HD200 - use line transfers in lcd_grey_data. This gives ~25% speedup.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@26856 a1c6a512-1295-4272-9138-f99709370657

1 files changed, 77 insertions, 7 deletions

diff --git a/firmware/target/coldfire/mpio/hd200/lcd-as-hd200.S b/firmware/target/coldfire/mpio/hd200/lcd-as-hd200.S
index 9709e210e2..a284896af7 100644
--- a/firmware/target/coldfire/mpio/hd200/lcd-as-hd200.S
+++ b/firmware/target/coldfire/mpio/hd200/lcd-as-hd200.S
@@ -125,16 +125,26 @@ lcd_mono_data:
     .type       lcd_grey_data,@function
 
 lcd_grey_data:
-    lea.l   (-6*4, %sp), %sp
-    movem.l  %d2-%d5/%a2-%a3, (%sp)
-    movem.l (6*4+4, %sp), %a0-%a2  /* values, phases, length */
+    lea.l   (-8*4, %sp), %sp
+    movem.l  %d2-%d7/%a2-%a3, (%sp) /* save some registers */
+    movem.l (8*4+4, %sp), %a0-%a2  /* values, phases, length */
     add.l   %a2, %a2
     lea.l   (%a1, %a2.l*4), %a2    /* end address */
     lea.l   LCD_BASE_ADDRESS+2, %a3 /* LCD data port address */
     moveq.l #24, %d4            /* shift count */
     move.l  #0x204081, %d5      /* bit shuffle factor */
 
-.ph_loop:
+    moveq.l #12, %d2
+    add.l   %a1, %d2
+    and.l   #0xfffffff0, %d2    /* first line bound */
+    cmp.l   %d2, %a2            /* end address lower than first line bound? */
+    bhs.s   1f
+    move.l  %a2, %d2            /* -> adjust end address of head loop */
+1:
+    cmp.l   %a1, %d2
+    bls.s   .g_head_tail_end
+
+.g_head_tail:
     movem.l (%a1), %d0-%d1      /* fetch 8 pixel phases */
 
     move.l  %d0, %d2
@@ -159,11 +169,71 @@ lcd_grey_data:
     addq.l  #8, %a1
 
     move.w  %d2, (%a3)          /* transfer second LCD byte */
+
+.g_head_tail_end:
+    cmp.l   %a1, %a2
+    bls.w   .g_end
+    lea.l   (-8, %a2), %a2
+    cmp.l   %a1, %a2
+    bls.s   .g_line_end
+
+.g_line_loop:
+    /* loop that utilize line transfers */
+    movem.l (%a1), %d0-%d3      /* fetch 2 * 8 pixels phases */
+
+    move.l  %d0, %d6
+    and.l   #0x80808080, %d6    /* %d6 = 0.......1.......2.......3....... */
+    eor.l   %d6, %d0
+    add.l   (%a0)+, %d0         /* add values to first 4 phases */
+
+    move.l  %d1, %d7
+    and.l   #0x80808080, %d7    /* %d7 = 4.......5.......6.......7....... */
+    eor.l   %d7, %d1
+    add.l   (%a0)+, %d1         /* add values to second 4 phases */
+
+    lsr.l   #4, %d7             /* %d7 = ....4.......5.......6.......7... */
+    or.l    %d7, %d6            /* %d6 = 0...4...1...5...2...6...3...7... */
+    mulu.l  %d5, %d6            /* %d6 = 01234567123.567.23..67..3...7... */
+    not.l   %d6                 /*       negate bits */
+    lsr.l   %d4, %d6            /* %d6 = ........................01234567 */
+
+    move.w  %d6, (%a3)          /* transfer first LCD byte */
+    move.w  %d6, (%a3)          /* transfer second LCD byte */
+
+    move.l  %d2, %d6
+    and.l   #0x80808080, %d6    /* %d6 = 0.......1.......2.......3....... */
+    eor.l   %d6, %d2
+    add.l   (%a0)+, %d2         /* add values to first 4 phases */
+
+    move.l  %d3, %d7
+    and.l   #0x80808080, %d7    /* %d7 = 4.......5.......6.......7....... */
+    eor.l   %d7, %d3
+    add.l   (%a0)+, %d3         /* add values to second 4 phases */
+
+    lsr.l   #4, %d7             /* %d7 = ....4.......5.......6.......7... */
+    or.l    %d7, %d6            /* %d6 = 0...4...1...5...2...6...3...7... */
+    mulu.l  %d5, %d6            /* %d6 = 01234567123.567.23..67..3...7... */
+    not.l   %d6                 /*       negate bits */
+    lsr.l   %d4, %d6            /* %d6 = ........................01234567 */
+
+    move.w  %d6, (%a3)          /* transfer first LCD byte */
+    move.w  %d6, (%a3)          /* transfer second LCD byte */
+
+    movem.l %d0-%d3, (%a1)      /* store 2 * 8 new pixel phases */
+    lea.l   (16, %a1), %a1      /* advance pointer */
+
     cmp.l   %a2, %a1
-    bls.s   .ph_loop
+    bls.s   .g_line_loop
+
+.g_line_end:
+    lea.l   (8, %a2), %a2
+    cmp.l   %a1, %a2
+    bls.s   .g_end
+    bra.w   .g_head_tail
 
-    movem.l (%sp), %d2-%d5/%a2-%a3
-    lea.l   (6*4, %sp), %sp
+.g_end:
+    movem.l (%sp), %d2-%d7/%a2-%a3
+    lea.l   (8*4, %sp), %sp
     rts
 .grey_end:
     .size    lcd_grey_data,.grey_end-lcd_grey_data