arm specific code
authorFabrice Bellard <fabrice@bellard.org>
Mon, 13 Aug 2001 21:38:25 +0000 (21:38 +0000)
committerFabrice Bellard <fabrice@bellard.org>
Mon, 13 Aug 2001 21:38:25 +0000 (21:38 +0000)
Originally committed as revision 79 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/armv4l/dsputil_arm.c [new file with mode: 0644]
libavcodec/armv4l/jrevdct_arm.S [new file with mode: 0644]

diff --git a/libavcodec/armv4l/dsputil_arm.c b/libavcodec/armv4l/dsputil_arm.c
new file mode 100644 (file)
index 0000000..1cf7b4f
--- /dev/null
@@ -0,0 +1,27 @@
+/*
+ * ARMv4L optimized DSP utils
+ * Copyright (c) 2001 Lionel Ulmer.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include "../dsputil.h"
+
+extern void j_rev_dct_ARM(DCTELEM *data);
+
+void dsputil_init_armv4l(void)
+{
+  ff_idct = j_rev_dct_ARM;
+}
diff --git a/libavcodec/armv4l/jrevdct_arm.S b/libavcodec/armv4l/jrevdct_arm.S
new file mode 100644 (file)
index 0000000..76eda57
--- /dev/null
@@ -0,0 +1,386 @@
+/* 
+   C-like prototype :
+       void j_rev_dct_ARM(DCTBLOCK data)
+
+   With DCTBLOCK being a pointer to an array of 64 'signed shorts'
+
+   Copyright (c) 2001 Lionel Ulmer (lionel.ulmer@free.fr / bbrox@bbrox.org)
+
+   Permission is hereby granted, free of charge, to any person obtaining a copy
+   of this software and associated documentation files (the "Software"), to deal
+   in the Software without restriction, including without limitation the rights
+   to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+   copies of the Software, and to permit persons to whom the Software is
+   furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+   COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+   IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+   CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+       
+*/
+#define FIX_0_298631336 2446
+#define FIX_0_541196100 4433
+#define FIX_0_765366865 6270
+#define FIX_1_175875602 9633
+#define FIX_1_501321110 12299
+#define FIX_2_053119869 16819
+#define FIX_3_072711026 25172
+#define FIX_M_0_390180644 -3196
+#define FIX_M_0_899976223 -7373
+#define FIX_M_1_847759065 -15137
+#define FIX_M_1_961570560 -16069
+#define FIX_M_2_562915447 -20995
+#define FIX_0xFFFF 0xFFFF      
+               
+#define FIX_0_298631336_ID      0
+#define FIX_0_541196100_ID      4
+#define FIX_0_765366865_ID      8
+#define FIX_1_175875602_ID     12
+#define FIX_1_501321110_ID     16
+#define FIX_2_053119869_ID     20
+#define FIX_3_072711026_ID     24
+#define FIX_M_0_390180644_ID   28
+#define FIX_M_0_899976223_ID   32
+#define FIX_M_1_847759065_ID   36
+#define FIX_M_1_961570560_ID   40
+#define FIX_M_2_562915447_ID   44
+#define FIX_0xFFFF_ID          48
+       .text
+       .align
+       
+       .global j_rev_dct_ARM
+j_rev_dct_ARM:
+       stmdb   sp!, { r4 - r12, lr }   @ all callee saved regs
+
+       sub sp, sp, #4                  @ reserve some space on the stack
+       str r0, [ sp ]                  @ save the DCT pointer to the stack
+
+       mov lr, r0                      @ lr = pointer to the current row
+       mov r12, #8                     @ r12 = row-counter
+       add r11, pc, #(const_array-.-8) @ r11 = base pointer to the constants array     
+row_loop:
+       ldrsh r0, [lr, # 0]             @ r0 = 'd0'
+       ldrsh r1, [lr, # 8]             @ r1 = 'd1'
+
+       @ Optimization for row that have all items except the first set to 0
+       @ (this works as the DCTELEMS are always 4-byte aligned)
+       ldr r5, [lr, # 0]
+       ldr r2, [lr, # 4]
+       ldr r3, [lr, # 8]
+       ldr r4, [lr, #12]
+       orr r3, r3, r4
