Luca Barbato
2004-Oct-06 09:34 UTC
[Flac-dev] flac-1.1.1 completely broken on linux/ppc and on macosx if built with the standard toolchain (not xcode)
Sadly the latest optimization broke completely everything.
The asm code isn't gas compliant. the libFLAC linker script has a typo,
disabling the asm optimization and/or altivec won't let a correct build
anyway.
Instant fixes for the asm stuff:
sed -i -e"s:;:\#:" on the lpc_asm.s
to load address instead of addis+ori you could use
lis and la and PLEASE use the @l(register) and @ha macros instead of the
not gas supported hi/lo16().
eg:
lis 31,LABEL@ha
la 31,LABEL@l(31)
append use -mregnames as as options (remove the other)
(the option there isn't present in the as manual, and I could just try
to figure what's supposed to do)
there are -rpath $(libdir) when -rpath,$(libdir) should be used
in src/libFLAC/Makefile
libFLAC.la: $(libFLAC_la_OBJECTS) $(libFLAC_la_DEPENDENCIES)
$(LINK) -rpath $(libdir) $(libFLAC_la_LDFLAGS)
$(libFLAC_la_OBJECTS) $(libFLAC_la_LIBADD) $(LIBS)
the final build still doesn't link correctly (the 2 altivec asm function
aren't properly linked in, I haven't the time to investigate the issue
yet=/)
More verbose report about the no altivec bug:
If I disable altivec optimization (with --disable-asm-optimizations
--disable-altivec)
the result is:
stream_decoder.c: In function `FLAC__stream_decoder_init':
stream_decoder.c:296: error: `FLAC__lpc_restore_signal' undeclared
(first use in this function)
stream_decoder.c:296: error: (Each undeclared identifier is reported
only once
stream_decoder.c:296: error: for each function it appears in.)
stream_decoder.c:297: error: `FLAC__lpc_restore_signal_wide' undeclared
(first use in this function)
make[4]: *** [stream_decoder.lo] Error 1
make[4]: Leaving directory
`/var/tmp/portage/flac-1.1.1/work/flac-1.1.1/src/libFLAC'
Could you please at least make sure that disabling altivec optimization
won't break the build?
That's a bit higher whish:
could you please write the altivec optimization using C intrinsics? That
way they should work with no/little modification even on ppc 970 in
ppc64 mode.
Regards
lu
PS:put my address in CC to the replies.
--
Luca Barbato
Developer
Gentoo Linux http://www.gentoo.org/~lu_zero
Josh Coalson
2004-Oct-06 09:47 UTC
[Flac-dev] flac-1.1.1 completely broken on linux/ppc and on macosx if built with the standard toolchain (not xcode)
thanks for the feedback, but it would really help if you supply a patch (diff -c), I didn't understand all the changes you described. someone reported a problem with src/libFLAC/include/private/lpc.h that was fixed in CVS and may fix the problem building with asm disabled http://cvs.sourceforge.net/viewcvs.py/*checkout*/flac/flac/src/libFLAC/include/private/lpc.h?rev=1.24 Josh --- Luca Barbato <lu_zero@gentoo.org> wrote:> Sadly the latest optimization broke completely everything. > > The asm code isn't gas compliant. the libFLAC linker script has a > typo, > disabling the asm optimization and/or altivec won't let a correct > build > anyway. > > Instant fixes for the asm stuff: > > sed -i -e"s:;:\#:" on the lpc_asm.s > to load address instead of addis+ori you could use > lis and la and PLEASE use the @l(register) and @ha macros instead of > the > not gas supported hi/lo16(). > > eg: > > lis 31,LABEL@ha > la 