This program is still to be converted. Because the Windows source uses ThreadTimer::SetThreadAffinityMask(), this will take some time.
ssepackedintegerhistogram.asm
; Name: ssepackedintegerhistogram.asm ; ; Build: g++ -c -m32 main.cpp -o main.o ; nasm -f elf32 -o ssepackedintegerhistogram.o ssepackedintegerhistogram.asm ; g++ -m32 -o ssepackedintegerhistogram ssepackedintegerhistogram.o main.o ../../commonfiles/xmmval.o ; ; Source: Modern x86 Assembly Language Programming p. 279 ; ; Remark: Until ImageBuffer.cpp is converted, this example is of little use. extern NUM_PIXELS_MAX global SsePiHistogram section .text ; extern bool SsePiHistogram(Uint32* histo, const Uint8* pixel_buff, Uint32 num_pixels); ; ; Description: The following function builds an image histogram. ; ; Returns: 0 = invalid argument value ; 1 = success ; ; Requires: SSE4.1 SsePiHistogram: push ebp mov ebp,esp and esp,0FFFFFFF0H ;align ESP to 16 byte boundary sub esp,1024 ;allocate histo2 mov edx,esp ;edx = histo2 push ebx push esi push edi ; Make sure num_pixels is valid xor eax,eax ;set error return code mov ecx,[ebp+16] ;ecx = num_pixels cmp ecx,[NUM_PIXELS_MAX] ja .done ;jump if num_pixels too big test ecx,1fh jnz .done ;jump if num_pixels % 32 != 0 ; Make sure histo & pixel_buff are properly aligned mov ebx,[ebp+8] ;ebx = histo test ebx,0fh jnz .done ;jump if misaligned mov esi,[ebp+12] ;esi = pixel_buff test esi,0fh jnz .done ;jump if misaligned ; Initialize the histogram buffers (set all entries to zero) mov edi,ebx ;edi = histo mov ecx,256 rep stosd ;initialize histo mov edi,edx ;edi = histo2 mov ecx,256 rep stosd ;initialize histo2 ; Perform processing loop initializations mov edi,edx ;edi = histo2 mov ecx,[ebp+16] ;ecx = number of pxiels shr ecx,5 ;ecx = number of pixel blocks ; Build the histograms ; Register usage: ebx = histo, edi = histo2, esi = pixel_buff align 16 ;align jump target .@1: movdqa xmm0,[esi] ;load pixel block movdqa xmm2,[esi+16] ;load pixel block movdqa xmm1,xmm0 movdqa xmm3,xmm2 ; Process pixels 0 - 3 pextrb eax,xmm0,0 ;extract & count pixel 0 add dword[ebx+eax*4],1 pextrb edx,xmm1,1 ;extract & count pixel 1 add dword[edi+edx*4],1 pextrb eax,xmm0,2 ;extract & count pixel 2 add dword[ebx+eax*4],1 pextrb edx,xmm1,3 ;extract & count pixel 3 add dword[edi+edx*4],1 ; Process pixels 4 - 7 pextrb eax,xmm0,4 ;extract & count pixel 4 add dword[ebx+eax*4],1 pextrb edx,xmm1,5 ;extract & count pixel 5 add dword[edi+edx*4],1 pextrb eax,xmm0,6 ;extract & count pixel 6 add dword[ebx+eax*4],1 pextrb edx,xmm1,7 ;extract & count pixel 7 add dword[edi+edx*4],1 ; Process pixels 8 - 11 pextrb eax,xmm0,8 ;extract & count pixel 8 add dword[ebx+eax*4],1 pextrb edx,xmm1,9 ;extract & count pixel 9 add dword[edi+edx*4],1 pextrb eax,xmm0,10 ;extract & count pixel 10 add dword[ebx+eax*4],1 pextrb edx,xmm1,11 ;extract & count pixel 11 add dword[edi+edx*4],1 ; Process pixels 12 - 15 pextrb eax,xmm0,12 ;extract & count pixel 12 add dword[ebx+eax*4],1 pextrb edx,xmm1,13 ;extract & count pixel 13 add dword[edi+edx*4],1 pextrb eax,xmm0,14 ;extract & count pixel 14 add dword[ebx+eax*4],1 pextrb edx,xmm1,15 ;extract & count pixel 15 add dword[edi+edx*4],1 ; Process pixels 16 - 19 pextrb eax,xmm2,0 ;extract & count pixel 16 add dword[ebx+eax*4],1 pextrb edx,xmm3,1 ;extract & count pixel 17 add dword[edi+edx*4],1 pextrb eax,xmm2,2 ;extract & count pixel 18 add dword[ebx+eax*4],1 pextrb edx,xmm3,3 ;extract & count pixel 19 add dword[edi+edx*4],1 ; Process pixels 20 - 23 pextrb eax,xmm2,4 ;extract & count pixel 20 add dword[ebx+eax*4],1 pextrb edx,xmm3,5 ;extract & count pixel 21 add dword[edi+edx*4],1 pextrb eax,xmm2,6 ;extract & count pixel 22 add dword[ebx+eax*4],1 pextrb edx,xmm3,7 ;extract & count pixel 23 add dword[edi+edx*4],1 ; Process pixels 24 - 27 pextrb eax,xmm2,8 ;extract & count pixel 24 add dword[ebx+eax*4],1 pextrb edx,xmm3,9 ;extract & count pixel 25 add dword[edi+edx*4],1 pextrb eax,xmm2,10 ;extract & count pixel 26 add dword[ebx+eax*4],1 pextrb edx,xmm3,11 ;extract & count pixel 27 add dword[edi+edx*4],1 ; Process pixels 28 - 31 pextrb eax,xmm2,12 ;extract & count pixel 28 add dword[ebx+eax*4],1 pextrb edx,xmm3,13 ;extract & count pixel 29 add dword[edi+edx*4],1 pextrb eax,xmm2,14 ;extract & count pixel 30 add dword[ebx+eax*4],1 pextrb edx,xmm3,15 ;extract & count pixel 31 add dword[edi+edx*4],1 add esi,32 ;esi =next pixel block sub ecx,1 ;update counter jnz .@1 ;repeat loop if not done ; Add histo2 to histo for final histogram. Note that each loop iteration ; adds 8 histogram entries. mov ecx,32 ;ecx = number of iterations xor eax,eax ;eax = offset for histo arrays .@2: movdqa xmm0,[ebx+eax] ;load histo counts movdqa xmm1,[ebx+eax+16] paddd xmm0,[edi+eax] ;add counts from histo2 paddd xmm1,[edi+eax+16] movdqa [ebx+eax],xmm0 ;save final histo counts movdqa [ebx+eax+16],xmm1 add eax,32 ;update array offset sub ecx,1 ;update counter jnz .@2 ;repeat loop if not done mov eax,1 ;set success return code .done: pop edi pop esi pop ebx mov esp,ebp pop ebp ret
build
nasm -f elf32 -o ssepackedintegerhistogram.o ssepackedintegerhistogram.asm