/* Intel SIMD SSE2 implementation of Viterbi ACS butterflies for 256-state (k=9) convolutional code Copyright 2004 Phil Karn, KA9Q This code may be used under the terms of the GNU Lesser General Public License (LGPL) Modifications for x86_64, 2012 Matthias P. Braendli, HB9EGM - changed registers to x86-64 equivalents - changed instructions accordingly - %rip indirect addressing needed for position independent code, which is required because x86-64 needs dynamic libs to be PIC. That still doesn't work void update_viterbi29_blk_sse2(struct v29 *vp,unsigned char *syms,int nbits) ; */ # SSE2 (128-bit integer SIMD) version # All X86-64 CPUs include SSE2 # These are offsets into struct v29, defined in viterbi29_av.c .set DP,512 .set OLDMETRICS,516 .set NEWMETRICS,520 .text .global update_viterbi29_blk_sse2,Branchtab29_sse2 .type update_viterbi29_blk_sse2,@function .align 16 update_viterbi29_blk_sse2: pushq %rbp movq %rsp,%rbp /* convention different between i386 and x86_64: rsi and rdi belong to called function, not caller */ /* Let's say we don't care (yet) */ pushq %rsi pushq %rdi pushq %rdx pushq %rbx movq 8(%rbp),%rdx # edx = vp testq %rdx,%rdx jnz 0f movq -1,%rax jmp err 0: movq OLDMETRICS(%rdx),%rsi # esi -> old metrics movq NEWMETRICS(%rdx),%rdi # edi -> new metrics movq DP(%rdx),%rdx # edx -> decisions 1: movq 16(%rbp),%rax # eax = nbits decq %rax jl 2f # passed zero, we're done movq %rax,16(%rbp) xorq %rax,%rax movq 12(%rbp),%rbx # ebx = syms movb (%rbx),%al movd %rax,%xmm6 # xmm6[0] = first symbol movb 1(%rbx),%al movd %rax,%xmm5 # xmm5[0] = second symbol addq $2,%rbx movq %rbx,12(%rbp) punpcklbw %xmm6,%xmm6 # xmm6[1] = xmm6[0] punpcklbw %xmm5,%xmm5 movdqa thirtyones(%rip),%xmm7 pshuflw $0,%xmm6,%xmm6 # copy low word to low 3 pshuflw $0,%xmm5,%xmm5 punpcklqdq %xmm6,%xmm6 # propagate to all 16 punpcklqdq %xmm5,%xmm5 # xmm6 now contains first symbol in each byte, xmm5 the second movdqa thirtyones(%rip),%xmm7 # each invocation of this macro does 16 butterflies in parallel .MACRO butterfly GROUP # compute branch metrics movdqa Branchtab29_sse2+(16*\GROUP)(%rip),%xmm4 movdqa Branchtab29_sse2+128+(16*\GROUP)(%rip),%xmm3 pxor %xmm6,%xmm4 pxor %xmm5,%xmm3 pavgb %xmm3,%xmm4 psrlw $3,%xmm4 pand %xmm7,%xmm4 # xmm4 contains branch metrics movdqa (16*\GROUP)(%esi),%xmm0 # Incoming path metric, high bit = 0 movdqa ((16*\GROUP)+128)(%esi),%xmm3 # Incoming path metric, high bit = 1 movdqa %xmm0,%xmm2 movdqa %xmm3,%xmm1 paddusb %xmm4,%xmm0 paddusb %xmm4,%xmm3 # invert branch metrics pxor %xmm7,%xmm4 paddusb %xmm4,%xmm1 paddusb %xmm4,%xmm2 # Find survivors, leave in mm0,2 pminub %xmm1,%xmm0 pminub %xmm3,%xmm2 # get decisions, leave in mm1,3 pcmpeqb %xmm0,%xmm1 pcmpeqb %xmm2,%xmm3 # interleave and store new branch metrics in mm0,2 movdqa %xmm0,%xmm4 punpckhbw %xmm2,%xmm0 # interleave second 16 new metrics punpcklbw %xmm2,%xmm4 # interleave first 16 new metrics movdqa %xmm0,(32*\GROUP+16)(%rdi) movdqa %xmm4,(32*\GROUP)(%rdi) # interleave decisions & store movdqa %xmm1,%xmm4 punpckhbw %xmm3,%xmm1 punpcklbw %xmm3,%xmm4 # work around bug in gas due to Intel