31,LABEL@l(31) > > > append use -mregnames as as options (remove the other) > > (the option there isn't present in the as manual, and I could just > try > to figure what's supposed to do) > > there are -rpath $(libdir) when -rpath,$(libdir) should be used > > in src/libFLAC/Makefile > libFLAC.la: $(libFLAC_la_OBJECTS) $(libFLAC_la_DEPENDENCIES) > $(LINK) -rpath $(libdir) $(libFLAC_la_LDFLAGS) > $(libFLAC_la_OBJECTS) $(libFLAC_la_LIBADD) $(LIBS) > > > the final build still doesn't link correctly (the 2 altivec asm > function > aren't properly linked in, I haven't the time to investigate the > issue > yet=/) > > > More verbose report about the no altivec bug: > > If I disable altivec optimization (with --disable-asm-optimizations > --disable-altivec) > > the result is: > > stream_decoder.c: In function `FLAC__stream_decoder_init': > stream_decoder.c:296: error: `FLAC__lpc_restore_signal' undeclared > (first use in this function) > stream_decoder.c:296: error: (Each undeclared identifier is reported > only once > stream_decoder.c:296: error: for each function it appears in.) > stream_decoder.c:297: error: `FLAC__lpc_restore_signal_wide' > undeclared > (first use in this function) > make[4]: *** [stream_decoder.lo] Error 1 > make[4]: Leaving directory > `/var/tmp/portage/flac-1.1.1/work/flac-1.1.1/src/libFLAC' > > Could you please at least make sure that disabling altivec > optimization > won't break the build? > That's a bit higher whish: > could you please write the altivec optimization using C intrinsics? > That > way they should work with no/little modification even on ppc 970 in > ppc64 mode. > > Regards > > lu > > PS:put my address in CC to the replies. > > -- > Luca Barbato > Developer > Gentoo Linux http://www.gentoo.org/~lu_zero_______________________________ Do you Yahoo!? Declare Yourself - Register online to vote today! http://vote.yahoo.com
Josh Coalson
2004-Oct-09 13:46 UTC
[Flac-dev] flac-1.1.1 completely broken on linux/ppc and on macosx if built with the standard toolchain (not xcode)
--- Luca Barbato <lu_zero@gentoo.org> wrote:> The asm code isn't gas compliant.went back and looked at this again. the lpc_asm.s in FLAC 1.1.1 compiles fine with OS X's cc, but not with gas. the patched one you sent doesn't compile with the native cc. is there a common syntax that both support? or do we have to have two versions of every PPC asm file? I want the code to be able to be built with the native OS X developer tools. Josh _______________________________ Do you Yahoo!? Declare Yourself - Register online to vote today! http://vote.yahoo.com
Luca Barbato
2004-Oct-15 05:10 UTC
[Flac-dev] flac-1.1.1 completely broken on linux/ppc and on macosx if built with the standard toolchain (not xcode)
Josh Coalson wrote:> thanks for the feedback, but it would really help if you supply > a patch (diff -c), I didn't understand all the changes you > described.I hope it helps. lu -- Luca Barbato Developer Gentoo Linux http://www.gentoo.org/~lu_zero -------------- next part -------------- *** /tmp/lpc_asm.s Wed Oct 6 14:06:11 2004 --- src/libFLAC/ppc/lpc_asm.s Tue Jul 27 21:32:05 2004 *************** *** 1,93 **** ! # libFLAC - Free Lossless Audio Codec library ! # Copyright (C) 2004 Josh Coalson ! # ! # Redistribution and use in source and binary forms, with or without ! # modification, are