nishi/milsko
1
Fork 0
mirror of https://gitea.nishi.boats/pyrite-dev/milsko synced 2026-01-22 09:04:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

closes #10

Browse Source 
git-svn-id: http://svn2.nishi.boats/svn/milsko/trunk@570 b9cfdab3-6d41-4d17-bbe4-086880011989
This commit is contained in:
NishiOwO
2025-11-02 19:15:38 +00:00
parent 92db649bda
commit 971ef1b827
15 changed files with 767 additions and 796 deletions
Download Patch File Download Diff File Expand all files Collapse all files

136
src/math/mmx.c
View File

@@ -1,111 +1,129 @@
/* $Id$ */
#include <Mw/LowLevelMath.h>
#ifdef MwLLMathMMX
#ifdef MwLLMath_x86
#include "math_internal.h"
#include <mmintrin.h>
#define DO_MMX_INTRINSIC(intrin, _ty, _rty, _tyn) \
__m64 m = intrin(*(__m64*)&a->un._ty, *(__m64*)&b->un._ty); \
struct _tyn* t = (struct _tyn*)&m; \
out->un._rty = *t;
#define DO_MMX_INTRINSIC(intrin, _ty, _cty) \
int i = 0; \
for(; i < a->size; i += 8) { \
__m64 m = intrin(*(__m64*)&a->buffer[i], *(__m64*)&b->buffer[i]); \
memcpy(&((_cty*)out->buffer)[i], &m, sizeof(m)); \
}
static void mmx_add8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_paddb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_paddb, u8, MwU8)
}
static void mmx_sub8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psubb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_psubb, u8, MwU8);
}
static void mmx_equal8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpeqb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_pcmpeqb, u8, MwU8);
}
static void mmx_add16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_paddw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_paddw, u16, MwU16);
}
static void mmx_sub16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psubw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_psubw, u16, MwU16);
}
static void mmx_shiftRight16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psrlw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_psrlw, u16, MwU16);
}
static void mmx_shiftLeft16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psllw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_psllw, u16, MwU16);
}
static void mmx_equal16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpeqw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_pcmpeqw, u16, MwU16);
}
static void mmx_add32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_paddd, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_paddd, u32, MwU32);
}
static void mmx_sub32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psubd, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_psubd, u32, MwU32);
}
static void mmx_shiftRight32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psrld, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_psrld, u32, MwU32);
}
static void mmx_shiftLeft32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pslld, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_pslld, u32, MwU32);
}
static void mmx_equal32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpeqw, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_pcmpeqw, u32, MwU32);
}
static void mmx_addi8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_paddsb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_paddsb, i8, MwI8);
}
static void mmx_subi8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psubsb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_psubsb, i8, MwI8);
}
static void mmx_addi16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_paddsw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_paddsw, i16, MwI16);