+       orr r3, r3, r2
+       orrs r5, r3, r5
+       beq end_of_row_loop             @ nothing to be done as ALL of them are '0'
+       orrs r2, r3, r1
+       beq empty_row
+       
+       ldrsh r2, [lr, # 2]             @ r2 = 'd2'
+       ldrsh r4, [lr, # 4]             @ r4 = 'd4'
+       ldrsh r6, [lr, # 6]             @ r6 = 'd6'
+       
+       ldr r3, [r11, #FIX_0_541196100_ID]
+       add r7, r2, r6
+       ldr r5, [r11, #FIX_M_1_847759065_ID]
+       mul r7, r3, r7                      @ r7 = z1
+       ldr r3, [r11, #FIX_0_765366865_ID]
+       mla r6, r5, r6, r7                  @ r6 = tmp2
+       add r5, r0, r4                      @ r5 = tmp0
+       mla r2, r3, r2, r7                  @ r2 = tmp3
+       sub r3, r0, r4                      @ r3 = tmp1
+
+       add r0, r2, r5, lsl #13             @ r0 = tmp10
+       rsb r2, r2, r5, lsl #13             @ r2 = tmp13
+       add r4, r6, r3, lsl #13             @ r4 = tmp11
+       rsb r3, r6, r3, lsl #13             @ r3 = tmp12
+
+       stmdb   sp!, { r0, r2, r3, r4 } @ save on the stack tmp10, tmp13, tmp12, tmp11
+       
+       ldrsh r3, [lr, #10]             @ r3 = 'd3'
+       ldrsh r5, [lr, #12]             @ r5 = 'd5'
+       ldrsh r7, [lr, #14]             @ r7 = 'd7'
+
+       add r0, r3, r5                  @ r0 = 'z2'
+       add r2, r1, r7                  @ r2 = 'z1'
+       add r4, r3, r7                  @ r4 = 'z3'
+       add r6, r1, r5                  @ r6 = 'z4'
+       ldr r9, [r11, #FIX_1_175875602_ID]
+       add r8, r4, r6                  @ r8 = z3 + z4
+       ldr r10, [r11, #FIX_M_0_899976223_ID]
+       mul r8, r9, r8                  @ r8 = 'z5'
+       ldr r9, [r11, #FIX_M_2_562915447_ID]
+       mul r2, r10, r2                 @ r2 = 'z1'
+       ldr r10, [r11, #FIX_M_1_961570560_ID]
+       mul r0, r9, r0                  @ r0 = 'z2'
+       ldr r9, [r11, #FIX_M_0_390180644_ID]
+       mla r4, r10, r4, r8             @ r4 = 'z3'
+       ldr r10, [r11, #FIX_0_298631336_ID]
+       mla r6, r9, r6, r8              @ r6 = 'z4'
+       ldr r9, [r11, #FIX_2_053119869_ID]
+       mla r7, r10, r7, r2             @ r7 = tmp0 + z1
+       ldr r10, [r11, #FIX_3_072711026_ID]
+       mla r5, r9, r5, r0              @ r5 = tmp1 + z2
+       ldr r9, [r11, #FIX_1_501321110_ID]
+       mla r3, r10, r3, r0             @ r3 = tmp2 + z2
+       add r7, r7, r4                  @ r7 = tmp0
+       mla r1, r9, r1, r2              @ r1 = tmp3 + z1
+       add r5, r5, r6                  @ r5 = tmp1
+       add r3, r3, r4                  @ r3 = tmp2
+       add r1, r1, r6                  @ r1 = tmp3
+
+       ldmia sp!, { r0, r2, r4, r6 } @ r0 = tmp10 / r2 = tmp13 / r4 = tmp12 / r6 = tmp11
+                                     @ r1 = tmp3  / r3 = tmp2  / r5 = tmp1  / r7 = tmp0
+       
+       @ Compute DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS)
+       add r8, r0, r1
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, # 0]
+       
+       @ Compute DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS)
+       sub r8, r0, r1
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, #14]
+       
+       @ Compute DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS)
+       add r8, r6, r3
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, # 2]
+       
+       @ Compute DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS)
+       sub r8, r6, r3
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, #12]
+       
+       @ Compute DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS)
+       add r8, r4, r5
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, # 4]