permitted provided that the following conditions ! # are met: ! # ! # - Redistributions of source code must retain the above copyright ! # notice, this list of conditions and the following disclaimer. ! # ! # - Redistributions in binary form must reproduce the above copyright ! # notice, this list of conditions and the following disclaimer in the ! # documentation and/or other materials provided with the distribution. ! # ! # - Neither the name of the Xiph.org Foundation nor the names of its ! # contributors may be used to endorse or promote products derived from ! # this software without specific prior written permission. ! # ! # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ! # ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT ! # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR ! # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR ! # CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, ! # EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, ! # PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR ! # PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF ! # LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING ! # NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS ! # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. .text .align 2 .globl _FLAC__lpc_restore_signal_asm_ppc_altivec_16 - .type _FLAC__lpc_restore_signal_asm_ppc_altivec_16, @function - .globl _FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8 - .type _FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8, @function _FLAC__lpc_restore_signal_asm_ppc_altivec_16: ! # r3: residual[] ! # r4: data_len ! # r5: qlp_coeff[] ! # r6: order ! # r7: lp_quantization ! # r8: data[] ! ! # see src/libFLAC/lpc.c:FLAC__lpc_restore_signal() ! # these is a PowerPC/Altivec assembly version which requires bps<=16 (or actual ! # bps<=15 for mid-side coding, since that uses an extra bit) ! ! # these should be fast; the inner loop is unrolled (it takes no more than ! # 3*(order%4) instructions, all of which are arithmetic), and all of the ! # coefficients and all relevant history stay in registers, so the outer loop ! # has only one load from memory (the residual) ! # I have not yet run this through simg4, so there may be some avoidable stalls, ! # and there may be a somewhat more clever way to do the outer loop ! # the branch mechanism may prevent dynamic loading; I still need to examine ! # this issue, and there may be a more elegant method stmw r31,-4(r1) addi r9,r1,-28 li r31,0xf ! andc r9,r9,r31 # for quadword-aligned stack data ! slwi r6,r6,2 # adjust for word size slwi r4,r4,2 ! add r4,r4,r8 # r4 = data+data_len ! mfspr r0,256 # cache old vrsave ! addis r31,0,0xffff ! ori r31,r31,0xfc00 ! mtspr 256,r31 # declare VRs in vrsave ! cmplw cr0,r8,r4 # i<data_len bc 4,0,L1400 ! # load coefficients into v0-v7 and initial history into v8-v15 li r31,0xf ! and r31,r8,r31 # r31: data%4 li r11,16 ! subf r31,r31,r11 # r31: 4-(data%4) ! slwi r31,r31,3 # convert to bits for vsro li r10,-4 stw r31,-4(r9) lvewx v0,r10,r9 vspltisb v18,-1 ! vsro v18,v18,v0 # v18: mask vector li r31,0x8 lvsl v0,0,r31 --- 1,90 ---- ! ; libFLAC - Free Lossless Audio Codec library ! ; Copyright (C) 2004 Josh Coalson ! ; ! ; Redistribution and use in source and binary forms, with or without ! ; modification, are permitted provided that the following conditions ! ; are met: ! ; ! ; - Redistributions of source code must retain the above copyright ! ; notice, this list of conditions and the following disclaimer. ! ; ! ; - Redistributions in binary form must reproduce the above copyright ! ; notice, this list of conditions and the following disclaimer in the ! ; documentation and/or other materials provided with the distribution. ! ; ! ; - Neither the name of the Xiph.org Foundation nor the names of its ! ; contributors may be used to endorse or promote products derived from ! ; this software without specific prior written permission. ! ; ! ; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS ! ; ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT ! ; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR ! ; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR ! ; CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, ! ; EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, ! ; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR ! ; PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF ! ; LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING ! ; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS ! ; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. .text .align 2 .globl _FLAC__lpc_restore_signal_asm_ppc_altivec_16 .globl _FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8 _FLAC__lpc_restore_signal_asm_ppc_altivec_16: ! ; r3: residual[] ! ; r4: data_len ! ; r5: qlp_coeff[] ! ; r6: order ! ; r7: lp_quantization ! ; r8: data[] ! ! ; see src/libFLAC/lpc.c:FLAC__lpc_restore_signal() ! ; these is a PowerPC/Altivec assembly version which requires bps<=16 (or actual ! ; bps<=15 for mid-side coding, since that uses an extra bit) ! ! ; these should be fast; the inner loop is unrolled (it takes no more than ! ; 3*(order%4) instructions, all of which are arithmetic), and all of the ! ; coefficients and all relevant history stay in registers, so the outer loop ! ; has only one load from memory (the residual) ! ; I have not yet run this through simg4, so there may be some avoidable stalls, ! ; and there may be a somewhat more clever way to do the outer loop ! ; the branch mechanism may prevent dynamic loading; I still need to examine ! ; this issue, and there may be a more elegant method stmw r31,-4(r1) addi r9,r1,-28 li r31,0xf ! andc r9,r9,r31 ; for quadword-aligned stack data ! slwi r6,r6,2 ; adjust for word size slwi r4,r4,2 ! add r4,r4,r8 ; r4 = data+data_len ! mfspr r0,256 ; cache old vrsave ! addis r31,0,hi16(0xfffffc00) ! ori r31,r31,lo16(0xfffffc00) ! mtspr 256,r31 ; declare VRs in vrsave ! cmplw cr0,r8,r4 ; i<data_len bc 4,0,L1400 ! ; load coefficients into v0-v7 and initial history into v8-v15 li r31,0xf ! and r31,r8,r31 ; r31: data%4 li r11,16 ! subf r31,r31,r11 ; r31: 4-(data%4) ! slwi r31,r31,3 ; convert to bits for vsro li r10,-4 stw r31,-4(r9) lvewx v0,r10,r9 vspltisb v18,-1 ! vsro v18,v18,v0 ; v18: mask vector li r31,0x8 lvsl v0,0,r31 *************** *** 97,110 **** vspltisb v2,0 vspltisb v3,-1 vmrglw v2,v2,v3 ! vsel v0,v1,v0,v2 # v0: reversal permutation vector add r10,r5,r6 ! lvsl v17,0,r5 # v17: coefficient alignment permutation vector ! vperm v17,v17,v17,v0 # v17: reversal coefficient alignment permutation vector