[FFmpeg-devel] suggestions for debugging ff_fft_calc_altivec?
Pavel Koshevoy
pkoshevoy at gmail.com
Sun Aug 11 07:10:30 CEST 2013
On 8/10/13 10:27 PM, Pavel Koshevoy wrote:
> Hi,
>
> I've decided to take a closer look at OSX 10.5 PPC build with
> gas-preprocessor.pl. The last time I tried it more than half a year
> ago the result was that some audio decoders didn't sound right (AAC,
> AC3 -- both sound like loud white noise).
>
> I tried again today and compared decoder output from builds with and
> without gas-preprocessor. I've narrowed it down to output of
> fft_calc_c (called in ff_imdct_half_c) being different from output of
> ff_fft_calc_altivec (called in imdct_half_altivec).
>
> I am not a PPC assembly expert, nor am I familiar with FFT
> implementation details to be able to easily spot what the problem is
> with ff_fft_calc_altivec. Can anyone offer some suggestions to
> proceed further?
>
I've attached the gas-preprocess'ed fft_altivec_s temp assembly file,
for reference.
Pavel.
-------------- next part --------------
# 1 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S"
# 1 "/Developer/ppc-debug-gas/ffmpeg-git-build-nfs-gas-debug//"
# 1 "<built-in>"
# 1 "<command line>"
# 1 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S"
# 41 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S"
# 1 "./config.h" 1
# 42 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S" 2
# 1 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/asm.S" 1
# 64 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/asm.S"
.macro extfunc name
.global _\name
.type _\name, STT_FUNC
_\name:
\name:
.endm
.macro movrel rd, sym, gp
lis \rd, \sym at ha
la \rd, \sym at l(\rd)
.endm
.macro get_got rd
.endm
# 43 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S" 2
.text
.macro addi2 ra, imm
.if \imm & 0xffff
addi \ra, \ra, \imm at l
.endif
.if (\imm+0x8000)>>16
addis \ra, \ra, \imm at ha
.endif
.endm
.macro FFT4 a0, a1, a2, a3
vperm \a2,\a0,\a1,v20
vperm \a3,\a0,\a1,v21
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vmrghw \a2,\a0,\a1
vperm \a3,\a0,\a1,v22
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vperm \a2,\a0,\a1,v23
vperm \a3,\a0,\a1,v24
.endm
.macro FFT4x2 a0, a1, b0, b1, a2, a3, b2, b3
vperm \a2,\a0,\a1,v20
vperm \a3,\a0,\a1,v21
vperm \b2,\b0,\b1,v20
vperm \b3,\b0,\b1,v21
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vaddfp \b0,\b2,\b3
vsubfp \b1,\b2,\b3
vmrghw \a2,\a0,\a1
vperm \a3,\a0,\a1,v22
