FaSTTUBe/GST-Tensordecoder-ov_ep
46
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
GST-Tensordecoder-ov_ep/subprojects/gst-plugins-good/gst/goom/ppc_zoom_ultimate.s

337 lines
10 KiB
ArmAsm
Raw Blame History

; PowerPC optimized zoom for Goom
; © 2001-2003 Guillaume Borios
; This library is free software; you can redistribute it and/or
; modify it under the terms of the GNU Library General Public
; License as published by the Free Software Foundation; either
; version 2 of the License, or (at your option) any later version.
;
; This library is distributed in the hope that it will be useful,
; but WITHOUT ANY WARRANTY; without even the implied warranty of
; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
; Library General Public License for more details.
;
; You should have received a copy of the GNU Library General Public
; License along with this library; if not, write to the
; Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
; Boston, MA 02110-1301, USA.
; Change log :
; 21 Dec 2003 : Use of altivec is now determined with a parameter
; Section definition : We use a read only section
.text
; name of the function to call by C program : ppc_zoom
; We declare this label as a global to extend its scope outside this file
.globl _ppc_zoom_generic
.globl _ppc_zoom_G4
; Description :
; This routine dynamically computes and applies a zoom filter
; parameters :
; r3 <=> unsigned int sizeX (in pixels)
; r4 <=> unsigned int sizeY (in pixels)
; r5 <=> unsigned int * frompixmap
; r6 <=> unsigned int * topixmap
; r7 <=> unsigned int * brutS
; r8 <=> unsigned int * brutD
; r9 <=> unsigned int buffratio
; r10 <=> int [16][16] precalccoeffs
; globals after init
; r5 <=> frompixmap - 1 byte needed for preincremental fetch (replaces r5)
; r6 <=> topixmap - 1 byte needed for preincremental fetch (replaces r6)
; r3 <=> ax = x max in 16th of pixels (replaces old r3)
; r4 <=> ay = y max in 16th of pixels (replaces old r4)
; r20 <=> row size in bytes
; r12 <=> 0xFF00FF (mask for parallel 32 bits pixs computing)
; r30 <=> brutS - 1 byte needed for preincremental fetch (replaces r7)
; r31 <=> brutD - 1 byte needed for preincremental fetch (replaces r8)
; ABI notes :
; r1 is the Stack Pointer (SP) => Do not use
; r13..r31 are non-volatiles => Do not use
_ppc_zoom_generic:
; Saves the used non volatile registers in the Mach-O stack s Red-Zone
stmw r18,-56(r1)
; init
li r18,0 ; Default value if out of range : 0 (Black)
mr r11,r10
lis r12,0xFF
mullw r2,r3,r4 ; Number of pixels to compute
subi r30,r8,0
slwi r20,r3,2
srawi r19,r20,2
ori r12,r12,0xFF
subi r3,r3,1
subi r4,r4,1
mtspr ctr,r2 ; Init the loop count (one loop per pixel computed)
subi r31,r7,0
subi r6,r6,4
slwi r3,r3,4
slwi r4,r4,4
;pre init for loop
lwz r2,0(r31) ; px
lwz r29,4(r31) ; py
lwz r8,0(r30) ; px2
lwz r10,4(r30) ; py2
b L1
.align 5
L1:
; computes dynamically the position to fetch
sub r8,r8,r2
sub r10,r10,r29
mullw r8,r8,r9
addi r31,r31,8
mullw r10,r10,r9
addi r30,r30,8
srawi r8,r8,16
srawi r10,r10,16
add r2,r2,r8
add r29,r29,r10
; if px>ax or py>ay goto outofrange
; computes the attenuation coeffs and the original point address
rlwinm r10,r2,6,28-6,31-6 ; r10 <- (r2 << 2) & 0x000002D0 (r10=(r2%16)*4*16)
cmpl cr4,0,r2,r3
rlwimi r10, r29, 2, 28-2, 31-2 ; r10 <- ((r29 << 2) & 0x0000002D) | (r10 & !0x0000002D) (r10=(r10%16)*4 | r10)
cmpl cr7,0,r29,r4
srawi r29,r29,4 ; pos computing
