diff options
author | Diego Biurrun <diego@biurrun.de> | 2005-12-22 01:10:11 +0000 |
---|---|---|
committer | Diego Biurrun <diego@biurrun.de> | 2005-12-22 01:10:11 +0000 |
commit | bb270c0896b39e1ae9277355e3c120ed3feb64a3 (patch) | |
tree | fc2fc2b1216d19acb3879abb6ea5a3b400f43fe4 /libavcodec/armv4l | |
parent | 50827fcf44f34521df4708cdb633809b56fb9df3 (diff) | |
download | ffmpeg-streaming-bb270c0896b39e1ae9277355e3c120ed3feb64a3.zip ffmpeg-streaming-bb270c0896b39e1ae9277355e3c120ed3feb64a3.tar.gz |
COSMETICS: tabs --> spaces, some prettyprinting
Originally committed as revision 4764 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/armv4l')
-rw-r--r-- | libavcodec/armv4l/dsputil_arm.c | 10 | ||||
-rw-r--r-- | libavcodec/armv4l/dsputil_iwmmxt.c | 8 | ||||
-rw-r--r-- | libavcodec/armv4l/jrevdct_arm.S | 632 | ||||
-rw-r--r-- | libavcodec/armv4l/simple_idct_arm.S | 326 |
4 files changed, 488 insertions, 488 deletions
diff --git a/libavcodec/armv4l/dsputil_arm.c b/libavcodec/armv4l/dsputil_arm.c index 0195c3c..18aebc0 100644 --- a/libavcodec/armv4l/dsputil_arm.c +++ b/libavcodec/armv4l/dsputil_arm.c @@ -205,13 +205,13 @@ void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx) #endif c->idct_put= j_rev_dct_ARM_put; c->idct_add= j_rev_dct_ARM_add; - c->idct = j_rev_dct_ARM; + c->idct = j_rev_dct_ARM; c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;/* FF_NO_IDCT_PERM */ } else if (idct_algo==FF_IDCT_SIMPLEARM){ - c->idct_put= simple_idct_ARM_put; - c->idct_add= simple_idct_ARM_add; - c->idct = simple_idct_ARM; - c->idct_permutation_type= FF_NO_IDCT_PERM; + c->idct_put= simple_idct_ARM_put; + c->idct_add= simple_idct_ARM_add; + c->idct = simple_idct_ARM; + c->idct_permutation_type= FF_NO_IDCT_PERM; #ifdef HAVE_IPP } else if (idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_IPP){ #else diff --git a/libavcodec/armv4l/dsputil_iwmmxt.c b/libavcodec/armv4l/dsputil_iwmmxt.c index 99a55e7..10fc697 100644 --- a/libavcodec/armv4l/dsputil_iwmmxt.c +++ b/libavcodec/armv4l/dsputil_iwmmxt.c @@ -138,10 +138,10 @@ void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx) mm_flags = mm_support(); if (avctx->dsp_mask) { - if (avctx->dsp_mask & FF_MM_FORCE) - mm_flags |= (avctx->dsp_mask & 0xffff); - else - mm_flags &= ~(avctx->dsp_mask & 0xffff); + if (avctx->dsp_mask & FF_MM_FORCE) + mm_flags |= (avctx->dsp_mask & 0xffff); + else + mm_flags &= ~(avctx->dsp_mask & 0xffff); } if (!