+       
+       @ Compute DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS)
+       sub r8, r4, r5
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, #10]
+       
+       @ Compute DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS)
+       add r8, r2, r7
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, # 6]
+       
+       @ Compute DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS)
+       sub r8, r2, r7
+       add r8, r8, #(1<<10)
+       mov r8, r8, asr #11
+       strh r8, [lr, # 8]
+
+       @ End of row loop
+       add lr, lr, #16
+       subs r12, r12, #1
+       bne row_loop
+       beq start_column_loop
+       
+empty_row:
+       ldr r1, [r11, #FIX_0xFFFF_ID]
+       mov r0, r0, lsl #2
+       and r0, r0, r1
+       add r0, r0, r0, lsl #16
+       str r0, [lr, # 0]
+       str r0, [lr, # 4]
+       str r0, [lr, # 8]
+       str r0, [lr, #12]
+
+end_of_row_loop:
+       @ End of loop
+       add lr, lr, #16
+       subs r12, r12, #1
+       bne row_loop
+
+start_column_loop:
+       @ Start of column loop
+       ldr lr, [ sp ]
+       mov r12, #8
+column_loop:
+       ldrsh r0, [lr, #( 0*8)]             @ r0 = 'd0'
+       ldrsh r2, [lr, #( 4*8)]             @ r2 = 'd2'
+       ldrsh r4, [lr, #( 8*8)]             @ r4 = 'd4'
+       ldrsh r6, [lr, #(12*8)]             @ r6 = 'd6'
+
+       ldr r3, [r11, #FIX_0_541196100_ID]
+       add r1, r2, r6
+       ldr r5, [r11, #FIX_M_1_847759065_ID]
+       mul r1, r3, r1                      @ r1 = z1
+       ldr r3, [r11, #FIX_0_765366865_ID]
+       mla r6, r5, r6, r1                  @ r6 = tmp2
+       add r5, r0, r4                      @ r5 = tmp0
+       mla r2, r3, r2, r1                  @ r2 = tmp3
+       sub r3, r0, r4                      @ r3 = tmp1
+
+       add r0, r2, r5, lsl #13             @ r0 = tmp10
+       rsb r2, r2, r5, lsl #13             @ r2 = tmp13
+       add r4, r6, r3, lsl #13             @ r4 = tmp11
+       rsb r6, r6, r3, lsl #13             @ r6 = tmp12
+
+       ldrsh r1, [lr, #( 2*8)]             @ r1 = 'd1'
+       ldrsh r3, [lr, #( 6*8)]             @ r3 = 'd3'
+       ldrsh r5, [lr, #(10*8)]             @ r5 = 'd5'
+       ldrsh r7, [lr, #(14*8)]             @ r7 = 'd7'
+
+       @ Check for empty odd column (happens about 20 to 25 % of the time according to my stats)
+       orr r9, r1, r3
+       orr r10, r5, r7
+       orrs r10, r9, r10
+       beq empty_odd_column
+
+       stmdb   sp!, { r0, r2, r4, r6 } @ save on the stack tmp10, tmp13, tmp12, tmp11
+               
+       add r0, r3, r5                  @ r0 = 'z2'
+       add r2, r1, r7                  @ r2 = 'z1'
+       add r4, r3, r7                  @ r4 = 'z3'
+       add r6, r1, r5                  @ r6 = 'z4'
+       ldr r9, [r11, #FIX_1_175875602_ID]
+       add r8, r4, r6
+       ldr r10, [r11, #FIX_M_0_899976223_ID]
+       mul r8, r9, r8                  @ r8 = 'z5'
+       ldr r9, [r11, #FIX_M_2_562915447_ID]
+       mul r2, r10, r2                 @ r2 = 'z1'
+       ldr r10, [r11, #FIX_M_1_961570560_ID]
+       mul r0, r9, r0                  @ r0 = 'z2'
+       ldr r9, [r11, #FIX_M_0_390180644_ID]
+       mla r4, r10, r4, r8             @ r4 = 'z3'
+       ldr r10, [r11, #FIX_0_298631336_ID]
+       mla r6, r9, r6, r8              @ r6 = 'z4'
+       ldr r9, [r11, #FIX_2_053119869_ID]
+       mla r7, r10, r7, r2             @ r7 = tmp0 + z1
+       ldr r10, [r11, #FIX_3_072711026_ID]
+       mla r5, r9, r5, r0              @ r5 = tmp1 + z2
+       ldr r9, [r11, #FIX_1_501321110_ID]
+       mla r3, r10, r3, r0             @ r3 = tmp2 + z2
