balhau/cpplab
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Replaced vmovdqa with vmovdqu because the memory is not aligned to 256 bits. Next align the memory and then use the aligned assembly instruction

Browse source
This commit is contained in:
balhau@balhau.net 2021-12-25 19:14:49 +00:00
parent cbf7ecee92
commit 7345dd2665
No known key found for this signature in database
GPG key ID: 1E666F326A121830
5 changed files with 36 additions and 35 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
app/sse.cpp
View file

@ -3,7 +3,6 @@
#include "../src/cpu/x86/sse.hpp" #include "../src/cpu/x86/sse.hpp"
#include "../src/cpu/naive.hpp" #include "../src/cpu/naive.hpp"
#include "../src/platform/timer.hpp" #include "../src/platform/timer.hpp"
#include <sys/time.h>
using namespace cpu; using namespace cpu;
using namespace platform; using namespace platform;

2
src/cpu/types.hpp
View file

@ -38,7 +38,7 @@
typedef unsigned char UChar; typedef unsigned char UChar;
typedef unsigned short UShort; typedef unsigned short UShort;
typedef unsigned int UInt; typedef unsigned int UInt;
typedef unsigned long ULong; typedef unsigned long int ULong;
// SSE DataTypes // SSE DataTypes
#define CHAR_LEN_128 16 #define CHAR_LEN_128 16

2
src/cpu/utils.hpp
View file

@ -50,7 +50,7 @@ namespace cpu
packedULong[0] = (ULong)packedUInteger[1] << 32 | packedUInteger[0]; packedULong[0] = (ULong)packedUInteger[1] << 32 | packedUInteger[0];
packedULong[1] = (ULong)packedUInteger[3] << 32 | packedUInteger[2]; packedULong[1] = (ULong)packedUInteger[3] << 32 | packedUInteger[2];
packedULong[2] = (ULong)packedUInteger[5] << 32 | packedUInteger[4]; packedULong[2] = (ULong)packedUInteger[5] << 32 | packedUInteger[4];
packedULong[3] = (ULong)packedUInteger[7] << 32 | packedUInteger[7]; packedULong[3] = (ULong)packedUInteger[7] << 32 | packedUInteger[6];
} }
/** /**

