Mirror/ppsspp
1
0 Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-10-01 14:58:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make fast_matrix_mul_4x4 inlineable

Browse Source
This commit is contained in:
Henrik Rydgård committed 2026-01-30 14:10:32 +01:00
1 parent 19e78ed122
commit 7d987cd78b
13 files changed
+152 -208
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

CMakeLists.txt
-1
View File
@@ -806,7 +806,6 @@ add_library(Common STATIC
Common/Input/KeyCodes.h
Common/Input/InputState.cpp
Common/Input/InputState.h
Common/Math/fast/fast_matrix.c
Common/Math/SIMDHeaders.h
Common/Math/SIMDHeaders.h
Common/Math/curves.cpp
Common/Common.h
+9
View File
@@ -90,3 +90,12 @@
// Easy way to printf string_views (note: The formatting specifier is "%.*s", not "%s")
#define STR_VIEW(sv) (int)(sv).size(), (sv).data()
// Restrict qualifier
#if defined(_MSC_VER)
#define RESTRICT __restrict
#elif defined(__GNUC__) || defined(__clang__)
#define RESTRICT __restrict__
#else
#define RESTRICT
#endif
Common/Common.vcxproj
-1
View File
@@ -918,7 +918,6 @@
<ClCompile Include="LoongArch64Emitter.cpp" />
<ClCompile Include="Math\curves.cpp" />
<ClCompile Include="Math\expression_parser.cpp" />
<ClCompile Include="Math\fast\fast_matrix.c" />
<ClCompile Include="Math\lin\matrix4x4.cpp" />
<ClCompile Include="Math\lin\vec3.cpp" />
<ClCompile Include="Math\math_util.cpp" />
Common/Common.vcxproj.filters
-3
View File
@@ -861,9 +861,6 @@
<ClCompile Include="Math\lin\vec3.cpp">
<Filter>Math\lin</Filter>
</ClCompile>
<ClCompile Include="Math\fast\fast_matrix.c">
<Filter>Math\fast</Filter>
</ClCompile>
<ClCompile Include="Data\Format\RIFF.cpp">
<Filter>Data\Format</Filter>
</ClCompile>
Common/Math/CrossSIMD.h
+1 -1
View File
@@ -198,7 +198,7 @@ struct Vec4F32 {
return Vec4F32 { _mm_mul_ps(_mm_cvtepi32_ps(value32), _mm_set1_ps(1.0f / 128.0f)) };
}
static Vec4F32 LoadS16Norm(const int16_t *src) { // Divides by 32768.0f
__m128i bits = _mm_castpd_si128(_mm_load_sd((const double *)src));
__m128i bits = _mm_loadl_epi64((const __m128i*)src);
// Sign extension. A bit ugly without SSE4.
bits = _mm_srai_epi32(_mm_unpacklo_epi16(bits, bits), 16);
return Vec4F32 { _mm_mul_ps(_mm_cvtepi32_ps(bits), _mm_set1_ps(1.0f / 32768.0f)) };
Common/Math/fast/fast_matrix.c
-159
View File
@@ -1,159 +0,0 @@
#include "ppsspp_config.h"
#include "Common/Math/SIMDHeaders.h"
#include "fast_matrix.h"
#if PPSSPP_ARCH(SSE2)
void fast_matrix_mul_4x4_sse(float *dest, const float *a, const float *b) {
int i;
__m128 a_col_1 = _mm_loadu_ps(a);
__m128 a_col_2 = _mm_loadu_ps(&a[4]);
__m128 a_col_3 = _mm_loadu_ps(&a[8]);
__m128 a_col_4 = _mm_loadu_ps(&a[12]);
for (i = 0; i < 16; i += 4) {
__m128 r_col = _mm_mul_ps(a_col_1, _mm_set1_ps(b[i]));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_2, _mm_set1_ps(b[i + 1])));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_3, _mm_set1_ps(b[i + 2])));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_4, _mm_set1_ps(b[i + 3])));
_mm_storeu_ps(&dest[i], r_col);