(mm_flags & MM_IWMMXT)) return; diff --git a/libavcodec/armv4l/jrevdct_arm.S b/libavcodec/armv4l/jrevdct_arm.S index 770091c..294ea47 100644 --- a/libavcodec/armv4l/jrevdct_arm.S +++ b/libavcodec/armv4l/jrevdct_arm.S @@ -1,6 +1,6 @@ /* C-like prototype : - void j_rev_dct_ARM(DCTBLOCK data) + void j_rev_dct_ARM(DCTBLOCK data) With DCTBLOCK being a pointer to an array of 64 'signed shorts' @@ -51,336 +51,336 @@ #define FIX_M_1_961570560_ID 40 #define FIX_M_2_562915447_ID 44 #define FIX_0xFFFF_ID 48 - .text - .align + .text + .align - .global j_rev_dct_ARM + .global j_rev_dct_ARM j_rev_dct_ARM: - stmdb sp!, { r4 - r12, lr } @ all callee saved regs + 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 + 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 + 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 + 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] + 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 + @ 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 + @ 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 + 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 + @ 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 + @ 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 + .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 diff --git a/libavcodec/armv4l/simple_idct_arm.S b/libavcodec/armv4l/simple_idct_arm.S index 00dcb58..70fdfac 100644 --- a/libavcodec/armv4l/simple_idct_arm.S +++ b/libavcodec/armv4l/simple_idct_arm.S @@ -51,9 +51,9 @@ #define COL_SHIFTED_1 524288 /* 1<< (COL_SHIFT-1) */ - .text - .align - .global simple_idct_ARM + .text + .align + .global simple_idct_ARM simple_idct_ARM: @@ void simple_idct_ARM(int16_t *block) @@ -120,8 +120,8 @@ __b_evaluation: ldr r11, [r12, #offW7] @ R11=W7 mul r5, r10, r7 @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) mul r7, r11, r7 @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) - teq r2, #0 @ if null avoid muls - mlane r0, r9, r2, r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) + teq r2, #0 @ if null avoid muls + mlane r0, r9, r2, r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) rsbne r2, r2, #0 @ R2=-ROWr16[3] mlane r1, r11, r2, r1 @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) mlane r5, r8, r2, r5 @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) @@ -147,7 +147,7 @@ __b_evaluation: @@ MAC16(b3, -W1, row[7]); @@ MAC16(b1, -W5, row[7]); mov r3, r3, asr #16 @ R3=ROWr16[5] - teq r3, #0 @ if null avoid muls + teq r3, #0 @ if null avoid muls mlane r0, r10, r3, r0 @ R0+=W5*ROWr16[5]=b0 mov r4, r4, asr #16 @ R4=ROWr16[7] mlane r5, r11, r3, r5 @ R5+=W7*ROWr16[5]=b2 @@ -155,7 +155,7 @@ __b_evaluation: rsbne r3, r3, #0 @ R3=-ROWr16[5] mlane r1, r8, r3, r1 @ R7-=W1*ROWr16[5]=b1 @@ R3 is free now - teq r4, #0 @ if null avoid muls + teq r4, #0 @ if null avoid muls mlane r0, r11, r4, r0 @ R0+=W7*ROWr16[7]=b0 mlane r5, r9, r4, r5 @ R5+=W3*ROWr16[7]=b2 rsbne r4, r4, #0 @ R4=-ROWr16[7] @@ -187,7 +187,7 @@ __a_evaluation: teq r2, #0 beq __end_bef_a_evaluation - add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) + add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) mul r11, r8, r4 @ R11=W2*ROWr16[2] sub r4, r6, r11 @ R4=a0-W2*ROWr16[2] (a3) add r6, r6, r11 @ R6=a0+W2*ROWr16[2] (a0) @@ -203,7 +203,7 @@ __a_evaluation: @@ a2 -= W4*row[4] @@ a3 += W4*row[4] ldrsh r11, [r14, #8] @ R11=ROWr16[4] - teq r11, #0 @ if null avoid muls + teq r11, #0 @ if null avoid muls mulne r11, r9, r11 @ R11=W4*ROWr16[4] @@ R9 is free now ldrsh r9, [r14, #12] @ R9=ROWr16[6] @@ -212,7 +212,7 @@ __a_evaluation: subne r3, r3, r11 @ R3-=W4*ROWr16[4] (a2) addne r4, r4, r11 @ R4+=W4*ROWr16[4] (a3) @@ W6 alone is no more useful, save W2*ROWr16[6] in it instead - teq r9, #0 @ if null avoid muls + teq r9, #0 @ if null avoid muls mulne r11, r10, r9 @ R11=W6*ROWr16[6] addne r6, r6, r11 @ R6+=W6*ROWr16[6] (a0) mulne r10, r8, r9 @ R10=W2*ROWr16[6] @@ -294,165 +294,165 @@ __end_row_loop: - @@ at this point, R0=block, R1-R11 (free) - @@ R12=__const_ptr_, R14=&block[n] - add r14, r0, #14 @ R14=&block[7], better start from the last col, and decrease the value until col=0, i.e. R14=block. + @@ at this point, R0=block, R1-R11 (free) + @@ R12=__const_ptr_, R14=&block[n] + add r14, r0, #14 @ R14=&block[7], better start from the last col, and decrease the value until col=0, i.e. R14=block. __col_loop: __b_evaluation2: - @@ at this point, R0=block (temp), R1-R11 (free) - @@ R12=__const_ptr_, R14=&block[n] - @@ proceed with b0-b3 first, followed by a0-a3 - @@ MUL16(b0, W1, col[8x1]); - @@ MUL16(b1, W3, col[8x1]); - @@ MUL16(b2, W5, col[8x1]); - @@ MUL16(b3, W7, col[8x1]); - @@ MAC16(b0, W3, col[8x3]); - @@ MAC16(b1, -W7, col[8x3]); - @@ MAC16(b2, -W1, col[8x3]); - @@ MAC16(b3, -W5, col[8x3]); - ldr r8, [r12, #offW1] @ R8=W1 - ldrsh r7, [r14, #16] - mul r0, r8, r7 @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) - ldr r9, [r12, #offW3] @ R9=W3 - ldr r10, [r12, #offW5] @ R10=W5 - mul r1, r9, r7 @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) - ldr r11, [r12, #offW7] @ R11=W7 - mul r5, r10, r7 @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) - ldrsh r2, [r14, #48] - mul r7, r11, r7 @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) - teq r2, #0 @ if 0, then avoid muls - mlane r0, r9, r2, r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) - rsbne r2, r2, #0 @ R2=-ROWr16[3] - mlane r1, r11, r2, r1 @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) - mlane r5, r8, r2, r5 @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) - mlane r7, r10, r2, r7 @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) - - @@ at this point, R0=b0, R1=b1, R2 (free), R3 (free), R4 (free), - @@ R5=b2, R6 (free), R7=b3, R8=W1, R9=W3, R10=W5, R11=W7, - @@ R12=__const_ptr_, R14=&block[n] - @@ MAC16(b0, W5, col[5x8]); - @@ MAC16(b2, W7, col[5x8]); - @@ MAC16(b3, W3, col[5x8]); - @@ MAC16(b1, -W1, col[5x8]); - @@ MAC16(b0, W7, col[7x8]); - @@ MAC16(b2, W3, col[7x8]); - @@ MAC16(b3, -W1, col[7x8]); - @@ MAC16(b1, -W5, col[7x8]); - ldrsh r3, [r14, #80] @ R3=COLr16[5x8] - teq r3, #0 @ if 0 then avoid muls - mlane r0, r10, r3, r0 @ R0+=W5*ROWr16[5x8]=b0 - mlane r5, r11, r3, r5 @ R5+=W7*ROWr16[5x8]=b2 - mlane r7, r9, r3, r7 @ R7+=W3*ROWr16[5x8]=b3 - rsbne r3, r3, #0 @ R3=-ROWr16[5x8] - ldrsh r4, [r14, #112] @ R4=COLr16[7x8] - mlane r1, r8, r3, r1 @ R7-=W1*ROWr16[5x8]=b1 - @@ R3 is free now - teq r4, #0 @ if 0 then avoid muls - mlane r0, r11, r4, r0 @ R0+=W7*ROWr16[7x8]=b0 - mlane r5, r9, r4, r5 @ R5+=W3*ROWr16[7x8]=b2 - rsbne r4, r4, #0 @ R4=-ROWr16[7x8] - mlane r7, r8, r4, r7 @ R7-=W1*ROWr16[7x8]=b3 - mlane r1, r10, r4, r1 @ R1-=W5*ROWr16[7x8]=b1 - @@ R4 is free now + @@ at this point, R0=block (temp), R1-R11 (free) + @@ R12=__const_ptr_, R14=&block[n] + @@ proceed with b0-b3 first, followed by a0-a3 + @@ MUL16(b0, W1, col[8x1]); + @@ MUL16(b1, W3, col[8x1]); + @@ MUL16(b2, W5, col[8x1]); + @@ MUL16(b3, W7, col[8x1]); + @@ MAC16(b0, W3, col[8x3]); + @@ MAC16(b1, -W7, col[8x3]); + @@ MAC16(b2, -W1, col[8x3]); + @@ MAC16(b3, -W5, col[8x3]); + ldr r8, [r12, #offW1] @ R8=W1 + ldrsh r7, [r14, #16] + mul r0, r8, r7 @ R0=W1*ROWr16[1]=b0 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) + ldr r9, [r12, #offW3] @ R9=W3 + ldr r10, [r12, #offW5] @ R10=W5 + mul r1, r9, r7 @ R1=W3*ROWr16[1]=b1 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) + ldr r11, [r12, #offW7] @ R11=W7 + mul r5, r10, r7 @ R5=W5*ROWr16[1]=b2 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) + ldrsh r2, [r14, #48] + mul r7, r11, r7 @ R7=W7*ROWr16[1]=b3 (ROWr16[1] must be the second arg, to have the possibility to save 1 cycle) + teq r2, #0 @ if 0, then avoid muls + mlane r0, r9, r2, r0 @ R0+=W3*ROWr16[3]=b0 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) + rsbne r2, r2, #0 @ R2=-ROWr16[3] + mlane r1, r11, r2, r1 @ R1-=W7*ROWr16[3]=b1 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) + mlane r5, r8, r2, r5 @ R5-=W1*ROWr16[3]=b2 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) + mlane r7, r10, r2, r7 @ R7-=W5*ROWr16[3]=b3 (ROWr16[3] must be the second arg, to have the possibility to save 1 cycle) + + @@ at this point, R0=b0, R1=b1, R2 (free), R3 (free), R4 (free), + @@ R5=b2, R6 (free), R7=b3, R8=W1, R9=W3, R10=W5, R11=W7, + @@ R12=__const_ptr_, R14=&block[n] + @@ MAC16(b0, W5, col[5x8]); + @@ MAC16(b2, W7, col[5x8]); + @@ MAC16(b3, W3, col[5x8]); + @@ MAC16(b1, -W1, col[5x8]); + @@ MAC16(b0, W7, col[7x8]); + @@ MAC16(b2, W3, col[7x8]); + @@ MAC16(b3, -W1, col[7x8]); + @@ MAC16(b1, -W5, col[7x8]); + ldrsh r3, [r14, #80] @ R3=COLr16[5x8] + teq r3, #0 @ if 0 then avoid muls + mlane r0, r10, r3, r0 @ R0+=W5*ROWr16[5x8]=b0 + mlane r5, r11, r3, r5 @ R5+=W7*ROWr16[5x8]=b2 + mlane r7, r9, r3, r7 @ R7+=W3*ROWr16[5x8]=b3 + rsbne r3, r3, #0 @ R3=-ROWr16[5x8] + ldrsh r4, [r14, #112] @ R4=COLr16[7x8] + mlane r1, r8, r3, r1 @ R7-=W1*ROWr16[5x8]=b1 + @@ R3 is free now + teq r4, #0 @ if 0 then avoid muls + mlane r0, r11, r4, r0 @ R0+=W7*ROWr16[7x8]=b0 + mlane r5, r9, r4, r5 @ R5+=W3*ROWr16[7x8]=b2 + rsbne r4, r4, #0 @ R4=-ROWr16[7x8] + mlane r7, r8, r4, r7 @ R7-=W1*ROWr16[7x8]=b3 + mlane r1, r10, r4, r1 @ R1-=W5*ROWr16[7x8]=b1 + @@ R4 is free now __end_b_evaluation2: - @@ at this point, R0=b0, R1=b1, R2 (free), R3 (free), R4 (free), - @@ R5=b2, R6 (free), R7=b3, R8 (free), R9 (free), R10 (free), R11 (free), - @@ R12=__const_ptr_, R14=&block[n] + @@ at this point, R0=b0, R1=b1, R2 (free), R3 (free), R4 (free), + @@ R5=b2, R6 (free), R7=b3, R8 (free), R9 (free), R10 (free), R11 (free), + @@ R12=__const_ptr_, R14=&block[n] __a_evaluation2: - @@ a0 = (W4 * col[8x0]) + (1 << (COL_SHIFT - 1)); - @@ a1 = a0 + W6 * row[2]; - @@ a2 = a0 - W6 * row[2]; - @@ a3 = a0 - W2 * row[2]; - @@ a0 = a0 + W2 * row[2]; - ldrsh r6, [r14, #0] - ldr r9, [r12, #offW4] @ R9=W4 - mul r6, r9, r6 @ R6=W4*ROWr16[0] - ldr r10, [r12, #offW6] @ R10=W6 - ldrsh r4, [r14, #32] @ R4=ROWr16[2] (a3 not defined yet) - add r6, r6, #COL_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(COL_SHIFT-1) (a0) - mul r11, r10, r4 @ R11=W6*ROWr16[2] - ldr r8, [r12, #offW2] @ R8=W2 - add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) - sub r3, r6, r11 @ R3=a0-W6*ROWr16[2] (a2) - mul r11, r8, r4 @ R11=W2*ROWr16[2] - sub r4, r6, r11 @ R4=a0-W2*ROWr16[2] (a3) - add r6, r6, r11 @ R6=a0+W2*ROWr16[2] (a0) - - @@ at this point, R0=b0, R1=b1, R2=a1, R3=a2, R4=a3, - @@ R5=b2, R6=a0, R7=b3, R8=W2, R9=W4, R10=W6, R11 (free), - @@ R12=__const_ptr_, R14=&block[n] - @@ a0 += W4*row[4] - @@ a1 -= W4*row[4] - @@ a2 -= W4*row[4] - @@ a3 += W4*row[4] - ldrsh r11, [r14, #64] @ R11=ROWr16[4] - teq r11, #0 @ if null avoid muls - mulne r11, r9, r11 @ R11=W4*ROWr16[4] - @@ R9 is free now - addne r6, r6, r11 @ R6+=W4*ROWr16[4] (a0) - subne r2, r2, r11 @ R2-=W4*ROWr16[4] (a1) - subne r3, r3, r11 @ R3-=W4*ROWr16[4] (a2) - ldrsh r9, [r14, #96] @ R9=ROWr16[6] - addne r4, r4, r11 @ R4+=W4*ROWr16[4] (a3) - @@ W6 alone is no more useful, save W2*ROWr16[6] in it instead - teq r9, #0 @ if null avoid muls - mulne r11, r10, r9 @ R11=W6*ROWr16[6] - addne r6, r6, r11 @ R6+=W6*ROWr16[6] (a0) - mulne r10, r8, r9 @ R10=W2*ROWr16[6] - @@ a0 += W6*row[6]; - @@ a3 -= W6*row[6]; - @@ a1 -= W2*row[6]; - @@ a2 += W2*row[6]; - subne r4, r4, r11 @ R4-=W6*ROWr16[6] (a3) - subne r2, r2, r10 @ R2-=W2*ROWr16[6] (a1) - addne r3, r3, r10 @ R3+=W2*ROWr16[6] (a2) + @@ a0 = (W4 * col[8x0]) + (1 << (COL_SHIFT - 1)); + @@ a1 = a0 + W6 * row[2]; + @@ a2 = a0 - W6 * row[2]; + @@ a3 = a0 - W2 * row[2]; + @@ a0 = a0 + W2 * row[2]; + ldrsh r6, [r14, #0] + ldr r9, [r12, #offW4] @ R9=W4 + mul r6, r9, r6 @ R6=W4*ROWr16[0] + ldr r10, [r12, #offW6] @ R10=W6 + ldrsh r4, [r14, #32] @ R4=ROWr16[2] (a3 not defined yet) + add r6, r6, #COL_SHIFTED_1 @ R6=W4*ROWr16[0] + 1<<(COL_SHIFT-1) (a0) + mul r11, r10, r4 @ R11=W6*ROWr16[2] + ldr r8, [r12, #offW2] @ R8=W2 + add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) + sub r3, r6, r11 @ R3=a0-W6*ROWr16[2] (a2) + mul r11, r8, r4 @ R11=W2*ROWr16[2] + sub r4, r6, r11 @ R4=a0-W2*ROWr16[2] (a3) + add r6, r6, r11 @ R6=a0+W2*ROWr16[2] (a0) + + @@ at this point, R0=b0, R1=b1, R2=a1, R3=a2, R4=a3, + @@ R5=b2, R6=a0, R7=b3, R8=W2, R9=W4, R10=W6, R11 (free), + @@ R12=__const_ptr_, R14=&block[n] + @@ a0 += W4*row[4] + @@ a1 -= W4*row[4] + @@ a2 -= W4*row[4] + @@ a3 += W4*row[4] + ldrsh r11, [r14, #64] @ R11=ROWr16[4] + teq r11, #0 @ if null avoid muls + mulne r11, r9, r11 @ R11=W4*ROWr16[4] + @@ R9 is free now + addne r6, r6, r11 @ R6+=W4*ROWr16[4] (a0) + subne r2, r2, r11 @ R2-=W4*ROWr16[4] (a1) + subne r3, r3, r11 @ R3-=W4*ROWr16[4] (a2) + ldrsh r9, [r14, #96] @ R9=ROWr16[6] + addne r4, r4, r11 @ R4+=W4*ROWr16[4] (a3) + @@ W6 alone is no more useful, save W2*ROWr16[6] in it instead + teq r9, #0 @ if null avoid muls + mulne r11, r10, r9 @ R11=W6*ROWr16[6] + addne r6, r6, r11 @ R6+=W6*ROWr16[6] (a0) + mulne r10, r8, r9 @ R10=W2*ROWr16[6] + @@ a0 += W6*row[6]; + @@ a3 -= W6*row[6]; + @@ a1 -= W2*row[6]; + @@ a2 += W2*row[6]; + subne r4, r4, r11 @ R4-=W6*ROWr16[6] (a3) + subne r2, r2, r10 @ R2-=W2*ROWr16[6] (a1) + addne r3, r3, r10 @ R3+=W2*ROWr16[6] (a2) __end_a_evaluation2: - @@ at this point, R0=b0, R1=b1, R2=a1, R3=a2, R4=a3, - @@ R5=b2, R6=a0, R7=b3, R8 (free), R9 (free), R10 (free), R11 (free), - @@ R12=__const_ptr_, R14=&block[n] - @@ col[0 ] = ((a0 + b0) >> COL_SHIFT); - @@ col[8 ] = ((a1 + b1) >> COL_SHIFT); - @@ col[16] = ((a2 + b2) >> COL_SHIFT); - @@ col[24] = ((a3 + b3) >> COL_SHIFT); - @@ col[32] = ((a3 - b3) >> COL_SHIFT); - @@ col[40] = ((a2 - b2) >> COL_SHIFT); - @@ col[48] = ((a1 - b1) >> COL_SHIFT); - @@ col[56] = ((a0 - b0) >> COL_SHIFT); - @@@@@ no optimisation here @@@@@ - add r8, r6, r0 @ R8=a0+b0 - add r9, r2, r1 @ R9=a1+b1 - mov r8, r8, asr #COL_SHIFT - mov r9, r9, asr #COL_SHIFT - strh r8, [r14, #0] - strh r9, [r14, #16] - add r8, r3, r5 @ R8=a2+b2 - add r9, r4, r7 @ R9=a3+b3 - mov r8, r8, asr #COL_SHIFT - mov r9, r9, asr #COL_SHIFT - strh r8, [r14, #32] - strh r9, [r14, #48] - sub r8, r4, r7 @ R8=a3-b3 - sub r9, r3, r5 @ R9=a2-b2 - mov r8, r8, asr #COL_SHIFT - mov r9, r9, asr #COL_SHIFT - strh r8, [r14, #64] - strh r9, [r14, #80] - sub r8, r2, r1 @ R8=a1-b1 - sub r9, r6, r0 @ R9=a0-b0 - mov r8, r8, asr #COL_SHIFT - mov r9, r9, asr #COL_SHIFT - strh r8, [r14, #96] - strh r9, [r14, #112] + @@ at this point, R0=b0, R1=b1, R2=a1, R3=a2, R4=a3, + @@ R5=b2, R6=a0, R7=b3, R8 (free), R9 (free), R10 (free), R11 (free), + @@ R12=__const_ptr_, R14=&block[n] + @@ col[0 ] = ((a0 + b0) >> COL_SHIFT); + @@ col[8 ] = ((a1 + b1) >> COL_SHIFT); + @@ col[16] = ((a2 + b2) >> COL_SHIFT); + @@ col[24] = ((a3 + b3) >> COL_SHIFT); + @@ col[32] = ((a3 - b3) >> COL_SHIFT); + @@ col[40] = ((a2 - b2) >> COL_SHIFT); + @@ col[48] = ((a1 - b1) >> COL_SHIFT); + @@ col[56] = ((a0 - b0) >> COL_SHIFT); + @@@@@ no optimisation here @@@@@ + add r8, r6, r0 @ R8=a0+b0 + add r9, r2, r1 @ R9=a1+b1 + mov r8, r8, asr #COL_SHIFT + mov r9, r9, asr #COL_SHIFT + strh r8, [r14, #0] + strh r9, [r14, #16] + add r8, r3, r5 @ R8=a2+b2 + add r9, r4, r7 @ R9=a3+b3 + mov r8, r8, asr #COL_SHIFT + mov r9, r9, asr #COL_SHIFT + strh r8, [r14, #32] + strh r9, [r14, #48] + sub r8, r4, r7 @ R8=a3-b3 + sub r9, r3, r5 @ R9=a2-b2 + mov r8, r8, asr #COL_SHIFT + mov r9, r9, asr #COL_SHIFT + strh r8, [r14, #64] + strh r9, [r14, #80] + sub r8, r2, r1 @ R8=a1-b1 + sub r9, r6, r0 @ R9=a0-b0 + mov r8, r8, asr #COL_SHIFT + mov r9, r9, asr #COL_SHIFT + strh r8, [r14, #96] + strh r9, [r14, #112] __end_col_loop: - @@ at this point, R0-R11 (free) - @@ R12=__const_ptr_, R14=&block[n] - ldr r0, [sp, #0] @ R0=block - teq r0, r14 @ compare current &block[n] to block, when block is reached, the loop is finished. - sub r14, r14, #2 - bne __col_loop + @@ at this point, R0-R11 (free) + @@ R12=__const_ptr_, R14=&block[n] + ldr r0, [sp, #0] @ R0=block + teq r0, r14 @ compare current &block[n] to block, when block is reached, the loop is finished. + sub r14, r14, #2 + bne __col_loop @@ -466,15 +466,15 @@ __end_simple_idct_ARM: @@ kind of sub-function, here not to overload the common case. __end_bef_a_evaluation: - add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) + add r2, r6, r11 @ R2=a0+W6*ROWr16[2] (a1) mul r11, r8, r4 @ R11=W2*ROWr16[2] sub r4, r6, r11 @ R4=a0-W2*ROWr16[2] (a3) add r6, r6, r11 @ R6=a0+W2*ROWr16[2] (a0) - bal __end_a_evaluation + bal __end_a_evaluation __constant_ptr__: @@ see #defines at the beginning of the source code for values. - .align + .align .word W1 .word W2 .word W3 |