aab652c (patch) | |
| tree | 29464a5abd1b2b6f7141a26374d45d03eaf17078 /arch/loongarch/lib/xor_simd.c | |
| parent | 2478e4b7593a2a55073a4a6bf23dc885c19befd8 (diff) | |
| download | linux-stable-75ded18a5e8e51ca2d26d55f010d60ae9aab652c.tar.gz linux-stable-75ded18a5e8e51ca2d26d55f010d60ae9aab652c.tar.bz2 linux-stable-75ded18a5e8e51ca2d26d55f010d60ae9aab652c.zip | |
LoongArch: Add SIMD-optimized XOR routines
Add LSX and LASX implementations of xor operations, operating on 64
bytes (one L1 cache line) at a time, for a balance between memory
utilization and instruction mix. Huacai confirmed that all future
LoongArch implementations by Loongson (that we care) will likely also
feature 64-byte cache lines, and experiments show no throughput
improvement with further unrolling.
Performance numbers measured during system boot on a 3A5000 @ 2.5GHz:
> 8regs : 12702 MB/sec
> 8regs_prefetch : 10920 MB/sec
> 32regs : 12686 MB/sec
> 32regs_prefetch : 10918 MB/sec
> lsx : 17589 MB/sec
> lasx : 26116 MB/sec
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: WANG Xuerui <git@xen0n.name>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
Diffstat (limited to 'arch/loongarch/lib/xor_simd.c')
| -rw-r--r-- | arch/loongarch/lib/xor_simd.c | 93 |
1 files changed, 93 insertions, 0 deletions
diff --git a/arch/loongarch/lib/xor_simd.c b/arch/loongarch/lib/xor_simd.c new file mode 100644 index 000000000000..84cd24b728c4 --- /dev/null +++ b/arch/loongarch/lib/xor_simd.c @@ -0,0 +1,93 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * LoongArch SIMD XOR operations + * + * Copyright (C) 2023 WANG Xuerui <git@xen0n.name> + */ + +#include "xor_simd.h" + +/* + * Process one cache line (64 bytes) per loop. This is assuming all future + * popular LoongArch cores are similar performance-characteristics-wise to the + * current models. + */ +#define LINE_WIDTH 64 + +#ifdef CONFIG_CPU_HAS_LSX + +#define LD(reg, base, offset) \ + "vld $vr" #reg ", %[" #base "], " #offset "\n\t" +#define ST(reg, base, offset) \ + "vst $vr" #reg ", %[" #base "], " #offset "\n\t" +#define XOR(dj, k) "vxor.v $vr" #dj ", $vr" #dj ", $vr" #k "\n\t" + +#define LD_INOUT_LINE(base) \ + LD(0, base, 0) \ + LD(1, base, 16) \ + LD(2, base, 32) \ + LD(3, base, 48) + +#define LD_AND_XOR_LINE(base) \ + LD(4, base, 0) \ + LD(5, base, 16) \ + LD(6, base, 32) \ + LD(7, base, 48) \ + XOR(0, 4) \ + XOR(1, 5) \ + XOR(2, 6) \ + XOR(3, 7) + +#define ST_LINE(base) \ + ST(0, base, 0) \ + ST(1, base, 16) \ + ST(2, base, 32) \ + ST(3, base, 48) + +#define XOR_FUNC_NAME(nr) __xor_lsx_##nr +#include "xor_template.c" + +#undef LD +#undef ST +#undef XOR +#undef LD_INOUT_LINE +#undef LD_AND_XOR_LINE +#undef ST_LINE +#undef XOR_FUNC_NAME + +#endif /* CONFIG_CPU_HAS_LSX */ + +#ifdef CONFIG_CPU_HAS_LASX + +#define LD(reg, base, offset) \ + "xvld $xr" #reg ", %[" #base "], " #offset "\n\t" +#define ST(reg, base, offset) \ + "xvst $xr" #reg ", %[" #base "], " #offset "\n\t" +#define XOR(dj, k) "xvxor.v $xr" #dj ", $xr" #dj ", $xr" #k "\n\t" + +#define LD_INOUT_LINE(base) \ + LD(0, base, 0) \ + LD(1, base, 32) + +#define LD_AND_XOR_LINE(base) \ + LD(2, base, 0) \ + LD(3, base, 32) \ + XOR(0, 2) \ + XOR(1, 3) + +#define ST_LINE(base) \ + ST(0, base, 0) \ + ST(1, base, 32) + +#define XOR_FUNC_NAME(nr) __xor_lasx_##nr +#include "xor_template.c" + +#undef LD +#undef ST +#undef XOR +#undef LD_INOUT_LINE +#undef LD_AND_XOR_LINE +#undef ST_LINE +#undef XOR_FUNC_NAME + +#endif /* CONFIG_CPU_HAS_LASX */ |