main.cpp
#include "../commonfiles/mmxval.h" // The order of the name constants in the following enum must match // the table that is defined in mmxAddition.asm. enum MmxAddOp : unsigned int { paddb, // packed byte addition with wraparound paddsb, // packed byte addition with signed saturation paddusb, // packed byte addition with unsigned saturation paddw, // packed word addition with wraparound paddsw, // packed word addition with signed saturation paddusw, // packed word addition with unsigned saturation paddd // packed doubleword addition with wrapround }; extern "C" bool MmxAdd(MmxVal a, MmxVal b, MmxAddOp op, MmxVal *c ); void MmxAddBytes(void) { MmxVal a, b, c; char buff [256]; // Packed byte addition - signed integers a.i8[0] = 50; b.i8[0] = 30; a.i8[1] = 80; b.i8[1] = 64; a.i8[2] = -27; b.i8[2] = -32; a.i8[3] = -70; b.i8[3] = -80; a.i8[4] = -42; b.i8[4] = 90; a.i8[5] = 60; b.i8[5] = -85; a.i8[6] = 64; b.i8[6] = 90; a.i8[7] = 100; b.i8[7] = -30; printf("\n\nPacked byte addition - signed integers\n"); printf("a: %s\n", a.ToString_i8(buff, sizeof(buff))); printf("b: %s\n", b.ToString_i8(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddb, &c); printf("\npaddb results\n"); printf("c: %s\n", c.ToString_i8(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddsb, &c); printf("\npaddsb results\n"); printf("c: %s\n", c.ToString_i8(buff, sizeof(buff))); // Packed byte addition - unsigned integers a.u8[0] = 50; b.u8[0] = 30; a.u8[1] = 80; b.u8[1] = 64; a.u8[2] = 132; b.u8[2] = 130; a.u8[3] = 200; b.u8[3] = 180; a.u8[4] = 42; b.u8[4] = 90; a.u8[5] = 60; b.u8[5] = 85; a.u8[6] = 140; b.u8[6] = 160; a.u8[7] = 10; b.u8[7] = 14; printf("\n\nPacked byte addition - unsigned integers\n"); printf("a: %s\n", a.ToString_u8(buff, sizeof(buff))); printf("b: %s\n", b.ToString_u8(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddb, &c); printf("\npaddb results\n"); printf("c: %s\n", c.ToString_u8(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddusb, &c); printf("\npaddusb results\n"); printf("c: %s\n", c.ToString_u8(buff, sizeof(buff))); } void MmxAddWords(void) { MmxVal a, b, c; char buff [256]; // Packed word addition - signed integers a.i16[0] = 550; b.i16[0] = 830; a.i16[1] = 30000; b.i16[1] =5000; a.i16[2] = -270; b.i16[2] = -320; a.i16[3] = -7000; b.i16[3] = -32000; printf("\n\nPacked word addition - signed integers\n"); printf("a: %s\n", a.ToString_i16(buff, sizeof(buff))); printf("b: %s\n", b.ToString_i16(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddw, &c); printf("\npaddw results\n"); printf("c: %s\n", c.ToString_i16(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddsw, &c); printf("\npaddsw results\n"); printf("c: %s\n", c.ToString_i16(buff, sizeof(buff))); // Packed word addition - unsigned integers a.u16[0] = 50; b.u16[0] = 30; a.u16[1] = 48000; b.u16[1] = 20000; a.u16[2] = 132; b.u16[2] = 130; a.u16[3] = 10000; b.u16[3] = 60000; printf("\n\nPacked word addition - unsigned integers\n"); printf("a: %s\n", a.ToString_u16(buff, sizeof(buff))); printf("b: %s\n", b.ToString_u16(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddw, &c); printf("\npaddw results\n"); printf("c: %s\n", c.ToString_u16(buff, sizeof(buff))); MmxAdd(a, b, MmxAddOp::paddusw, &c); printf("\npaddusw results\n"); printf("c: %s\n", c.ToString_u16(buff, sizeof(buff))); } int main(int argc, char* argv[]) { MmxAddBytes(); MmxAddWords(); return 0; }
mmxaddition.asm
; extern "C" bool MmxAdd(MmxVal a, MmxVal b, MmxAddOp op, MmxVal *c ); ; ; Description: The following function demonstrates use of the ; padd* instructions. ; ; Returns: c contains the calculated result. ; eax returns true if success otherwise false ; ; Name: mmxaddition.asm ; ; Build: g++ -c -m32 main.cpp -o main.o -std=c++11 ; nasm -f elf32 -o mmxaddition.o mmxaddition.asm ; g++ -m32 -o mmxaddition mmxaddition.o main.o ../commonfiles/mmxval.o ; ; Source: Modern X86 Assembly Language Programming p.149 global MmxAdd section .data align 4 ; The order of the labels in the following table must match ; the enum that is defined in MmxAddition.cpp. %define DWORD_SIZE 4 AddOpTable: dd MmxAdd.mmxPaddb dd MmxAdd.mmxPaddsb dd MmxAdd.mmxPaddusb dd MmxAdd.mmxPaddw dd MmxAdd.mmxPaddsw dd MmxAdd.mmxPaddusw dd MmxAdd.mmxPaddd AddOpTableCount equ ($ - AddOpTable) / DWORD_SIZE section .text MmxAdd: %define a [ebp+8] %define b [ebp+16] %define add_op [ebp+24] %define c [ebp+28] push ebp mov ebp,esp ; Make sure 'add_op' is valid mov eax,add_op ;load 'add_op' cmp eax,AddOpTableCount ;compare to table count jae .badAddOp ;jump if 'add_op' is invalid ; Load parameters and execute specified instruction movq mm0,a ;load 'a' movq mm1,b ;load 'b' jmp [AddOpTable+eax*DWORD_SIZE] ;jump to specified 'add_op' .mmxPaddb: paddb mm0,mm1 ;packed byte addition using jmp .saveResult ;wraparound .mmxPaddsb: paddsb mm0,mm1 ;packed byte addition using jmp .saveResult ;signed saturation .mmxPaddusb: paddusb mm0,mm1 ;packed byte addition using jmp .saveResult ;unsigned saturation .mmxPaddw: paddw mm0,mm1 ;packed word addition using jmp .saveResult ;wraparound .mmxPaddsw: paddsw mm0,mm1 ;packed word addition using jmp .saveResult ;signed saturation .mmxPaddusw: paddusw mm0,mm1 ;packed word addition using jmp .saveResult ;unsigned saturation .mmxPaddd: paddd mm0,mm1 ;packed dword addition using jmp .saveResult ;wraparound .badAddOp: pxor mm0,mm0 ;return 0 if 'add_op' is bad pxor mm2,mm2 xor eax,eax ;return value false jmp .done ; Move final result into edx:eax .saveResult: ; g++ doesn't return 64 bit values in edx:eax therefor the patch to store the value in c mov eax, 1 ;return value true pshufw mm2,mm0,01001110b ;swap high & low dwords .done: ; save values of mmo and mm2 in c mov edx,c movq [edx],mm0 movq [edx+8],mm2 emms ;clear MMX state pop ebp ret
build
g++ -c -m32 main.cpp -o main.o -std=c++11 nasm -f elf32 -o mmxaddition.o mmxaddition.asm g++ -m32 -o mmxaddition mmxaddition.o main.o ../commonfiles/mmxval.o