}
}
#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
void fast_matrix_mul_4x4_neon(float *C, const float *A, const float *B) {
// these are the columns A
float32x4_t A0;
float32x4_t A1;
float32x4_t A2;
float32x4_t A3;
// these are the columns B
float32x4_t B0;
float32x4_t B1;
float32x4_t B2;
float32x4_t B3;
// these are the columns C
float32x4_t C0;
float32x4_t C1;
float32x4_t C2;
float32x4_t C3;
A0 = vld1q_f32(A);
A1 = vld1q_f32(A + 4);
A2 = vld1q_f32(A + 8);
A3 = vld1q_f32(A + 12);
// Multiply accumulate in 4x1 blocks, i.e. each column in C
B0 = vld1q_f32(B);
C0 = vmulq_laneq_f32(A0, B0, 0);
C0 = vfmaq_laneq_f32(C0, A1, B0, 1);
C0 = vfmaq_laneq_f32(C0, A2, B0, 2);
C0 = vfmaq_laneq_f32(C0, A3, B0, 3);
vst1q_f32(C, C0);
B1 = vld1q_f32(B + 4);
C1 = vmulq_laneq_f32(A0, B1, 0);
C1 = vfmaq_laneq_f32(C1, A1, B1, 1);
C1 = vfmaq_laneq_f32(C1, A2, B1, 2);
C1 = vfmaq_laneq_f32(C1, A3, B1, 3);
vst1q_f32(C + 4, C1);
B2 = vld1q_f32(B + 8);
C2 = vmulq_laneq_f32(A0, B2, 0);
C2 = vfmaq_laneq_f32(C2, A1, B2, 1);
C2 = vfmaq_laneq_f32(C2, A2, B2, 2);
C2 = vfmaq_laneq_f32(C2, A3, B2, 3);
vst1q_f32(C + 8, C2);
B3 = vld1q_f32(B + 12);
C3 = vmulq_laneq_f32(A0, B3, 0);
C3 = vfmaq_laneq_f32(C3, A1, B3, 1);
C3 = vfmaq_laneq_f32(C3, A2, B3, 2);
C3 = vfmaq_laneq_f32(C3, A3, B3, 3);
vst1q_f32(C + 12, C3);
}
#else
#define xx 0
#define xy 1
#define xz 2
#define xw 3
#define yx 4
#define yy 5
#define yz 6
#define yw 7
#define zx 8
#define zy 9
#define zz 10
#define zw 11
#define wx 12
#define wy 13
#define wz 14
#define ww 15
void fast_matrix_mul_4x4_c(float *dest, const float *a, const float *b) {
dest[xx] = b[xx] * a[xx] + b[xy] * a[yx] + b[xz] * a[zx] + b[xw] * a[wx];
dest[xy] = b[xx] * a[xy] + b[xy] * a[yy] + b[xz] * a[zy] + b[xw] * a[wy];
dest[xz] = b[xx] * a[xz] + b[xy] * a[yz] + b[xz] * a[zz] + b[xw] * a[wz];
dest[xw] = b[xx] * a[xw] + b[xy] * a[yw] + b[xz] * a[zw] + b[xw] * a[ww];
dest[yx] = b[yx] * a[xx] + b[yy] * a[yx] + b[yz] * a[zx] + b[yw] * a[wx];
dest[yy] = b[yx] * a[xy] + b[yy] * a[yy] + b[yz] * a[zy] + b[yw] * a[wy];
dest[yz] = b[yx] * a[xz] + b[yy] * a[yz] + b[yz] * a[zz] + b[yw] * a[wz];
dest[yw] = b[yx] * a[xw] + b[yy] * a[yw] + b[yz] * a[zw] + b[yw] * a[ww];
dest[zx] = b[zx] * a[xx] + b[zy] * a[yx] + b[zz] * a[zx] + b[zw] * a[wx];
dest[zy] = b[zx] * a[xy] + b[zy] * a[yy] + b[zz] * a[zy] + b[zw] * a[wy];
dest[zz] = b[zx] * a[xz] + b[zy] * a[yz] + b[zz] * a[zz] + b[zw] * a[wz];
dest[zw] = b[zx] * a[xw] + b[zy] * a[yw] + b[zz] * a[zw] + b[zw] * a[ww];
dest[wx] = b[wx] * a[xx] + b[wy] * a[yx] + b[wz] * a[zx] + b[ww] * a[wx];
dest[wy] = b[wx] * a[xy] + b[wy] * a[yy] + b[wz] * a[zy] + b[ww] * a[wy];
dest[wz] = b[wx] * a[xz] + b[wy] * a[yz] + b[wz] * a[zz] + b[ww] * a[wz];
dest[ww] = b[wx] * a[xw] + b[wy] * a[yw] + b[wz] * a[zw] + b[ww] * a[ww];
}
#endif
Common/Math/fast/fast_matrix.h
+135 -18
View File
@@ -2,28 +2,145 @@
#include "ppsspp_config.h"
#ifdef __cplusplus
extern "C" {
#endif
// A mini library of 4x4 matrix muls.
