claunia/flac
1
0
Fork 0
mirror of https://github.com/claunia/flac.git synced 2025-12-16 18:54:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix bug in SSE version of ...autocorrelation...

Browse Source
This commit is contained in:
Josh Coalson
2001-05-30 19:08:50 +00:00
parent 714216128b
commit 7fc68ffc81
1 changed files with 28 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
src/libFLAC/i386/lpc_asm.s
View File

@@ -57,10 +57,10 @@ cglobal FLAC__lpc_restore_signal_asm_i386_mmx
ALIGN 16 ALIGN 16
cident FLAC__lpc_compute_autocorrelation_asm_i386 cident FLAC__lpc_compute_autocorrelation_asm_i386
; esp + 32 == autoc[] ;[esp + 32] == autoc[]
; esp + 28 == lag ;[esp + 28] == lag
; esp + 24 == data_len ;[esp + 24] == data_len
; esp + 20 == data[] ;[esp + 20] == data[]
push ebp push ebp
push ebx push ebx
@@ -208,10 +208,10 @@ cident FLAC__lpc_compute_autocorrelation_asm_i386
ALIGN 16 ALIGN 16
cident FLAC__lpc_compute_autocorrelation_asm_i386_sse cident FLAC__lpc_compute_autocorrelation_asm_i386_sse
; esp + 16 == autoc[] ;[esp + 16] == autoc[]
; esp + 12 == lag ;[esp + 12] == lag
; esp + 8 == data_len ;[esp + 8] == data_len
; esp + 4 == data[] ;[esp + 4] == data[]
; for(coeff = 0; coeff < lag; coeff++) ; for(coeff = 0; coeff < lag; coeff++)
; autoc[coeff] = 0.0; ; autoc[coeff] = 0.0;
@@ -227,45 +227,38 @@ cident FLAC__lpc_compute_autocorrelation_asm_i386_sse
shufps xmm0, xmm0, 0 ; xmm0 = data[0],data[0],data[0],data[0] shufps xmm0, xmm0, 0 ; xmm0 = data[0],data[0],data[0],data[0]
movaps xmm1, xmm0 ; xmm1 = data[0],data[0],data[0],data[0] movaps xmm1, xmm0 ; xmm1 = data[0],data[0],data[0],data[0]
xorps xmm3, xmm3 ; xmm3 = 0,0,0,0 xorps xmm3, xmm3 ; xmm3 = 0,0,0,0
.warmup: ; xmm3:xmm2 = data[sample-[7..0]] .warmup: ; xmm3:xmm2 == data[sample-7],data[sample-6],...,data[sample]
movaps xmm4, xmm0 mulps xmm0, xmm2
movaps xmm5, xmm1 ; xmm5:xmm4 = xmm1:xmm0 = data[sample]*8 mulps xmm1, xmm3 ; xmm5:xmm4 = xmm1:xmm0 * xmm3:xmm2
mulps xmm4, xmm2 addps xmm6, xmm0
mulps xmm5, xmm3 ; xmm5:xmm4 = xmm1:xmm0 * xmm3:xmm2 addps xmm7, xmm1 ; xmm7:xmm6 += xmm1:xmm0 * xmm3:xmm2
addps xmm6, xmm4
addps xmm7, xmm5 ; xmm7:xmm6 += xmm1:xmm0 * xmm3:xmm2
dec edx dec edx
;* there's no need to even check for this because we know that lag == 8 ;* there's no need to even check for this because we know that lag == 8
;* and data_len >= lag, so our 1-sample warmup cannot finish the loop ;* and data_len >= lag, so our 1-sample warmup cannot finish the loop
; jz .loop_end ; jz .loop_end
ALIGN 16 ALIGN 16
.loop_8: .loop_8:
; read the next sample ; start by reading the next sample
movss xmm0, [eax] ; xmm0 = 0,0,0,data[sample] movss xmm0, [eax] ; xmm0 = 0,0,0,data[sample]
add eax, 4 ; here we reorder the instructions; see the (#) indexes for a logical order
shufps xmm0, xmm0, 0 ; xmm0 = data[sample],data[sample],data[sample],data[sample] shufps xmm2, xmm2, 93h ; (3) 93h=2-1-0-3 => xmm2 gets rotated left by one float
movaps xmm1, xmm0 ; xmm1 = data[sample],data[sample],data[sample],data[sample] add eax, 4 ; (0)
; now shift the lagged samples shufps xmm3, xmm3, 93h ; (4) 93h=2-1-0-3 => xmm3 gets rotated left by one float
movaps xmm4, xmm2 shufps xmm0, xmm0, 0 ; (1) xmm0 = data[sample],data[sample],data[sample],data[sample]
movaps xmm5, xmm3 movss xmm3, xmm2 ; (5)
shufps xmm2, xmm4, 93h ; 93h=2-1-0-3 => xmm2 gets rotated left by one float movaps xmm1, xmm0 ; (2) xmm1 = data[sample],data[sample],data[sample],data[sample]
shufps xmm3, xmm5, 93h ; 93h=2-1-0-3 => xmm3 gets rotated left by one float movss xmm2, xmm0 ; (6)
movss xmm3, xmm2 mulps xmm1, xmm3 ; (8)
movss xmm2, xmm0 mulps xmm0, xmm2 ; (7) xmm5:xmm4 = xmm1:xmm0 * xmm3:xmm2
addps xmm7, xmm1 ; (10)
movaps xmm4, xmm0 addps xmm6, xmm0 ; (9) xmm7:xmm6 += xmm1:xmm0 * xmm3:xmm2
movaps xmm5, xmm1 ; xmm5:xmm4 = xmm1:xmm0 = data[sample]*8
mulps xmm4, xmm2
mulps xmm5, xmm3 ; xmm5:xmm4 = xmm1:xmm0 * xmm3:xmm2
addps xmm6, xmm4
addps xmm7, xmm5 ; xmm7:xmm6 += xmm1:xmm0 * xmm3:xmm2
dec edx dec edx
jnz .loop_8 jnz .loop_8
.loop_end: .loop_end:
; store autoc ; store autoc
mov edx, [esp + 16] ; edx == autoc mov edx, [esp + 16] ; edx == autoc
movups xmm6, [edx] movups [edx], xmm6
movups xmm7, [edx + 4] movups [edx + 4], xmm7
.end: .end:
ret ret