mr r11,r8 ! lvsl v16,0,r11 # v16: history alignment permutation vector lvx v0,0,r5 addi r5,r5,16 --- 94,107 ---- vspltisb v2,0 vspltisb v3,-1 vmrglw v2,v2,v3 ! vsel v0,v1,v0,v2 ; v0: reversal permutation vector add r10,r5,r6 ! lvsl v17,0,r5 ; v17: coefficient alignment permutation vector ! vperm v17,v17,v17,v0 ; v17: reversal coefficient alignment permutation vector mr r11,r8 ! lvsl v16,0,r11 ; v16: history alignment permutation vector lvx v0,0,r5 addi r5,r5,16 *************** *** 117,124 **** cmplw cr0,r5,r10 bc 12,0,L1101 vand v0,v0,v18 ! addis r31,0,L1307@ha ! ori r31,r31,L1307@l b L1199 L1101: --- 114,121 ---- cmplw cr0,r5,r10 bc 12,0,L1101 vand v0,v0,v18 ! addis r31,0,hi16(L1307) ! ori r31,r31,lo16(L1307) b L1199 L1101: *************** *** 131,138 **** cmplw cr0,r5,r10 bc 12,0,L1102 vand v1,v1,v18 ! addis r31,0,L1306@ha ! ori r31,r31,L1306@l b L1199 L1102: --- 128,135 ---- cmplw cr0,r5,r10 bc 12,0,L1102 vand v1,v1,v18 ! addis r31,0,hi16(L1306) ! ori r31,r31,lo16(L1306) b L1199 L1102: *************** *** 145,152 **** cmplw cr0,r5,r10 bc 12,0,L1103 vand v2,v2,v18 ! lis r31,L1305@ha ! la r31,L1305@l(r31) b L1199 L1103: --- 142,149 ---- cmplw cr0,r5,r10 bc 12,0,L1103 vand v2,v2,v18 ! addis r31,0,hi16(L1305) ! ori r31,r31,lo16(L1305) b L1199 L1103: *************** *** 159,166 **** cmplw cr0,r5,r10 bc 12,0,L1104 vand v3,v3,v18 ! lis r31,L1304@ha ! la r31,L1304@l(r31) b L1199 L1104: --- 156,163 ---- cmplw cr0,r5,r10 bc 12,0,L1104 vand v3,v3,v18 ! addis r31,0,hi16(L1304) ! ori r31,r31,lo16(L1304) b L1199 L1104: *************** *** 173,180 **** cmplw cr0,r5,r10 bc 12,0,L1105 vand v4,v4,v18 ! lis r31,L1303@ha ! la r31,L1303@l(r31) b L1199 L1105: --- 170,177 ---- cmplw cr0,r5,r10 bc 12,0,L1105 vand v4,v4,v18 ! addis r31,0,hi16(L1303) ! ori r31,r31,lo16(L1303) b L1199 L1105: *************** *** 187,194 **** cmplw cr0,r5,r10 bc 12,0,L1106 vand v5,v5,v18 ! lis r31,L1302@ha ! la r31,L1302@l(r31) b L1199 L1106: --- 184,191 ---- cmplw cr0,r5,r10 bc 12,0,L1106 vand v5,v5,v18 ! addis r31,0,hi16(L1302) ! ori r31,r31,lo16(L1302) b L1199 L1106: *************** *** 201,208 **** cmplw cr0,r5,r10 bc 12,0,L1107 vand v6,v6,v18 ! lis r31,L1301@ha ! la r31,L1301@l(r31) b L1199 L1107: --- 198,205 ---- cmplw cr0,r5,r10 bc 12,0,L1107 vand v6,v6,v18 ! addis r31,0,hi16(L1301) ! ori r31,r31,lo16(L1301) b L1199 L1107: *************** *** 213,242 **** lvx v19,0,r11 vperm v15,v19,v15,v16 vand v7,v7,v18 ! lis r31,L1300@ha ! la r31,L1300@l(r31) L1199: mtctr r31 ! # set up invariant vectors ! vspltish v16,0 # v16: zero vector li r10,-12 ! lvsr v17,r10,r8 # v17: result shift vector ! lvsl v18,r10,r3 # v18: residual shift back vector li r10,-4 stw r7,-4(r9) ! lvewx v19,r10,r9 # v19: lp_quantization vector L1200: ! vmulosh v20,v0,v8 # v20: sum vector bcctr 20,0 L1300: vmulosh v21,v7,v15 ! vsldoi v15,v15,v14,4 # increment history vaddsws v20,v20,v21 L1301: --- 210,239 ---- lvx v19,0,r11 vperm v15,v19,v15,v16 vand v7,v7,v18 ! addis r31,0,hi16(L1300) ! ori r31,r31,lo16(L1300) L1199: mtctr r31 ! ; set up invariant vectors ! vspltish v16,0 ; v16: zero vector li r10,-12 ! lvsr v17,r10,r8 ; v17: result shift vector ! lvsl v18,r10,r3 ; v18: residual shift back vector li r10,-4 stw r7,-4(r9) ! lvewx v19,r10,r9 ; v19: lp_quantization vector L1200: ! vmulosh v20,v0,v8 ; v20: sum vector bcctr 20,0 L1300: vmulosh