#include "Common/Math/SIMDHeaders.h"
#include "Common/Common.h"
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(SSE2)
inline void fast_matrix_mul_4x4(float *dest, const float *a, const float *b) {
int i;
__m128 a_col_1 = _mm_loadu_ps(a);
__m128 a_col_2 = _mm_loadu_ps(&a[4]);
__m128 a_col_3 = _mm_loadu_ps(&a[8]);
__m128 a_col_4 = _mm_loadu_ps(&a[12]);
for (i = 0; i < 16; i += 4) {
__m128 r_col = _mm_mul_ps(a_col_1, _mm_set1_ps(b[i]));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_2, _mm_set1_ps(b[i + 1])));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_3, _mm_set1_ps(b[i + 2])));
r_col = _mm_add_ps(r_col, _mm_mul_ps(a_col_4, _mm_set1_ps(b[i + 3])));
_mm_storeu_ps(&dest[i], r_col);
}
}
#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
inline __m128 __lsx_vreplfr2vr_s(float val) {
typedef union {
int32_t i;
float f;
} FloatInt;
FloatInt tmpval = {.f = val};
return (__m128)__lsx_vreplgr2vr_w(tmpval.i);
}
inline void fast_matrix_mul_4x4(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)
#ifdef B0
#undef B0
#endif
#ifdef __cplusplus
} // extern "C"
// From https://developer.arm.com/documentation/102467/0100/Matrix-multiplication-example
inline void fast_matrix_mul_4x4(float *C, const float *A, const float *B) {
// these are the columns A
float32x4_t A0;
float32x4_t A1;
float32x4_t A2;
float32x4_t A3;
// these are the columns B
float32x4_t B0;
float32x4_t B1;
float32x4_t B2;
float32x4_t B3;
// these are the columns C
float32x4_t C0;
float32x4_t C1;
float32x4_t C2;
float32x4_t C3;
A0 = vld1q_f32(A);
A1 = vld1q_f32(A + 4);
A2 = vld1q_f32(A + 8);
A3 = vld1q_f32(A + 12);
// Multiply accumulate in 4x1 blocks, i.e. each column in C
B0 = vld1q_f32(B);
B1 = vld1q_f32(B + 4);
B2 = vld1q_f32(B + 8);
B3 = vld1q_f32(B + 12);
C0 = vmulq_laneq_f32(A0, B0, 0);
C0 = vfmaq_laneq_f32(C0, A1, B0, 1);
C0 = vfmaq_laneq_f32(C0, A2, B0, 2);
C0 = vfmaq_laneq_f32(C0, A3, B0, 3);
vst1q_f32(C, C0);
C1 = vmulq_laneq_f32(A0, B1, 0);
C1 = vfmaq_laneq_f32(C1, A1, B1, 1);
C1 = vfmaq_laneq_f32(C1, A2, B1, 2);
C1 = vfmaq_laneq_f32(C1, A3, B1, 3);
vst1q_f32(C + 4, C1);
C2 = vmulq_laneq_f32(A0, B2, 0);
C2 = vfmaq_laneq_f32(C2, A1, B2, 1);
C2 = vfmaq_laneq_f32(C2, A2, B2, 2);
C2 = vfmaq_laneq_f32(C2, A3, B2, 3);
vst1q_f32(C + 8, C2);
C3 = vmulq_laneq_f32(A0, B3, 0);
C3 = vfmaq_laneq_f32(C3, A1, B3, 1);
C3 = vfmaq_laneq_f32(C3, A2, B3, 2);
C3 = vfmaq_laneq_f32(C3, A3, B3, 3);
vst1q_f32(C + 12, C3);
}
#else
inline void fast_matrix_mul_4x4(float * RESTRICT dest, const float * RESTRICT a, const float * RESTRICT b) {