vmrghw \b2,\b0,\b1
vperm \b3,\b0,\b1,v22
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vaddfp \b0,\b2,\b3
vsubfp \b1,\b2,\b3
vperm \a2,\a0,\a1,v23
vperm \a3,\a0,\a1,v24
vperm \b2,\b0,\b1,v23
vperm \b3,\b0,\b1,v24
.endm
.macro FFT8 a0, a1, b0, b1, a2, a3, b2, b3, b4
vmrghw \b2,\b0,\b1
vmrglw \b3,\b0,\b1
vperm \a2,\a0,\a1,v20
vperm \a3,\a0,\a1,v21
vaddfp \b0,\b2,\b3
vsubfp \b1,\b2,\b3
vperm \b4,\b1,\b1,v25
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vmaddfp \b1,\b1,v17,v14
vmaddfp \b1,\b4,v18,\b1
vmrghw \a2,\a0,\a1
vperm \a3,\a0,\a1,v22
vperm \b2,\b0,\b1,v26
vperm \b3,\b0,\b1,v27
vaddfp \a0,\a2,\a3
vsubfp \a1,\a2,\a3
vaddfp \b0,\b2,\b3
vsubfp \b1,\b2,\b3
vperm \a2,\a0,\a1,v23
vperm \a3,\a0,\a1,v24
vperm \b2,\b0,\b1,v28
vperm \b3,\b0,\b1,v29
vsubfp \b0,\a2,\b2
vsubfp \b1,\a3,\b3
vaddfp \a0,\a2,\b2
vaddfp \a1,\a3,\b3
.endm
.macro BF d0,d1,s0,s1
vsubfp \d1,\s0,\s1
vaddfp \d0,\s0,\s1
.endm
.macro zip d0,d1,s0,s1
vmrghw \d0,\s0,\s1
vmrglw \d1,\s0,\s1
.endm
.macro def_fft4 interleave
fft4\interleave\()_altivec:
lvx v0, 0,r3
lvx v1,r9,r3
FFT4 v0,v1,v2,v3
.ifnb \interleave
zip v0,v1,v2,v3
stvx v0, 0,r3
stvx v1,r9,r3
.else
stvx v2, 0,r3
stvx v3,r9,r3
.endif
blr
.endm
.macro def_fft8 interleave
fft8\interleave\()_altivec:
addi r4,r3,32
lvx v0, 0,r3
lvx v1,r9,r3
lvx v2, 0,r4
lvx v3,r9,r4
FFT8 v0,v1,v2,v3,v4,v5,v6,v7,v8
.ifnb \interleave
zip v4,v5,v0,v1
zip v6,v7,v2,v3
stvx v4, 0,r3
stvx v5,r9,r3
stvx v6, 0,r4
stvx v7,r9,r4
.else
stvx v0, 0,r3
stvx v1,r9,r3
stvx v2, 0,r4
stvx v3,r9,r4
.endif
blr
.endm
.macro def_fft16 interleave
fft16\interleave\()_altivec:
addi r5,r3,64
addi r6,r3,96
addi r4,r3,32
lvx v0, 0,r5
lvx v1,r9,r5
lvx v2, 0,r6
lvx v3,r9,r6
FFT4x2 v0,v1,v2,v3,v4,v5,v6,v7
lvx v0, 0,r3
lvx v1,r9,r3
lvx v2, 0,r4
lvx v3,r9,r4
FFT8 v0,v1,v2,v3,v8,v9,v10,v11,v12
vmaddfp v8,v4,v15,v14
vmaddfp v9,v5,v15,v14
vmaddfp v10,v6,v15,v14
vmaddfp v11,v7,v15,v14
vmaddfp v8,v5,v16,v8
vnmsubfp v9,v4,v16,v9
vnmsubfp v10,v7,v16,v10
vmaddfp v11,v6,v16,v11
BF v10,v12,v10,v8
BF v11,v13,v9,v11
BF v0,v4,v0,v10
BF v3,v7,v3,v12
BF v1,v5,v1,v11
BF v2,v6,v2,v13
.ifnb \interleave
zip v8, v9,v0,v1
zip v10,v11,v2,v3
zip v12,v13,v4,v5
zip v14,v15,v6,v7
stvx v8, 0,r3
stvx v9,r9,r3
stvx v10, 0,r4
stvx v11,r9,r4
stvx v12, 0,r5
stvx v13,r9,r5
stvx v14, 0,r6
stvx v15,r9,r6
.else
stvx v0, 0,r3
stvx v4, 0,r5
stvx v3,r9,r4
stvx v7,r9,r6
stvx v1,r9,r3
stvx v5,r9,r5
stvx v2, 0,r4
stvx v6, 0,r6
.endif
blr
.endm
.macro PASS interleave, suffix
