From 75bd4adbc22356b890d61d1e6dfca60013b03d04 Mon Sep 17 00:00:00 2001 From: Jens Arnold Date: Mon, 1 Dec 2008 13:21:06 +0000 Subject: [PATCH] Shuffling around register allocation allows to keep decoded0 and decoded1 in registers, for a slight speedup. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19287 a1c6a512-1295-4272-9138-f99709370657 --- apps/codecs/demac/libdemac/predictor-cf.S | 223 +++++++++++----------- 1 file changed, 108 insertions(+), 115 deletions(-) diff --git a/apps/codecs/demac/libdemac/predictor-cf.S b/apps/codecs/demac/libdemac/predictor-cf.S index 0a1ffe9ead..cd2e07fd5e 100644 --- a/apps/codecs/demac/libdemac/predictor-cf.S +++ b/apps/codecs/demac/libdemac/predictor-cf.S @@ -69,16 +69,15 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA | int count) predictor_decode_stereo: - lea.l (-14*4,%sp), %sp - movem.l %d2-%d7/%a2-%a6, (3*4,%sp) + lea.l (-12*4,%sp), %sp + movem.l %d2-%d7/%a2-%a6, (4,%sp) - movem.l (14*4+8,%sp), %d0-%d2 - movem.l %d0-%d2, (%sp) | (%sp) = decoded0 - | (4,%sp) = decoded1 - | (8,%sp) = count + movem.l (12*4+8,%sp), %a3-%a5 | %a3 = decoded0 + | %a4 = decoded1 + move.l %a5, (%sp) | (%sp) = count move.l #0, %macsr | signed integer mode - move.l (14*4+4,%sp), %a6 | %a6 = p + move.l (12*4+4,%sp), %a6 | %a6 = p move.l (%a6), %a5 | %a5 = p->buf .loop: @@ -98,15 +97,15 @@ predictor_decode_stereo: move.l %d2, (YDELAYA-4,%a5) | p->buf[YDELAYA-1] = %d2 - movem.l (YcoeffsA,%a6), %a0-%a3 | %a0 = p->YcoeffsA[0] - | %a1 = p->YcoeffsA[1] - | %a2 = p->YcoeffsA[2] - | %a3 = p->YcoeffsA[3] + movem.l (YcoeffsA,%a6), %d4-%d7 | %d4 = p->YcoeffsA[0] + | %d5 = p->YcoeffsA[1] + | %d6 = p->YcoeffsA[2] + | %d7 = p->YcoeffsA[3] - mac.l %d3, %a0, %acc0 | %acc0 = p->buf[YDELAYA] * p->YcoeffsA[0] - mac.l %d2, %a1, %acc0 | %acc0 += p->buf[YDELAYA-1] * p->YcoeffsA[1] - mac.l %d1, %a2, %acc0 | %acc0 += p->buf[YDELAYA-2] * p->YcoeffsA[2] - mac.l %d0, %a3, %acc0 | %acc0 += p->buf[YDELAYA-3] * p->YcoeffsA[3] + mac.l %d3, %d4, %acc0 | %acc0 = p->buf[YDELAYA] * p->YcoeffsA[0] + mac.l %d2, %d5, %acc0 | %acc0 += p->buf[YDELAYA-1] * p->YcoeffsA[1] + mac.l %d1, %d6, %acc0 | %acc0 += p->buf[YDELAYA-2] * p->YcoeffsA[2] + mac.l %d0, %d7, %acc0 | %acc0 += p->buf[YDELAYA-3] * p->YcoeffsA[3] move.l %d3, (YDELAYA,%a5) | p->buf[YDELAYA] = %d3 @@ -151,17 +150,17 @@ predictor_decode_stereo: move.l %d7, (YDELAYB-4,%a5) | p->buf[YDELAYB-1] = %d7 - movem.l (YcoeffsB,%a6), %d2/%a0-%a3 | %d2 = p->YcoeffsB[0] - | %a0 = p->YcoeffsB[1] - | %a1 = p->YcoeffsB[2] - | %a2 = p->YcoeffsB[3] - | %a3 = p->YcoeffsB[4] + movem.l (YcoeffsB,%a6), %d1-%d2/%a0-%a2 | %d1 = p->YcoeffsB[0] + | %d2 = p->YcoeffsB[1] + | %a0 = p->YcoeffsB[2] + | %a1 = p->YcoeffsB[3] + | %a2 = p->YcoeffsB[4] - mac.l %d3, %d2, %acc0 | %acc0 = p->buf[YDELAYB] * p->YcoeffsB[0] - mac.l %d7, %a0, %acc0 | %acc0 += p->buf[YDELAYB-1] * p->YcoeffsB[1] - mac.l %d6, %a1, %acc0 | %acc0 += p->buf[YDELAYB-2] * p->YcoeffsB[2] - mac.l %d5, %a2, %acc0 | %acc0 += p->buf[YDELAYB-3] * p->YcoeffsB[3] - mac.l %d4, %a3, %acc0 | %acc0 += p->buf[YDELAYB-4] * p->YcoeffsB[4] + mac.l %d3, %d1, %acc0 | %acc0 = p->buf[YDELAYB] * p->YcoeffsB[0] + mac.l %d7, %d2, %acc0 | %acc0 += p->buf[YDELAYB-1] * p->YcoeffsB[1] + mac.l %d6, %a0, %acc0 | %acc0 += p->buf[YDELAYB-2] * p->YcoeffsB[2] + mac.l %d5, %a1, %acc0 | %acc0 += p->buf[YDELAYB-3] * p->YcoeffsB[3] + mac.l %d4, %a2, %acc0 | %acc0 += p->buf[YDELAYB-4] * p->YcoeffsB[4] move.l %d3, (YDELAYB, %a5) | p->buf[YDELAYB] = %d3 @@ -180,40 +179,37 @@ predictor_decode_stereo: 1: | %d3 = SIGN(%d3) move.l %d3, (YADAPTCOEFFSB, %a5) | p->buf[YADAPTCOEFFSB] = %d3 - movclr.l %acc0, %d1 + movclr.l %acc0, %d4 | %d0 still contains predictionA - | %d1 contains predictionB + | %d4 contains predictionB | Finish Predictor Y - move.l (%sp), %a4 | %a4 = decoded0 - asr.l #1, %d1 - add.l %d1, %d0 | %d0 += (%d1 >> 1) - move.l (%a4), %d5 | %d5 = *decoded0 - move.l %d5, %d1 | %d1 = %d5 + asr.l #1, %d4 + add.l %d4, %d0 | %d0 += (%d1 >> 1) + move.l (%a3), %d5 | %d5 = *decoded0 + move.l %d5, %d4 | %d4 = %d5 asr.l #8, %d0 asr.l #2, %d0 | %d0 >>= 10 - add.l %d0, %d1 | %d1 += %d0 - move.l %d1, (YlastA,%a6) | p->YlastA = %d1 + add.l %d0, %d4 | %d4 += %d0 + move.l %d4, (YlastA,%a6) | p->YlastA = %d4 move.l (YfilterA,%a6), %d6 | %d6 = p->YfilterA move.l %d6, %d0 lsl.l #5, %d6 sub.l %d0, %d6 | %d6 = 31 * %d6 asr.l #5, %d6 | %d6 >>= 5 - add.l %d6, %d1 - move.l %d1, (YfilterA,%a6) | p->YfilterA = %d1 + add.l %d6, %d4 + move.l %d4, (YfilterA,%a6) | p->YfilterA = %d4 - | %d1 contains p->YfilterA - | %a4 contains decoded0 + | %d4 contains p->YfilterA | %d5 contains *decoded0 - | %d2, %a0, %a1, %a2, %a3 