dest[0] = b[0] * a[0] + b[1] * a[4] + b[2] * a[8] + b[3] * a[12];
dest[1] = b[0] * a[1] + b[1] * a[5] + b[2] * a[9] + b[3] * a[13];
dest[2] = b[0] * a[2] + b[1] * a[6] + b[2] * a[10] + b[3] * a[14];
dest[3] = b[0] * a[3] + b[1] * a[7] + b[2] * a[11] + b[3] * a[15];
dest[4] = b[4] * a[0] + b[5] * a[4] + b[6] * a[8] + b[7] * a[12];
dest[5] = b[4] * a[1] + b[5] * a[5] + b[6] * a[9] + b[7] * a[13];
dest[6] = b[4] * a[2] + b[5] * a[6] + b[6] * a[10] + b[7] * a[14];
dest[7] = b[4] * a[3] + b[5] * a[7] + b[6] * a[11] + b[7] * a[15];
dest[8] = b[8] * a[0] + b[9] * a[4] + b[10] * a[8] + b[11] * a[12];
dest[9] = b[8] * a[1] + b[9] * a[5] + b[10] * a[9] + b[11] * a[13];
dest[10] = b[8] * a[2] + b[9] * a[6] + b[10] * a[10] + b[11] * a[14];
dest[11] = b[8] * a[3] + b[9] * a[7] + b[10] * a[11] + b[11] * a[15];
dest[12] = b[12] * a[0] + b[13] * a[4] + b[14] * a[8] + b[15] * a[12];
dest[13] = b[12] * a[1] + b[13] * a[5] + b[14] * a[9] + b[15] * a[13];
dest[14] = b[12] * a[2] + b[13] * a[6] + b[14] * a[10] + b[15] * a[14];
dest[15] = b[12] * a[3] + b[13] * a[7] + b[14] * a[11] + b[15] * a[15];
}
#endif
Core/KeyMap.cpp
+2 -1
View File
@@ -848,7 +848,8 @@ void NotifyPadConnected(InputDeviceID deviceId, std::string_view name) {
g_padNames[deviceId] = name;
// Don't notify within the first 5 seconds, to avoid notification spam on startup.
if (time_now_d() >= 5.0) {
// Also for some reason we get some strange things on Android... "Virtual"?
if (time_now_d() >= 5.0 && name != "Virtual") {
auto co = GetI18NCategory(I18NCat::CONTROLS);
g_OSD.Show(OSDType::MESSAGE_SUCCESS, ApplySafeSubstitutions("%1: %2", co->T("Game controller connected"), name), "", "I_CONTROLLER", 2.0f, "controller_connected");
}
GPU/Common/TextureCacheCommon.cpp
+1 -1
View File
@@ -802,7 +802,7 @@ bool TextureCacheCommon::GetBestFramebufferCandidate(const TextureDefinition &en
if (bestIndex != -1) {
if (logging) {
WARN_LOG(Log::G3D, "Chose candidate %d:\n%s (%dx%d)\n", (int)bestIndex, candidates[bestIndex].ToString().c_str());
WARN_LOG(Log::G3D, "Chose candidate %d:\n%s", (int)bestIndex, candidates[bestIndex].ToString().c_str());
}
*bestCandidate = candidates[bestIndex];
return true;
UWP/CommonUWP/CommonUWP.vcxproj
+1 -6
View File
@@ -330,7 +330,6 @@
<ClCompile Include="..\..\Common\Input\InputState.cpp" />
<ClCompile Include="..\..\Common\Math\curves.cpp" />
<ClCompile Include="..\..\Common\Math\expression_parser.cpp" />
<ClCompile Include="..\..\Common\Math\fast\fast_matrix.c" />
<ClCompile Include="..\..\Common\Math\lin\matrix4x4.cpp" />