fft_pass\suffix\()_altivec:
mtctr r5
slwi r0,r5,4
slwi r7,r5,6
slwi r5,r5,5
add r10,r5,r7
add r0,r4,r0
addi r6,r5,16
addi r8,r7,16
addi r11,r10,16
1:
lvx v8, 0,r4
lvx v10, 0,r0
sub r0,r0,r9
lvx v9, 0,r0
vperm v9,v9,v10,v19
lvx v4,r3,r7
lvx v5,r3,r8
lvx v6,r3,r10
lvx v7,r3,r11
vmaddfp v10,v4,v8,v14
vmaddfp v11,v5,v8,v14
vmaddfp v12,v6,v8,v14
vmaddfp v13,v7,v8,v14
lvx v0, 0,r3
lvx v3,r3,r6
vmaddfp v10,v5,v9,v10
vnmsubfp v11,v4,v9,v11
vnmsubfp v12,v7,v9,v12
vmaddfp v13,v6,v9,v13
lvx v1,r3,r9
lvx v2,r3,r5
BF v12,v8,v12,v10
BF v13,v9,v11,v13
BF v0,v4,v0,v12
BF v3,v7,v3,v8
.if !\interleave
stvx v0, 0,r3
stvx v4,r3,r7
stvx v3,r3,r6
stvx v7,r3,r11
.endif
BF v1,v5,v1,v13
BF v2,v6,v2,v9
.if !\interleave
stvx v1,r3,r9
stvx v2,r3,r5
stvx v5,r3,r8
stvx v6,r3,r10
.else
vmrghw v8,v0,v1
vmrglw v9,v0,v1
stvx v8, 0,r3
stvx v9,r3,r9
vmrghw v8,v2,v3
vmrglw v9,v2,v3
stvx v8,r3,r5
stvx v9,r3,r6
vmrghw v8,v4,v5
vmrglw v9,v4,v5
stvx v8,r3,r7
stvx v9,r3,r8
vmrghw v8,v6,v7
vmrglw v9,v6,v7
stvx v8,r3,r10
stvx v9,r3,r11
.endif
addi r3,r3,32
addi r4,r4,16
bdnz 1b
sub r3,r3,r5
blr
.endm
# 315 "/nfs/scratch/Developer/ffmpeg-git-src/libavcodec/ppc/fft_altivec_s.S"
.rodata
.align 4
fft_data:
.float 0, 0, 0, 0
.float 1, 0.92387953, 0.70710678118654752440, 0.38268343
.float 0, 0.38268343, 0.70710678118654752440, 0.92387953
.float -0.70710678118654752440, 0.70710678118654752440, 0.70710678118654752440,-0.70710678118654752440
.float 0.70710678118654752440, 0.70710678118654752440, 0.70710678118654752440, 0.70710678118654752440
.byte 0x10,0x11,0x12,0x13, 0x0c,0x0d,0x0e,0x0f, 0x08,0x09,0x0a,0x0b, 0x04,0x05,0x06,0x07
.byte 0x00,0x01,0x02,0x03, 0x04,0x05,0x06,0x07, 0x18,0x19,0x1a,0x1b, 0x14,0x15,0x16,0x17
.byte 0x08,0x09,0x0a,0x0b, 0x0c,0x0d,0x0e,0x0f, 0x10,0x11,0x12,0x13, 0x1c,0x1d,0x1e,0x1f
.byte 0x08,0x09,0x0a,0x0b, 0x1c,0x1d,0x1e,0x1f, 0x0c,0x0d,0x0e,0x0f, 0x18,0x19,0x1a,0x1b
.byte 0x00,0x01,0x02,0x03, 0x04,0x05,0x06,0x07, 0x10,0x11,0x12,0x13, 0x14,0x15,0x16,0x17
.byte 0x08,0x09,0x0a,0x0b, 0x0c,0x0d,0x0e,0x0f, 0x18,0x19,0x1a,0x1b, 0x1c,0x1d,0x1e,0x1f
.byte 0x08,0x09,0x0a,0x0b, 0x0c,0x0d,0x0e,0x0f, 0x00,0x01,0x02,0x03, 0x04,0x05,0x06,0x07
.byte 0x04,0x05,0x06,0x07, 0x08,0x09,0x0a,0x0b, 0x1c,0x1d,0x1e,0x1f, 0x10,0x11,0x12,0x13
.byte 0x00,0x01,0x02,0x03, 0x0c,0x0d,0x0e,0x0f, 0x18,0x19,0x1a,0x1b, 0x14,0x15,0x16,0x17
.byte 0x00,0x01,0x02,0x03, 0x08,0x09,0x0a,0x0b, 0x14,0x15,0x16,0x17, 0x1c,0x1d,0x1e,0x1f