bge- cr4,L4
srawi r2,r2,4 ; pos computing
mullw r29, r29,r19 ; pos computing
bge- cr7,L4
; Channels notation : 00112233 (AARRVVBB)
add r2,r2,r29 ; pos computing
lwzx r10,r11,r10 ; Loads coefs
slwi r2,r2,2 ; pos computing
add r2,r2,r5 ; pos computing
rlwinm r21,r10,0,24,31 ; Isolates coef1 (??????11 -> 00000011)
lwz r25,0(r2) ; Loads col1 -> r25
lwz r26,4(r2) ; Loads col2 -> r26
rlwinm r22,r10,24,24,31 ; Isolates coef2 (????22?? -> 00000022)
rlwinm r23,r10,16,24,31 ; Isolates coef3 (??33???? -> 00000033)
add r2,r2,r20 ; Adds one line for future load of col3 and col4
and r8, r25,r12 ; Masks col1 channels 1 & 3 : 0x00XX00XX
rlwinm r24,r10,8,24,31 ; Isolates coef4 (44?????? -> 00000044)
andi. r25,r25,0xFF00 ; Masks col1 channel 2 : 0x0000XX00
mullw r8, r8, r21 ; Applies coef1 on col1 channels 1 & 3
; computes final pixel color
and r10,r26,r12 ; Masks col2 channels 1 & 3 : 0x00XX00XX
lwz r27,0(r2) ; Loads col3 -> r27
mullw r10,r10,r22 ; Applies coef2 on col2 channels 1 & 3
mullw r25,r25,r21 ; Applies coef1 on col1 channel 2
andi. r29,r26,0xFF00 ; Masks col2 channel 2 : 0x0000XX00
mullw r29,r29,r22 ; Applies coef2 on col2 channel 2
lwz r28,4(r2) ; Loads col4 -> r28
add r8 ,r8 ,r10 ; Adds col1 & col2 channels 1 & 3
and r10,r27,r12 ; Masks col3 channels 1 & 3 : 0x00XX00XX
add r25,r25,r29 ; Adds col1 & col2 channel 2
mullw r10,r10,r23 ; Applies coef3 on col3 channels 1 & 3
andi. r29,r27,0xFF00 ; Masks col3 channel 2 : 0x0000XX00
mullw r29,r29,r23 ; Applies coef3 on col3 channel 2
lwz r2,0(r31) ; px
add r7 ,r8 ,r10 ; Adds col3 to (col1 + col2) channels 1 & 3
and r10,r28,r12 ; Masks col4 channels 1 & 3 : 0x00XX00XX
mullw r10,r10,r24 ; Applies coef4 on col4 channels 1 & 3
add r25,r25,r29 ; Adds col 3 to (col1 + col2) channel 2
lwz r8,0(r30) ; px2
andi. r28,r28,0xFF00 ; Masks col4 channel 2 : 0x0000XX00
add r7 ,r7 ,r10 ; Adds col4 to (col1 + col2 + col3) channels 1 & 3
lwz r10,4(r30) ; py2
mullw r28,r28,r24 ; Applies coef4 on col4 channel 2
srawi r7, r7, 8 ; (sum of channels 1 & 3) >> 8
lwz r29,4(r31) ; py
add r25,r25,r28 ; Adds col 4 to (col1 + col2 + col3) channel 2
rlwimi r7, r25, 24, 16, 23 ; (((sum of channels 2) >> 8 ) & 0x0000FF00) | ((sum of channels 1 and 3) & 0xFFFF00FF)
stwu r7,4(r6) ; Stores the computed pixel
bdnz L1 ; Iterate again if needed
b L3 ;goto end ; If not, returns from the function
; if out of range
L4:
stwu r18,4(r6)
lwz r8,0(r30) ; px2
lwz r10,4(r30) ; py2
lwz r2,0(r31) ; px
lwz r29,4(r31) ; py
bdnz L1
L3:
; Restore saved registers and return
lmw r18,-56(r1)
blr
_ppc_zoom_G4:
; Saves the used non volatile registers in the Mach-O stack s Red-Zone
stmw r17,-60(r1)
; init
li r18,0 ; Default value if out of range : 0 (Black)
mr r11,r10
lis r12,0xFF
mullw r2,r3,r4 ; Number of pixels to compute
subi r30,r8,0
slwi r20,r3,2
srawi r19,r20,2
ori r12,r12,0xFF
subi r3,r3,1
subi r4,r4,1
mtspr ctr,r2 ; Init the loop count (one loop per pixel computed)
subi r31,r7,0
subi r6,r6,4
slwi r3,r3,4
slwi r4,r4,4
;pre init for loop
lwz r2,0(r31) ; px
lwz r29,4(r31) ; py
lwz r8,0(r30) ; px2
lwz r10,4(r30) ; py2
;*********************
lis r17,0x0F01
b L100
.align 5
L100:
addi r6,r6,4
; Optimization to ensure the destination buffer
; won't be loaded into the data cache
rlwinm. r0,r6,0,27,31
bne+ L500
dcbz 0,r6
;dcba 0,r6
L500:
; computes dynamically the position to fetch
;mullw r8,r8,r29
;mullw r2,r2,r29
;add r2,r8,r2