contain p->YcoeffsB[0..4] + | %d1, %d2, %a0, %a1, %a2 contain p->YcoeffsB[0..4] | %d7, %d3 contain p->buf[YADAPTCOEFFSB-1] and p->buf[YADAPTCOEFFSB] - move.l %d1, (%a4)+ | *(decoded0++) = %d1 (p->YfilterA) - move.l %a4, (%sp) | save decoded0 + move.l %d4, (%a3)+ | *(decoded0++) = %d1 (p->YfilterA) tst.l %d5 beq.s 3f @@ -225,13 +221,13 @@ predictor_decode_stereo: | *decoded0 > 0 - sub.l %d3, %d2 | %d2 = p->YcoeffsB[0] - p->buf[YADAPTCOEFFSB] - sub.l %d7, %a0 | %a0 = p->YcoeffsB[1] - p->buf[YADAPTCOEFFSB-1] - sub.l %d6, %a1 | %a1 = p->YcoeffsB[2] - p->buf[YADAPTCOEFFSB-2] - sub.l %d5, %a2 | %a2 = p->YcoeffsB[3] - p->buf[YADAPTCOEFFSB-3] - sub.l %d4, %a3 | %a3 = p->YcoeffsB[4] - p->buf[YADAPTCOEFFSB-4] - - movem.l %d2/%a0-%a3, (YcoeffsB,%a6) | Save p->YcoeffsB[] + sub.l %d3, %d1 | %d1 = p->YcoeffsB[0] - p->buf[YADAPTCOEFFSB] + sub.l %d7, %d2 | %d2 = p->YcoeffsB[1] - p->buf[YADAPTCOEFFSB-1] + sub.l %d6, %a0 | %a0 = p->YcoeffsB[2] - p->buf[YADAPTCOEFFSB-2] + sub.l %d5, %a1 | %a1 = p->YcoeffsB[3] - p->buf[YADAPTCOEFFSB-3] + sub.l %d4, %a2 | %a2 = p->YcoeffsB[4] - p->buf[YADAPTCOEFFSB-4] + + movem.l %d1-%d2/%a0-%a2, (YcoeffsB,%a6) | Save p->YcoeffsB[] movem.l (YcoeffsA,%a6), %d4-%d7 | %d4 = p->YcoeffsA[0] | %d5 = p->YcoeffsA[1] @@ -252,13 +248,13 @@ predictor_decode_stereo: 1: | *decoded0 < 0 - add.l %d3, %d2 | %d2 = p->YcoeffsB[0] + p->buf[YADAPTCOEFFSB] - add.l %d7, %a0 | %a0 = p->YcoeffsB[1] + p->buf[YADAPTCOEFFSB-1] - add.l %d6, %a1 | %a1 = p->YcoeffsB[2] + p->buf[YADAPTCOEFFSB-2] - add.l %d5, %a2 | %a2 = p->YcoeffsB[3] + p->buf[YADAPTCOEFFSB-3] - add.l %d4, %a3 | %a3 = p->YcoeffsB[4] + p->buf[YADAPTCOEFFSB-4] - - movem.l %d2/%a0-%a3, (YcoeffsB,%a6) | Save p->YcoeffsB[] + add.l %d3, %d1 | %d1 = p->YcoeffsB[0] + p->buf[YADAPTCOEFFSB] + add.l %d7, %d2 | %d2 = p->YcoeffsB[1] + p->buf[YADAPTCOEFFSB-1] + add.l %d6, %a0 | %a0 = p->YcoeffsB[2] + p->buf[YADAPTCOEFFSB-2] + add.l %d5, %a1 | %a1 = p->YcoeffsB[3] + p->buf[YADAPTCOEFFSB-3] + add.l %d4, %a2 | %a2 = p->YcoeffsB[4] + p->buf[YADAPTCOEFFSB-4] + + movem.l %d1-%d2/%a0-%a2, (YcoeffsB,%a6) | Save p->YcoeffsB[] movem.l (YcoeffsA,%a6), %d4-%d7 | %d4 = p->YcoeffsA[0] | %d5 = p->YcoeffsA[1] @@ -295,15 +291,15 @@ predictor_decode_stereo: move.l %d2, (XDELAYA-4,%a5) | p->buf[XDELAYA-1] = %d2 - movem.l (XcoeffsA,%a6), %a0-%a3 | %a0 = p->XcoeffsA[0] - | %a1 = p->XcoeffsA[1] - | %a2 = p->XcoeffsA[2] - | %a3 = p->XcoeffsA[3] + movem.l (XcoeffsA,%a6), %d4-%d7 | %d4 = p->XcoeffsA[0] + | %d5 = p->XcoeffsA[1] + | %d6 = p->XcoeffsA[2] + | %d7 = p->XcoeffsA[3] - mac.l %d3, %a0, %acc0 | %acc0 = p->buf[XDELAYA] * p->XcoeffsA[0] - mac.l %d2, %a1, %acc0 | %acc0 += p->buf[XDELAYA-1] * p->XcoeffsA[1] - mac.l %d1, %a2, %acc0 | %acc0 += p->buf[XDELAYA-2] * p->XcoeffsA[2] - mac.l %d0, %a3, %acc0 | %acc0 += p->buf[XDELAYA-3] * p->XcoeffsA[3] + mac.l %d3, %d4, %acc0 | %acc0 = p->buf[XDELAYA] * p->XcoeffsA[0] + mac.l %d2, %d5, %acc0 | %acc0 += p->buf[XDELAYA-1] * p->XcoeffsA[1] + mac.l %d1, %d6, %acc0 | %acc0 += p->buf[XDELAYA-2] * p->XcoeffsA[2] + mac.l %d0, %d7, %acc0 | %acc0 += p->buf[XDELAYA-3] * p->XcoeffsA[3] move.l %d3, (XDELAYA,%a5) | p->buf[XDELAYA] = %d3 @@ -348,17 +344,17 @@ predictor_decode_stereo: move.l %d7, (XDELAYB-4,%a5) | p->buf[XDELAYB-1] = %d7 - movem.l (XcoeffsB,%a6), %d2/%a0-%a3 | %d2 = p->XcoeffsB[0] - | %a0 = p->XcoeffsB[1] - | %a1 = p->XcoeffsB[2] - | %a2 = p->XcoeffsB[3] - | %a3 = p->XcoeffsB[4] + movem.l (XcoeffsB,%a6), %d1-%d2/%a0-%a2 | %d1 = p->XcoeffsB[0] + | %d2 = p->XcoeffsB[1] + | %a0 = p->XcoeffsB[2] + | %a1 = p->XcoeffsB[3] + | %a2 = p->XcoeffsB[4] - mac.l %d3, %d2, %acc0 | %acc0 = p->buf[XDELAYB] * p->XcoeffsB[0] - mac.l %d7, %a0, %acc0 | %acc0 += p->buf[XDELAYB-1] * p->XcoeffsB[1] - mac.l %d6, %a1, %acc0 | %acc0 += p->buf[XDELAYB-2] * p->XcoeffsB[2] - mac.l %d5, %a2, %acc0 | %acc0 += p->buf[XDELAYB-3] * p->XcoeffsB[3] - mac.l %d4, %a3, %acc0 | %acc0 += p->buf[XDELAYB-4] * p->XcoeffsB[4] + mac.l %d3, %d1, %acc0 | %acc0 = p->buf[XDELAYB] * p->XcoeffsB[0] + mac.l %d7, %d2, %acc0 | %acc0 += p->buf[XDELAYB-1] * p->XcoeffsB[1] + mac.l %d6, %a0, %acc0 | %acc0 += p->buf[XDELAYB-2] * p->XcoeffsB[2] + mac.l %d5, %a1, %acc0 | %acc0 += p->buf[XDELAYB-3] * p->XcoeffsB[3] + mac.l %d4, %a2, %acc0 | %acc0 += p->buf[XDELAYB-4] * p->XcoeffsB[4] move.l %d3, (XDELAYB, %a5) | p->buf[XDELAYB] = %d3 @@ -378,40 +374,37 @@ predictor_decode_stereo: 1: | %d3 = SIGN(%d3) move.l %d3, (XADAPTCOEFFSB, %a5) | p->buf[XADAPTCOEFFSB] = %d3 - movclr.l %acc0, %d1 + movclr.l %acc0, %d4 | %d0 still contains predictionA - | %d1 contains predictionB + | %d4 contains predictionB | Finish Predictor X - move.l (4,%sp), %a4 | %a4 = decoded1 - asr.l #1, %d1 - add.l %d1, %d0 | %d0 += (%d1 >> 1) + asr.l #1, %d4 + add.l %d4, %d0 | %d0 += (%d1 >> 1) move.l (%a4), %d5 | %d5 = *decoded1 - move.l %d5, %d1 | %d1 = %d5 + move.l %d5, %d4 | %d4 = %d5 asr.l #8, %d0 asr.l #2, %d0 | %d0 >>= 10 - add.l %d0, %d1 | %d1 += %d0 - move.l %d1, (XlastA,%a6) | p->XlastA = %d1 + add.l %d0, %d4 | %d4 += %d0 + move.l %d4, (XlastA,%a6) | p->XlastA = %d1 move.l (XfilterA,%a6), %d6 | %d6 = p->XfilterA move.l %d6, %d0 lsl.l #5, %d6 sub.l %d0, %d6 | %d6 = 31 * %d6 asr.l #5, %d6 | %d6 >>= 5 - add.l %d6, %d1 - move.l %d1, (XfilterA,%a6) | p->XfilterA = %d6 + add.l %d6, %d4 + move.l %d4, (XfilterA,%a6) | p->XfilterA = %d6 - | %d1 contains p->XfilterA - | %a4 contains decoded1 + | %d4 contains p->XfilterA | %d5 contains *decoded1 - | %d2, %a0, %a1, %a2, %a3 contain p->XcoeffsB[0..4] + | %d1, %d2, %a0, %a1, %a2 contain p->XcoeffsB[0..4] | %d7, %d3 contain p->buf[XADAPTCOEFFSB-1] and p->buf[XADAPTCOEFFSB] - move.l %d1, (%a4)+ | *(decoded1++) = %d1 (p->XfilterA) - move.l %a4, (4,%sp) | save decoded1 + move.l %d4, (%a4)+ | *(decoded1++) = %d1 (p->XfilterA) tst.l %d5 beq.s 3f @@ -423,13 +416,13 @@ predictor_decode_stereo: | *decoded1 > 0 - sub.l %d3, %d2 | %d2 = p->XcoeffsB[0] - p->buf[XADAPTCOEFFSB] - sub.l %d7, %a0 | %a0 = p->XcoeffsB[1] - p->buf[XADAPTCOEFFSB-1] - sub.l %d6, %a1 | %a1 = p->XcoeffsB[2] - p->buf[XADAPTCOEFFSB-2] - sub.l %d5, %a2 | %a2 = p->XcoeffsB[3] - p->buf[XADAPTCOEFFSB-3] - sub.l %d4, %a3 | %a3 = p->XcoeffsB[4] - p->buf[XADAPTCOEFFSB-4] + sub.l %d3, %d1 | %d1 = p->XcoeffsB[0] - p->buf[XADAPTCOEFFSB] + sub.l %d7, %d2 | %d2 = p->XcoeffsB[1] - p->buf[XADAPTCOEFFSB-1] + sub.l %d6, %a0 | %a0 = p->XcoeffsB[2] - p->buf[XADAPTCOEFFSB-2] + sub.l %d5, %a1 | %a1 = p->XcoeffsB[3] - p->buf[XADAPTCOEFFSB-3] + sub.l %d4, %a2 | %a2 = p->XcoeffsB[4] - p->buf[XADAPTCOEFFSB-4] - movem.l %d2/%a0-%a3, (XcoeffsB,%a6) | Save p->XcoeffsB[] + movem.l %d1-%d2/%a0-%a2, (XcoeffsB,%a6) | Save p->XcoeffsB[] movem.l (XcoeffsA,%a6), %d4-%d7 | %d4 = p->XcoeffsA[0] | %d5 = p->XcoeffsA[1] @@ -450,13 +443,13 @@ predictor_decode_stereo: 1: | *decoded1 < 0 - add.l %d3, %d2 | %d2 = p->XcoeffsB[0] + p->buf[XADAPTCOEFFSB] - add.l %d7, %a0 | %a0 = p->XcoeffsB[1] + p->buf[XADAPTCOEFFSB-1] - add.l %d6, %a1 | %a1 = p->XcoeffsB[2] + p->buf[XADAPTCOEFFSB-2] - add.l %d5, %a2 | %a2 = p->XcoeffsB[3] + p->buf[XADAPTCOEFFSB-3] - add.l %d4, %a3 | %a3 = p->XcoeffsB[4] + p->buf[XADAPTCOEFFSB-4] + add.l %d3, %d1 | %d1 = p->XcoeffsB[0] + p->buf[XADAPTCOEFFSB] + add.l %d7, %d2 | %d2 = p->XcoeffsB[1] + p->buf[XADAPTCOEFFSB-1] + add.l %d6, %a0 | %a0 = p->XcoeffsB[2] + p->buf[XADAPTCOEFFSB-2] + add.l %d5, %a1 | %a1 = p->XcoeffsB[3] + p->buf[XADAPTCOEFFSB-3] + add.l %d4, %a2 | %a2 = p->XcoeffsB[4] + p->buf[XADAPTCOEFFSB-4] - movem.l %d2/%a0-%a3, (XcoeffsB,%a6) | Save p->XcoeffsB[] + movem.l %d1-%d2/%a0-%a2, (XcoeffsB,%a6) | Save p->XcoeffsB[] movem.l (XcoeffsA,%a6), %d4-%d7 | %d4 = p->XcoeffsA[0] | %d5 = p->XcoeffsA[1] @@ -482,42 +475,42 @@ predictor_decode_stereo: addq.l #4, %a5 | p->buf++ - lea.l (historybuffer+PREDICTOR_HISTORY_SIZE*4,%a6), %a3 - | %a3 = &p->historybuffer[PREDICTOR_HISTORY_SIZE] + lea.l (historybuffer+PREDICTOR_HISTORY_SIZE*4,%a6), %a2 + | %a2 = &p->historybuffer[PREDICTOR_HISTORY_SIZE] - cmp.l %a3, %a5 + cmp.l %a2, %a5 beq.s .move_hist | The history buffer is full, we need to do a memmove - subq.l #1, (8,%sp) | decrease loop count + subq.l #1, (%sp) | decrease loop count bne.w .loop .done: move.l %a5, (%a6) | Save value of p->buf - movem.l (3*4,%sp), %d2-%d7/%a2-%a6 - lea.l (14*4,%sp), %sp + movem.l (4,%sp), %d2-%d7/%a2-%a6 + lea.l (12*4,%sp), %sp rts .move_hist: - lea.l (historybuffer,%a6), %a3 + lea.l (historybuffer,%a6), %a2 - | dest = %a3 (p->historybuffer) + | dest = %a2 (p->historybuffer) | src = %a5 (p->buf) | n = 200 movem.l (%a5), %d0-%d7/%a0-%a1 | 40 bytes - movem.l %d0-%d7/%a0-%a1, (%a3) + movem.l %d0-%d7/%a0-%a1, (%a2) movem.l (40,%a5), %d0-%d7/%a0-%a1 | 40 bytes - movem.l %d0-%d7/%a0-%a1, (40,%a3) + movem.l %d0-%d7/%a0-%a1, (40,%a2) movem.l (80,%a5), %d0-%d7/%a0-%a1 | 40 bytes - movem.l %d0-%d7/%a0-%a1, (80,%a3) + movem.l %d0-%d7/%a0-%a1, (80,%a2) movem.l (120,%a5), %d0-%d7/%a0-%a1 | 40 bytes - movem.l %d0-%d7/%a0-%a1, (120,%a3) + movem.l %d0-%d7/%a0-%a1, (120,%a2) movem.l (160,%a5), %d0-%d7/%a0-%a1 | 40 bytes - movem.l %d0-%d7/%a0-%a1, (160,%a3) + movem.l %d0-%d7/%a0-%a1, (160,%a2) - move.l %a3, %a5 | p->buf = &p->historybuffer[0] + move.l %a2, %a5 | p->buf = &p->historybuffer[0] - subq.l #1, (8,%sp) | decrease loop count + subq.l #1, (%sp) | decrease loop count bne.w .loop bra.s .done