doc error .byte 0x66,0x0f,0xd7,0xd9 # pmovmskb %xmm1,%ebx shlq $16,%rbx .byte 0x66,0x0f,0xd7,0xc4 # pmovmskb %xmm4,%eax orq %rax,%rbx movq %rbx,(4*\GROUP)(%rdx) .endm # invoke macro 8 times for a total of 128 butterflies butterfly GROUP=0 butterfly GROUP=1 butterfly GROUP=2 butterfly GROUP=3 butterfly GROUP=4 butterfly GROUP=5 butterfly GROUP=6 butterfly GROUP=7 addq $32,%rdx # bump decision pointer # see if we have to normalize movq (%rdi),%rax # extract first output metric andq $255,%rax cmp $50,%rax # is it greater than 50? movq $0,%rax jle done # No, no need to normalize # Normalize by finding smallest metric and subtracting it # from all metrics movdqa (%rdi),%xmm0 pminub 16(%rdi),%xmm0 pminub 32(%rdi),%xmm0 pminub 48(%rdi),%xmm0 pminub 64(%rdi),%xmm0 pminub 80(%rdi),%xmm0 pminub 96(%rdi),%xmm0 pminub 112(%rdi),%xmm0 pminub 128(%rdi),%xmm0 pminub 144(%rdi),%xmm0 pminub 160(%rdi),%xmm0 pminub 176(%rdi),%xmm0 pminub 192(%rdi),%xmm0 pminub 208(%rdi),%xmm0 pminub 224(%rdi),%xmm0 pminub 240(%rdi),%xmm0 # crunch down to single lowest metric movdqa %xmm0,%xmm1 psrldq $8,%xmm0 # the count to psrldq is bytes, not bits! pminub %xmm1,%xmm0 movdqa %xmm0,%xmm1 psrlq $32,%xmm0 pminub %xmm1,%xmm0 movdqa %xmm0,%xmm1 psrlq $16,%xmm0 pminub %xmm1,%xmm0 movdqa %xmm0,%xmm1 psrlq $8,%xmm0 pminub %xmm1,%xmm0 punpcklbw %xmm0,%xmm0 # lowest 2 bytes pshuflw $0,%xmm0,%xmm0 # lowest 8 bytes punpcklqdq %xmm0,%xmm0 # all 16 bytes # xmm0 now contains lowest metric in all 16 bytes # subtract it from every output metric movdqa (%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,(%rdi) movdqa 16(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,16(%rdi) movdqa 32(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,32(%rdi) movdqa 48(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,48(%rdi) movdqa 64(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,64(%rdi) movdqa 80(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,80(%rdi) movdqa 96(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,96(%rdi) movdqa 112(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,112(%rdi) movdqa 128(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,128(%rdi) movdqa 144(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,144(%rdi) movdqa 160(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,160(%rdi) movdqa 176(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,176(%rdi) movdqa 192(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,192(%rdi) movdqa 208(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,208(%rdi) movdqa 224(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,224(%rdi) movdqa 240(%rdi),%xmm1 psubusb %xmm0,%xmm1 movdqa %xmm1,240(%rdi) done: # swap metrics movq %rsi,%rax movq %rdi,%rsi movq %rax,%rdi jmp 1b 2: movq 8(%rbp),%rbx # ebx = vp # stash metric pointers movq %rsi,OLDMETRICS(%rbx) movq %rdi,NEWMETRICS(%rbx) movq %rdx,DP(%rbx) # stash incremented value of vp->dp xorq %rax,%rax err: popq %rbx popq %rdx popq %rdi popq %rsi popq %rbp ret .data .align 16 thirtyones: .byte 31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31