.byte 0x04,0x05,0x06,0x07, 0x0c,0x0d,0x0e,0x0f, 0x10,0x11,0x12,0x13, 0x18,0x19,0x1a,0x1b
.macro lvm b, r, regs:vararg
lvx \r, 0, \b
addi \b, \b, 16
.ifnb \regs
lvm \b, \regs
.endif
.endm
.macro stvm b, r, regs:vararg
stvx \r, 0, \b
addi \b, \b, 16
.ifnb \regs
stvm \b, \regs
.endif
.endm
.macro fft_calc interleave
extfunc ff_fft_calc\interleave\()_altivec
mflr r0
stw r0, 2*4(r1)
stwu r1, -(160+16*4)(r1)
get_got r11
addi r6, r1, 16*4
stvm r6, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29
mfvrsave r0
stw r0, 15*4(r1)
li r6, 0xfffffffc
mtvrsave r6
movrel r6, fft_data, r11
lvm r6, v14, v15, v16, v17, v18, v19, v20, v21
lvm r6, v22, v23, v24, v25, v26, v27, v28, v29
li r9, 16
movrel r12, _ff_cos_tabs, r11
movrel r6, fft_dispatch_tab\interleave\()_altivec, r11
lwz r3, 0(r3)
subi r3, r3, 2
slwi r3, r3, 2+0
lwzx r3, r3, r6
mtctr r3
mr r3, r4
bctrl
addi r6, r1, 16*4
lvm r6, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29
lwz r6, 15*4(r1)
mtvrsave r6
lwz r1, 0(r1)
lwz r0, 2*4(r1)
mtlr r0
blr
.endm
.macro DECL_FFT suffix, bits, n, n2, n4
fft\n\suffix\()_altivec:
mflr r0
stw r0,4*(\bits-3)(r1)
bl fft\n2\()_altivec
addi2 r3,\n*4
bl fft\n4\()_altivec
addi2 r3,\n*2
bl fft\n4\()_altivec
addi2 r3,\n*-6
lwz r0,4*(\bits-3)(r1)
lwz r4,\bits*4(r12)
mtlr r0
li r5,\n/16
b fft_pass\suffix\()_altivec
.endm
.macro DECL_FFTS interleave, suffix
.text
def_fft4 \suffix
def_fft8 \suffix
def_fft16 \suffix
PASS \interleave, \suffix
DECL_FFT \suffix, 5, 32, 16, 8
DECL_FFT \suffix, 6, 64, 32, 16
DECL_FFT \suffix, 7, 128, 64, 32
DECL_FFT \suffix, 8, 256, 128, 64
DECL_FFT \suffix, 9, 512, 256, 128
DECL_FFT \suffix,10, 1024, 512, 256
DECL_FFT \suffix,11, 2048, 1024, 512
DECL_FFT \suffix,12, 4096, 2048, 1024
DECL_FFT \suffix,13, 8192, 4096, 2048
DECL_FFT \suffix,14,16384, 8192, 4096
DECL_FFT \suffix,15,32768,16384, 8192
DECL_FFT \suffix,16,65536,32768,16384
fft_calc \suffix
.rodata
.align 3
fft_dispatch_tab\suffix\()_altivec:
.int fft4\suffix\()_altivec
.int fft8\suffix\()_altivec
.int fft16\suffix\()_altivec
.int fft32\suffix\()_altivec
.int fft64\suffix\()_altivec
.int fft128\suffix\()_altivec
.int fft256\suffix\()_altivec
.int fft512\suffix\()_altivec
.int fft1024\suffix\()_altivec
.int fft2048\suffix\()_altivec
.int fft4096\suffix\()_altivec
.int fft8192\suffix\()_altivec
.int fft16384\suffix\()_altivec
.int fft32768\suffix\()_altivec
.int fft65536\suffix\()_altivec
.endm
DECL_FFTS 0
DECL_FFTS 1, _interleave
More information about the ffmpeg-devel
mailing list