50
src/cpu/x86/sse.cpp
View file

@ -1,9 +1,11 @@
#include "sse.hpp" #include "sse.hpp"
#ifdef ARCH_X86 #ifdef ARCH_X86
#include <iostream> #include <iostream>
void cpu::SSE::add_128(UChar *a,UChar *b){ void cpu::SSE::add_128(UChar *a,UChar *b){
__asm__ __volatile__ ( asm volatile (
"movdqa %0,%%xmm1\n" "movdqa %0,%%xmm1\n"
"paddb %1,%%xmm1\n" "paddb %1,%%xmm1\n"
"movdqa %%xmm1,%0" "movdqa %%xmm1,%0"
@ -14,9 +16,9 @@
//X86 Assembly to add two 128 bit numbers in the form of packed integers 32bit //X86 Assembly to add two 128 bit numbers in the form of packed integers 32bit
void cpu::SSE::add_128(UInt *a,UInt *b) { void cpu::SSE::add_128(UInt *a,UInt *b) {
__asm__ __volatile__ ( asm volatile (
"movdqa %0, %%xmm1\n" "movdqa %0,%%xmm1\n"
"paddw %1, %%xmm1\n" "paddw %1,%%xmm1\n"
"movdqa %%xmm1, %0" "movdqa %%xmm1, %0"
: "=m"(*a) : "=m"(*a)
: "m"(*b) : "m"(*b)
@ -26,9 +28,9 @@
//X86 Assembly to add two 128 bit numbers in the form of packed long 64bit //X86 Assembly to add two 128 bit numbers in the form of packed long 64bit
void cpu::SSE::add_128(ULong *a,ULong *b) { void cpu::SSE::add_128(ULong *a,ULong *b) {
__asm__ __volatile__ ( asm volatile (
"movdqa %0, %%xmm1\n" "movdqa %0,%%xmm1\n"
"paddd %1, %%xmm1\n" "paddd %1,%%xmm1\n"
"movdqa %%xmm1, %0" "movdqa %%xmm1, %0"
: "=m"(*a) : "=m"(*a)
: "m"(*b) : "m"(*b)
@ -37,35 +39,35 @@
//X86 Assembly to add two 256 bit numbers in the form of packed byte vector //X86 Assembly to add two 256 bit numbers in the form of packed byte vector
void cpu::SSE::add_256(UChar *a,UChar *b) { void cpu::SSE::add_256(UChar *a,UChar *b) {
__asm__ __volatile__ ( asm volatile (
"vmovdqa %0, %%ymm1\n" "vmovdqu %0,%%ymm1\n"
"vmovdqa %1, %%ymm2\n" "vmovdqu %1,%%ymm2\n"
"vpaddb %%ymm1, %%ymm1, %%ymm2\n" "vpaddb %%ymm3,%%ymm2,%%ymm1\n"
"vmovdqa %%ymm1, %0" "vmovdqu %%ymm1,%0"
: "=m"(*a) : "=m"(*a)
: "m"(*b) : "m"(*b)
); );
}; };
//X86 Assembly to add two 128 bit numbers in the form of packed long 32bit //X86 Assembly to add two 256 bit numbers in the form of packed int 32bit
void cpu::SSE::add_256(UInt *a,UInt *b) { void cpu::SSE::add_256(UInt *a,UInt *b) {
__asm__ __volatile__( asm volatile(
"vmovdqa %0, %%xmm1\n" "vmovdqu %0,%%ymm1\n"
"vmovdqa %1, %%ymm2\n" "vmovdqu %1,%%ymm2\n"
"vpaddw %%ymm1, %%ymm1, %%ymm2\n" "vpaddw %%ymm1, %%ymm2, %%ymm1\n"
"vmovdqa %%ymm1, %0" "vmovdqu %%ymm1,%0"
: "=m"(*a) : "=m"(*a)
: "m"(*b) : "m"(*b)
); );
}; };
//X86 Assembly to add two 128 bit numbers in the form of packed long 64bit //X86 Assembly to add two 256 bit numbers in the form of packed long 64bit
void cpu::SSE::add_256(ULong *a,ULong *b) { void cpu::SSE::add_256(ULong *a,ULong *b) {
__asm__ __volatile__( asm volatile(
"vmovdqa %0, %%ymm1\n" "vmovdqu %0, %%ymm1\n"
"vmovdqa %1, %%ymm2\n" "vmovdqu %1, %%ymm2\n"
"vpaddd %%ymm1, %%ymm1, %%ymm2\n" "vpaddd %%ymm1,%%ymm2,%%ymm1\n"
"vmovdqa %%ymm1, %0" "vmovdqu %%ymm1, %0"
: "=m"(*a) : "=m"(*a)
: "m"(*b) : "m"(*b)
); );

16
src/platform/timer.hpp
View file

@ -2,6 +2,7 @@
#include <iostream> #include <iostream>
#include <sys/time.h> #include <sys/time.h>
#include <unistd.h>
namespace platform namespace platform
{ {
@ -11,26 +12,25 @@ namespace platform
std::string label; std::string label;
long int start; long int start;
long int end; long int end;
long int gettime() long int gettime()
{ {
struct timeval tp; struct timeval tp;
gettimeofday(&tp, NULL); gettimeofday(&tp, NULL);
long int ms = tp.tv_sec * 1000 + tp.tv_usec / 1000; return (double)(tp.tv_sec * 1000 + (double)tp.tv_usec / 1000);
return ms;
}; };
public: public:
Timer(const char* lbl) : label(lbl) Timer(const char *lbl)
{ {
this->label=std::string(lbl); this->label = std::string(lbl);
this->start = gettime(); this->start = this->gettime();
} }
~Timer() ~Timer()
{ {
this->end=this->gettime(); this->end = this->gettime();
unsigned long diff = this->end - this->start; unsigned long diff = this->end - this->start;
std::cout << this->label << diff << std::endl; std::cout << this->label << diff << std::endl;
} }
}; };