}
static void mmx_subi16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_psubsw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_psubsw, i16, MwI16);
}
static void mmx_greaterTheni8(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpgtb, u8, u8, _MwLLVecDataU8x8);
DO_MMX_INTRINSIC(_m_pcmpgtb, i8, MwI8);
}
static void mmx_greaterTheni16(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpgtw, u16, u16, _MwLLVecDataU16x4);
DO_MMX_INTRINSIC(_m_pcmpgtw, i16, MwI16);
}
static void mmx_greaterTheni32(MwLLVec* a, MwLLVec* b, MwLLVec* out) {
DO_MMX_INTRINSIC(_m_pcmpgtd, u32, u32, _MwLLVecDataU32x2);
DO_MMX_INTRINSIC(_m_pcmpgtd, i32, MwI16);
}
void mmx_apply(MwLLMathVTable** t) {
t[_MwLLVecTypeU8x8]->Add = mmx_add8;
t[_MwLLVecTypeU8x8]->Sub = mmx_sub8;
// t[_MwLLVecTypeU8x8]->GreaterThen = mmx_greaterThen8;
t[_MwLLVecTypeU8x8]->Equal = mmx_equal8;
t[_MwLLVecTypeU16x4]->Add = mmx_add16;
t[_MwLLVecTypeU16x4]->Sub = mmx_sub16;
t[_MwLLVecTypeU16x4]->ShiftLeft = mmx_shiftLeft16;
t[_MwLLVecTypeU16x4]->ShiftRight = mmx_shiftRight16;
// t[_MwLLVecTypeU16x4]->GreaterThen = mmx_greaterThen16;
t[_MwLLVecTypeU16x4]->Equal = mmx_equal16;
t[_MwLLVecTypeU32x2]->Add = mmx_add32;
t[_MwLLVecTypeU32x2]->Sub = mmx_sub32;
t[_MwLLVecTypeU32x2]->ShiftLeft = mmx_shiftLeft32;
t[_MwLLVecTypeU32x2]->ShiftRight = mmx_shiftRight32;
// t[_MwLLVecTypeU32x2]->GreaterThen = mmx_greaterThen32;
t[_MwLLVecTypeU32x2]->Equal = mmx_equal32;
t[_MwLLVecTypeU8x8]->Add = mmx_addi8;
t[_MwLLVecTypeU8x8]->Sub = mmx_subi8;
t[_MwLLVecTypeI8x8]->GreaterThen = mmx_greaterTheni8;
t[_MwLLVecTypeI8x8]->Equal = mmx_equal8;
t[_MwLLVecTypeU16x4]->Add = mmx_addi16;
t[_MwLLVecTypeU16x4]->Sub = mmx_subi16;
t[_MwLLVecTypeI16x4]->ShiftLeft = mmx_shiftLeft16;
t[_MwLLVecTypeI16x4]->ShiftRight = mmx_shiftRight16;
t[_MwLLVecTypeI16x4]->GreaterThen = mmx_greaterTheni16;
t[_MwLLVecTypeI16x4]->Equal = mmx_equal16;
t[_MwLLVecTypeI32x2]->Add = mmx_add32;
t[_MwLLVecTypeI32x2]->Sub = mmx_sub32;
t[_MwLLVecTypeI32x2]->ShiftLeft = mmx_shiftLeft32;
t[_MwLLVecTypeI32x2]->ShiftRight = mmx_shiftRight32;
t[_MwLLVecTypeI32x2]->GreaterThen = mmx_greaterTheni32;
t[_MwLLVecTypeI32x2]->Equal = mmx_equal32;
void mmx_apply(MwLLVec* vec) {
switch(vec->ty) {
case MwLLVecTypeU8:
vec->vtable.add = mmx_add8;
vec->vtable.sub = mmx_sub8;
vec->vtable.equal = mmx_equal8;
break;
case MwLLVecTypeU16:
vec->vtable.add = mmx_add16;
vec->vtable.sub = mmx_sub16;
vec->vtable.shiftLeft = mmx_shiftLeft16;
vec->vtable.shiftRight = mmx_shiftRight16;
vec->vtable.equal = mmx_equal16;
break;
case MwLLVecTypeU32:
vec->vtable.add = mmx_add32;
vec->vtable.sub = mmx_sub32;
vec->vtable.shiftLeft = mmx_shiftLeft32;
vec->vtable.shiftRight = mmx_shiftRight32;
vec->vtable.equal = mmx_equal32;
break;
case MwLLVecTypeU64:
break;
case MwLLVecTypeI8:
vec->vtable.add = mmx_addi8;
vec->vtable.sub = mmx_subi8;
vec->vtable.greaterThen = mmx_greaterTheni8;
vec->vtable.equal = mmx_equal8;
break;
case MwLLVecTypeI16:
vec->vtable.add = mmx_addi16;
vec->vtable.sub = mmx_subi16;
vec->vtable.shiftLeft = mmx_shiftLeft16;
vec->vtable.shiftRight = mmx_shiftRight16;
vec->vtable.greaterThen = mmx_greaterTheni16;
vec->vtable.equal = mmx_equal16;
break;
case MwLLVecTypeI32:
vec->vtable.add = mmx_add32;
vec->vtable.sub = mmx_sub32;
vec->vtable.shiftLeft = mmx_shiftLeft32;
vec->vtable.shiftRight = mmx_shiftRight32;
vec->vtable.greaterThen = mmx_greaterTheni32;
vec->vtable.equal = mmx_equal32;
break;
case MwLLVecTypeI64:
break;
case MwLLVecType_Max:
break;
}
}
#endif