v21,v7,v15 ! vsldoi v15,v15,v14,4 ; increment history vaddsws v20,v20,v21 L1301: *************** *** 270,342 **** vaddsws v20,v20,v21 L1307: ! vsumsws v20,v20,v16 # v20[3]: sum ! vsraw v20,v20,v19 # v20[3]: sum >> lp_quantization ! lvewx v21,0,r3 # v21[n]: *residual ! vperm v21,v21,v21,v18 # v21[3]: *residual ! vaddsws v20,v21,v20 # v20[3]: *residual + (sum >> lp_quantization) ! vsldoi v18,v18,v18,4 # increment shift vector ! vperm v21,v20,v20,v17 # v21[n]: shift for storage ! vsldoi v17,v17,v17,12 # increment shift vector stvewx v21,0,r8 vsldoi v20,v20,v20,12 ! vsldoi v8,v8,v20,4 # insert value onto history addi r3,r3,4 addi r8,r8,4 ! cmplw cr0,r8,r4 # i<data_len bc 12,0,L1200 L1400: ! mtspr 256,r0 # restore old vrsave lmw r31,-4(r1) blr _FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8: ! # r3: residual[] ! # r4: data_len ! # r5: qlp_coeff[] ! # r6: order ! # r7: lp_quantization ! # r8: data[] ! ! # see _FLAC__lpc_restore_signal_asm_ppc_altivec_16() above ! # this version assumes order<=8; it uses fewer vector registers, which should ! # save time in context switches, and has less code, which may improve ! # instruction caching stmw r31,-4(r1) addi r9,r1,-28 li r31,0xf ! andc r9,r9,r31 # for quadword-aligned stack data ! slwi r6,r6,2 # adjust for word size slwi r4,r4,2 ! add r4,r4,r8 # r4 = data+data_len ! mfspr r0,256 # cache old vrsave ! addis r31,0,0xffc0 ! ori r31,r31,0x0000 ! mtspr 256,r31 # declare VRs in vrsave ! cmplw cr0,r8,r4 # i<data_len bc 4,0,L2400 ! # load coefficients into v0-v1 and initial history into v2-v3 li r31,0xf ! and r31,r8,r31 # r31: data%4 li r11,16 ! subf r31,r31,r11 # r31: 4-(data%4) ! slwi r31,r31,3 # convert to bits for vsro li r10,-4 stw r31,-4(r9) lvewx v0,r10,r9 vspltisb v6,-1 ! vsro v6,v6,v0 # v6: mask vector li r31,0x8 lvsl v0,0,r31 --- 267,339 ---- vaddsws v20,v20,v21 L1307: ! vsumsws v20,v20,v16 ; v20[3]: sum ! vsraw v20,v20,v19 ; v20[3]: sum >> lp_quantization ! lvewx v21,0,r3 ; v21[n]: *residual ! vperm v21,v21,v21,v18 ; v21[3]: *residual ! vaddsws v20,v21,v20 ; v20[3]: *residual + (sum >> lp_quantization) ! vsldoi v18,v18,v18,4 ; increment shift vector ! vperm v21,v20,v20,v17 ; v21[n]: shift for storage ! vsldoi v17,v17,v17,12 ; increment shift vector stvewx v21,0,r8 vsldoi v20,v20,v20,12 ! vsldoi v8,v8,v20,4 ; insert value onto history addi r3,r3,4 addi r8,r8,4 ! cmplw cr0,r8,r4 ; i<data_len bc 12,0,L1200 L1400: ! mtspr 256,r0 ; restore old vrsave lmw r31,-4(r1) blr _FLAC__lpc_restore_signal_asm_ppc_altivec_16_order8: ! ; r3: residual[] ! ; r4: data_len ! ; r5: qlp_coeff[] ! ; r6: order ! ; r7: lp_quantization ! ; r8: data[] ! ! ; see _FLAC__lpc_restore_signal_asm_ppc_altivec_16() above ! ; this version assumes order<=8; it uses fewer vector registers, which should ! ; save time in context switches, and has less code, which may improve ! ; instruction caching stmw r31,-4(r1) addi r9,r1,-28 li r31,0xf ! andc r9,r9,r31 ; for quadword-aligned stack data ! slwi r6,r6,2 ; adjust for word size slwi r4,r4,2 ! add r4,r4,r8 ; r4 = data+data_len ! mfspr r0,256 ; cache old vrsave ! addis r31,0,hi16(0xffc00000) ! ori r31,r31,lo16(0xffc00000) ! mtspr 256,r31 ; declare VRs in vrsave ! cmplw cr0,r8,r4 ; i<data_len bc 4,0,L2400 ! ; load coefficients into v0-v1 and initial history into v2-v3 li r31,0xf ! and r31,r8,r31 ; r31: