Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Add fast_matrix_mul_4x4_lsx function for LoongArch64

Browse source
This commit is contained in:
Wang Xiang 2025-02-08 15:53:00 +08:00
parent d5bd7d24aa
commit 66f5ac9897
5 changed files with 48 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
CMakeLists.txt
View file

@ -69,6 +69,8 @@ if(CMAKE_SYSTEM_PROCESSOR)
set(RISCV64_DEVICE ON)
elseif(${CMAKE_SYSTEM_PROCESSOR} MATCHES "^loongarch64")
set(LOONGARCH64_DEVICE ON)
add_compile_options(-mlsx)
add_compile_options(-mlasx)
else()
message("Unknown CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}")
endif()

6
Common/Math/SIMDHeaders.h
View file

@ -31,6 +31,12 @@
#endif
#endif
#if PPSSPP_ARCH(LOONGARCH64)
#if PPSSPP_ARCH(LOONGARCH64_LSX)
#include <lsxintrin.h>
#endif
#endif
// Basic types
#if PPSSPP_ARCH(ARM64_NEON)

36
Common/Math/fast/fast_matrix.c
View file

@ -22,6 +22,42 @@ void fast_matrix_mul_4x4_sse(float *dest, const float *a, const float *b) {
}
}
#elif PPSSPP_ARCH(LOONGARCH64_LSX)
typedef union
{
int32_t i;
float f;
} FloatInt;
static __m128 __lsx_vreplfr2vr_s(float val)
{
FloatInt tmpval = {.f = val};
return (__m128)__lsx_vreplgr2vr_w(tmpval.i);
}
void fast_matrix_mul_4x4_lsx(float *dest, const float *a, const float *b) {
__m128 a_col_1 = (__m128)__lsx_vld(a, 0);
__m128 a_col_2 = (__m128)__lsx_vld(a + 4, 0);
__m128 a_col_3 = (__m128)__lsx_vld(a + 8, 0);
__m128 a_col_4 = (__m128)__lsx_vld(a + 12, 0);
for (int i = 0; i < 16; i += 4) {
__m128 b1 = __lsx_vreplfr2vr_s(b[i]);
__m128 b2 = __lsx_vreplfr2vr_s(b[i + 1]);
__m128 b3 = __lsx_vreplfr2vr_s(b[i + 2]);
__m128 b4 = __lsx_vreplfr2vr_s(b[i + 3]);
__m128 result = __lsx_vfmul_s(a_col_1, b1);
result = __lsx_vfmadd_s(a_col_2, b2, result);
result = __lsx_vfmadd_s(a_col_3, b3, result);
result = __lsx_vfmadd_s(a_col_4, b4, result);
__lsx_vst(result, &dest[i], 0);
}
}
#elif PPSSPP_ARCH(ARM_NEON)
// From https://developer.arm.com/documentation/102467/0100/Matrix-multiplication-example

3
Common/Math/fast/fast_matrix.h
View file

@ -11,12 +11,15 @@ extern "C" {
extern void fast_matrix_mul_4x4_c(float *dest, const float *a, const float *b);
extern void fast_matrix_mul_4x4_neon(float *dest, const float *a, const float *b);
extern void fast_matrix_mul_4x4_sse(float *dest, const float *a, const float *b);
extern void fast_matrix_mul_4x4_lsx(float *dest, const float *a, const float *b);
#if PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)
// Hard link to SSE implementations on x86/amd64
#define fast_matrix_mul_4x4 fast_matrix_mul_4x4_sse
#elif PPSSPP_ARCH(ARM_NEON)
#define fast_matrix_mul_4x4 fast_matrix_mul_4x4_neon
#elif PPSSPP_ARCH(LOONGARCH64_LSX)
#define fast_matrix_mul_4x4 fast_matrix_mul_4x4_lsx
#else
#define fast_matrix_mul_4x4 fast_matrix_mul_4x4_c
#endif

1
ppsspp_config.h
View file

@ -81,6 +81,7 @@
//https://github.com/gcc-mirror/gcc/blob/master/gcc/config/loongarch/loongarch-c.cc
#define PPSSPP_ARCH_LOONGARCH64 1
#define PPSSPP_ARCH_64BIT 1
#define PPSSPP_ARCH_LOONGARCH64_LSX 1
#endif
// PLATFORM defines