+       add r7, r7, r4                  @ r7 = tmp0
+       mla r1, r9, r1, r2              @ r1 = tmp3 + z1
+       add r5, r5, r6                  @ r5 = tmp1
+       add r3, r3, r4                  @ r3 = tmp2
+       add r1, r1, r6                  @ r1 = tmp3     
+       
+       ldmia sp!, { r0, r2, r4, r6 } @ r0 = tmp10 / r2 = tmp13 / r4 = tmp11 / r6 = tmp12
+                                     @ r1 = tmp3  / r3 = tmp2  / r5 = tmp1  / r7 = tmp0        
+
+       @ Compute DESCALE(tmp10 + tmp3, CONST_BITS+PASS1_BITS+3)
+       add r8, r0, r1
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #( 0*8)]
+       
+       @ Compute DESCALE(tmp10 - tmp3, CONST_BITS+PASS1_BITS+3)
+       sub r8, r0, r1
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #(14*8)]
+       
+       @ Compute DESCALE(tmp11 + tmp2, CONST_BITS+PASS1_BITS+3)
+       add r8, r4, r3
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #( 2*8)]
+       
+       @ Compute DESCALE(tmp11 - tmp2, CONST_BITS+PASS1_BITS+3)
+       sub r8, r4, r3
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #(12*8)]
+       
+       @ Compute DESCALE(tmp12 + tmp1, CONST_BITS+PASS1_BITS+3)
+       add r8, r6, r5
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #( 4*8)]
+       
+       @ Compute DESCALE(tmp12 - tmp1, CONST_BITS+PASS1_BITS+3)
+       sub r8, r6, r5
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #(10*8)]
+       
+       @ Compute DESCALE(tmp13 + tmp0, CONST_BITS+PASS1_BITS+3)
+       add r8, r2, r7
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #( 6*8)]
+       
+       @ Compute DESCALE(tmp13 - tmp0, CONST_BITS+PASS1_BITS+3)
+       sub r8, r2, r7
+       add r8, r8, #(1<<17)
+       mov r8, r8, asr #18
+       strh r8, [lr, #( 8*8)]
+
+       @ End of row loop
+       add lr, lr, #2
+       subs r12, r12, #1
+       bne column_loop
+       beq the_end
+       
+empty_odd_column:
+       @ Compute DESCALE(tmp10 + tmp3, CONST_BITS+PASS1_BITS+3)
+       @ Compute DESCALE(tmp10 - tmp3, CONST_BITS+PASS1_BITS+3)
+       add r0, r0, #(1<<17)
+       mov r0, r0, asr #18
+       strh r0, [lr, #( 0*8)]
+       strh r0, [lr, #(14*8)]
+       
+       @ Compute DESCALE(tmp11 + tmp2, CONST_BITS+PASS1_BITS+3)
+       @ Compute DESCALE(tmp11 - tmp2, CONST_BITS+PASS1_BITS+3)
+       add r4, r4, #(1<<17)
+       mov r4, r4, asr #18
+       strh r4, [lr, #( 2*8)]
+       strh r4, [lr, #(12*8)]
+       
+       @ Compute DESCALE(tmp12 + tmp1, CONST_BITS+PASS1_BITS+3)
+       @ Compute DESCALE(tmp12 - tmp1, CONST_BITS+PASS1_BITS+3)
+       add r6, r6, #(1<<17)
+       mov r6, r6, asr #18
+       strh r6, [lr, #( 4*8)]
+       strh r6, [lr, #(10*8)]
+       
+       @ Compute DESCALE(tmp13 + tmp0, CONST_BITS+PASS1_BITS+3)
+       @ Compute DESCALE(tmp13 - tmp0, CONST_BITS+PASS1_BITS+3)
+       add r2, r2, #(1<<17)
+       mov r2, r2, asr #18
+       strh r2, [lr, #( 6*8)]
+       strh r2, [lr, #( 8*8)]
+
+       @ End of row loop
+       add lr, lr, #2
+       subs r12, r12, #1
+       bne column_loop
+               
+the_end:       
+       @ The end....
+       add sp, sp, #4
+       ldmia   sp!, { r4 - r12, pc }   @ restore callee saved regs and return
+
+const_array:
+       .align
+       .word FIX_0_298631336
+       .word FIX_0_541196100
+       .word FIX_0_765366865
+       .word FIX_1_175875602
+       .word FIX_1_501321110
+       .word FIX_2_053119869
+       .word FIX_3_072711026
+       .word FIX_M_0_390180644
+       .word FIX_M_0_899976223
+       .word FIX_M_1_847759065
+       .word FIX_M_1_961570560
+       .word FIX_M_2_562915447
+       .word FIX_0xFFFF