<ClCompile Include="..\..\Common\Math\lin\vec3.cpp" />
<ClCompile Include="..\..\Common\Math\math_util.cpp" />
@@ -461,10 +460,6 @@
<None Include="..\..\ext\basis_universal\basisu_transcoder_tables_pvrtc2_alpha_33.inc" />
<None Include="..\..\ext\libpng17\CMakeLists.txt" />
</ItemGroup>
<ItemGroup>
<None Include="..\..\Common\Math\fast\fast_matrix_neon.S" />
<None Include="..\..\Common\Math\lin\matrix_neon.s" />
</ItemGroup>
<ItemGroup>
<ProjectReference Include="..\cpu_features_UWP\cpu_features_UWP.vcxproj">
<Project>{c249f016-7f82-45cf-bb6e-0642a988c4d3}</Project>
@@ -477,4 +472,4 @@
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
<ImportGroup Label="ExtensionTargets">
</ImportGroup>
</Project>
</Project>
UWP/CommonUWP/CommonUWP.vcxproj.filters
+1 -13
View File
@@ -250,9 +250,6 @@
<ClCompile Include="..\..\Common\Math\math_util.cpp">
<Filter>Math</Filter>
</ClCompile>
<ClCompile Include="..\..\Common\Math\fast\fast_matrix.c">
<Filter>Math\fast</Filter>
</ClCompile>
<ClCompile Include="..\..\Common\Math\lin\matrix4x4.cpp">
<Filter>Math\lin</Filter>
</ClCompile>
@@ -702,9 +699,6 @@
<ClInclude Include="..\..\Common\Math\math_util.h">
<Filter>Math</Filter>
</ClInclude>
<ClInclude Include="..\..\Common\Math\fast\fast_matrix.h">
<Filter>Math\fast</Filter>
</ClInclude>
<ClInclude Include="..\..\Common\Math\lin\matrix4x4.h">
<Filter>Math\lin</Filter>
</ClInclude>
@@ -1049,12 +1043,6 @@
</ClInclude>
</ItemGroup>
<ItemGroup>
<None Include="..\..\Common\Math\fast\fast_matrix_neon.S">
<Filter>Math\fast</Filter>
</None>
<None Include="..\..\Common\Math\lin\matrix_neon.s">
<Filter>Math\lin</Filter>
</None>
<None Include="..\..\ext\libpng17\CMakeLists.txt" />
<None Include="..\..\ext\basis_universal\basisu_transcoder_tables_astc.inc">
<Filter>ext\basis_universal</Filter>
@@ -1095,4 +1083,4 @@
<Filter>ext\at3_standalone</Filter>
</Text>
</ItemGroup>
</Project>
</Project>
android/jni/Android.mk
-1
View File
@@ -330,7 +330,6 @@ EXEC_AND_LIB_FILES := \
$(SRC)/Common/Render/Text/draw_text_android.cpp \
$(SRC)/Common/Input/GestureDetector.cpp \
$(SRC)/Common/Input/InputState.cpp \
$(SRC)/Common/Math/fast/fast_matrix.c \
$(SRC)/Common/Math/math_util.cpp \
$(SRC)/Common/Math/Statistics.cpp \
$(SRC)/Common/Math/curves.cpp \
libretro/Makefile.common
+2 -3
View File
@@ -561,9 +561,8 @@ SOURCES_CXX += \
$(COMMONDIR)/VR/VRInput.cpp \
$(COMMONDIR)/VR/VRRenderer.cpp
SOURCES_C +=\
$(COMMONDIR)/GPU/OpenGL/gl3stub.c \
$(COMMONDIR)/Math/fast/fast_matrix.c
SOURCES_C += \
$(COMMONDIR)/GPU/OpenGL/gl3stub.c
SOURCES_CXX += \
$(GPUCOMMONDIR)/Draw2D.cpp \