data%4 li r11,16 ! subf r31,r31,r11 ; r31: 4-(data%4) ! slwi r31,r31,3 ; convert to bits for vsro li r10,-4 stw r31,-4(r9) lvewx v0,r10,r9 vspltisb v6,-1 ! vsro v6,v6,v0 ; v6: mask vector li r31,0x8 lvsl v0,0,r31 *************** *** 346,359 **** vspltisb v2,0 vspltisb v3,-1 vmrglw v2,v2,v3 ! vsel v0,v1,v0,v2 # v0: reversal permutation vector add r10,r5,r6 ! lvsl v5,0,r5 # v5: coefficient alignment permutation vector ! vperm v5,v5,v5,v0 # v5: reversal coefficient alignment permutation vector mr r11,r8 ! lvsl v4,0,r11 # v4: history alignment permutation vector lvx v0,0,r5 addi r5,r5,16 --- 343,356 ---- vspltisb v2,0 vspltisb v3,-1 vmrglw v2,v2,v3 ! vsel v0,v1,v0,v2 ; v0: reversal permutation vector add r10,r5,r6 ! lvsl v5,0,r5 ; v5: coefficient alignment permutation vector ! vperm v5,v5,v5,v0 ; v5: reversal coefficient alignment permutation vector mr r11,r8 ! lvsl v4,0,r11 ; v4: history alignment permutation vector lvx v0,0,r5 addi r5,r5,16 *************** *** 366,373 **** cmplw cr0,r5,r10 bc 12,0,L2101 vand v0,v0,v6 ! lis r31,L2301@ha ! la r31,L2301@l(r31) b L2199 L2101: --- 363,370 ---- cmplw cr0,r5,r10 bc 12,0,L2101 vand v0,v0,v6 ! addis r31,0,hi16(L2301) ! ori r31,r31,lo16(L2301) b L2199 L2101: *************** *** 378,402 **** lvx v7,0,r11 vperm v3,v7,v3,v4 vand v1,v1,v6 ! lis r31,L2300@ha ! la r31,L2300@l(r31) L2199: mtctr r31 ! # set up invariant vectors ! vspltish v4,0 # v4: zero vector li r10,-12 ! lvsr v5,r10,r8 # v5: result shift vector ! lvsl v6,r10,r3 # v6: residual shift back vector li r10,-4 stw r7,-4(r9) ! lvewx v7,r10,r9 # v7: lp_quantization vector L2200: ! vmulosh v8,v0,v2 # v8: sum vector bcctr 20,0 L2300: --- 375,399 ---- lvx v7,0,r11 vperm v3,v7,v3,v4 vand v1,v1,v6 ! addis r31,0,hi16(L2300) ! ori r31,r31,lo16(L2300) L2199: mtctr r31 ! ; set up invariant vectors ! vspltish v4,0 ; v4: zero vector li r10,-12 ! lvsr v5,r10,r8 ; v5: result shift vector ! lvsl v6,r10,r3 ; v6: residual shift back vector li r10,-4 stw r7,-4(r9) ! lvewx v7,r10,r9 ; v7: lp_quantization vector L2200: ! vmulosh v8,v0,v2 ; v8: sum vector bcctr 20,0 L2300: *************** *** 405,431 **** vaddsws v8,v8,v9 L2301: ! vsumsws v8,v8,v4 # v8[3]: sum ! vsraw v8,v8,v7 # v8[3]: sum >> lp_quantization ! lvewx v9,0,r3 # v9[n]: *residual ! vperm v9,v9,v9,v6 # v9[3]: *residual ! vaddsws v8,v9,v8 # v8[3]: *residual + (sum >> lp_quantization) ! vsldoi v6,v6,v6,4 # increment shift vector ! vperm v9,v8,v8,v5 # v9[n]: shift for storage ! vsldoi v5,v5,v5,12 # increment shift vector stvewx v9,0,r8 vsldoi v8,v8,v8,12 ! vsldoi v2,v2,v8,4 # insert value onto history addi r3,r3,4 addi r8,r8,4 ! cmplw cr0,r8,r4 # i<data_len bc 12,0,L2200 L2400: ! mtspr 256,r0 # restore old vrsave lmw r31,-4(r1) blr --- 402,428 ---- vaddsws v8,v8,v9 L2301: ! vsumsws v8,v8,v4 ; v8[3]: sum ! vsraw v8,v8,v7 ; v8[3]: sum >> lp_quantization ! lvewx v9,0,r3 ; v9[n]: *residual ! vperm v9,v9,v9,v6 ; v9[3]: *residual ! vaddsws v8,v9,v8 ; v8[3]: *residual + (sum >> lp_quantization) ! vsldoi v6,v6,v6,4 ; increment shift vector ! vperm v9,v8,v8,v5 ; v9[n]: shift for storage ! vsldoi v5,v5,v5,12 ; increment shift vector stvewx v9,0,r8 vsldoi v8,v8,v8,12 ! vsldoi v2,v2,v8,4 ; insert value onto history addi r3,r3,4 addi r8,r8,4 ! cmplw cr0,r8,r4 ; i<data_len bc 12,0,L2200 L2400: ! mtspr 256,r0 ; restore old vrsave lmw r31,-4(r1) blr