;srawi r2,r2,17
sub r8,r8,r2
sub r10,r10,r29
mullw r8,r8,r9
addi r31,r31,8
mullw r10,r10,r9
addi r30,r30,8
dst r30,r17,0
srawi r8,r8,16
srawi r10,r10,16
add r2,r2,r8
add r29,r29,r10
dst r31,r17,1
; if px>ax or py>ay goto outofrange
; computes the attenuation coeffs and the original point address
rlwinm r10,r2,6,28-6,31-6 ; r10 <- (r2 << 2) & 0x000002D0 (r10=(r2%16)*4*16)
cmpl cr4,0,r2,r3
rlwimi r10, r29, 2, 28-2, 31-2 ; r10 <- ((r29 << 2) & 0x0000002D) | (r10 & !0x0000002D) (r10=(r29%16)*4 | r10)
cmpl cr7,0,r29,r4
srawi r29,r29,4 ; pos computing
bge- cr4,L400
srawi r2,r2,4 ; pos computing
mullw r29, r29,r19 ; pos computing
bge- cr7,L400
; Channels notation : 00112233 (AARRVVBB)
add r2,r2,r29 ; pos computing
lwzx r10,r11,r10 ; Loads coefs
slwi r2,r2,2 ; pos computing
add r2,r2,r5 ; pos computing
rlwinm r21,r10,0,24,31 ; Isolates coef1 (??????11 -> 00000011)
lwz r25,0(r2) ; Loads col1 -> r25
lwz r26,4(r2) ; Loads col2 -> r26
rlwinm r22,r10,24,24,31 ; Isolates coef2 (????22?? -> 00000022)
rlwinm r23,r10,16,24,31 ; Isolates coef3 (??33???? -> 00000033)
add r2,r2,r20 ; Adds one line for future load of col3 and col4
and r8, r25,r12 ; Masks col1 channels 1 & 3 : 0x00XX00XX
rlwinm r24,r10,8,24,31 ; Isolates coef4 (44?????? -> 00000044)
dst r2,r17,2
rlwinm r25,r25,0,16,23 ; Masks col1 channel 2 : 0x0000XX00
;andi. r25,r25,0xFF00 ; Masks col1 channel 2 : 0x0000XX00
mullw r8, r8, r21 ; Applies coef1 on col1 channels 1 & 3
; computes final pixel color
and r10,r26,r12 ; Masks col2 channels 1 & 3 : 0x00XX00XX
lwz r27,0(r2) ; Loads col3 -> r27
mullw r10,r10,r22 ; Applies coef2 on col2 channels 1 & 3
mullw r25,r25,r21 ; Applies coef1 on col1 channel 2
rlwinm r29,r26,0,16,23 ; Masks col2 channel 2 : 0x0000XX00
;andi. r29,r26,0xFF00 ; Masks col2 channel 2 : 0x0000XX00
mullw r29,r29,r22 ; Applies coef2 on col2 channel 2
lwz r28,4(r2) ; Loads col4 -> r28
add r8 ,r8 ,r10 ; Adds col1 & col2 channels 1 & 3
and r10,r27,r12 ; Masks col3 channels 1 & 3 : 0x00XX00XX
add r25,r25,r29 ; Adds col1 & col2 channel 2
mullw r10,r10,r23 ; Applies coef3 on col3 channels 1 & 3
rlwinm r29,r27,0,16,23 ; Masks col3 channel 2 : 0x0000XX00
;andi. r29,r27,0xFF00 ; Masks col3 channel 2 : 0x0000XX00
mullw r29,r29,r23 ; Applies coef3 on col3 channel 2
lwz r2,0(r31) ; px
add r7 ,r8 ,r10 ; Adds col3 to (col1 + col2) channels 1 & 3
and r10,r28,r12 ; Masks col4 channels 1 & 3 : 0x00XX00XX
mullw r10,r10,r24 ; Applies coef4 on col4 channels 1 & 3
add r25,r25,r29 ; Adds col 3 to (col1 + col2) channel 2
lwz r8,0(r30) ; px2
rlwinm r28,r28,0,16,23 ; Masks col4 channel 2 : 0x0000XX00
;andi. r28,r28,0xFF00 ; Masks col4 channel 2 : 0x0000XX00
add r7 ,r7 ,r10 ; Adds col4 to (col1 + col2 + col3) channels 1 & 3
lwz r10,4(r30) ; py2
mullw r28,r28,r24 ; Applies coef4 on col4 channel 2
srawi r7, r7, 8 ; (sum of channels 1 & 3) >> 8
lwz r29,4(r31) ; py
add r25,r25,r28 ; Adds col 4 to (col1 + col2 + col3) channel 2
rlwimi r7, r25, 24, 16, 23 ; (((sum of channels 2) >> 8 ) & 0x0000FF00) | ((sum of channels 1 and 3) & 0xFFFF00FF)
stw r7,0(r6) ; Stores the computed pixel
bdnz L100 ; Iterate again if needed
b L300 ;goto end ; If not, returns from the function
; if out of range
L400:
stw r18,0(r6)
lwz r8,0(r30) ; px2
lwz r10,4(r30) ; py2
lwz r2,0(r31) ; px
lwz r29,4(r31) ; py
bdnz L100
L300:
; Restore saved registers and return
lmw r17,-60(r1)
blr
Reference in New Issue View Git Blame Copy Permalink