X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Fvideo%2FSDL_blit_neon.S;h=dcbfd74310f9841043f45e7060b020922514f461;hb=199f36ec4a30dfc5a736a2d0c6e0840d83f7e9e7;hp=af9af365a493ffe0aa4fc8fd0907c31371c9693e;hpb=a1eff5db323aedd15ed6fbfd13d06273b54d2d6f;p=sdl_omap.git

diff --git a/src/video/SDL_blit_neon.S b/src/video/SDL_blit_neon.S
index af9af36..dcbfd74 100644
--- a/src/video/SDL_blit_neon.S
+++ b/src/video/SDL_blit_neon.S
@@ -11,11 +11,20 @@
 .text
 .align 2
 
+#define func(name) \
+    .global name; \
+    name
+
 @ void *dst, const void *src, int count, uint abits
 .macro do_argb bgr2rgb
     vdup.i8    d0, r3
 0:
+    cmp        r2, #8
+    pld        [r1, #64*2]
+    blt        3f
+1:
     vld4.8     {d4-d7}, [r1]!
+2:
 .if \bgr2rgb
     vswp       d4, d6		@ BGR->RGB
 .endif
@@ -26,6 +35,23 @@
     bxeq       lr
     nop
     b          0b
+
+3:
+    @ unaligned ending nastiness :(
+    add        r12, r1, #8*4
+    lsr        r12, #12
+    cmp        r12, r1, lsr #12 @ crossing page?
+    beq        1b               @ nope, overreading is safe
+
+    @ _wb_'s bad luck, do some slow stuff here
+    push       {r0-r2,lr}
+    sub        sp, #8*4
+    mov        r0, sp
+    lsl        r2, #2
+    bl         memcpy
+    vld4.8     {d4-d7}, [sp]!
+    pop        {r0-r2,lr}
+    b          2b
 .endm
 
 @ void *dst, const void *src, int count, uint global_alpha
@@ -35,11 +61,16 @@
     vdup.16    q11, r3
 .endif
     vdup.i16   q12, r12
+
 0:
-    pld        [r1, #64*2]
-    pld        [r0, #64*2]
+    cmp        r2, #8
+    blt        3f
+1:
     vld4.8     {d4-d7}, [r1]!
+    pld        [r1, #64*2]
     vld4.8     {d0-d3}, [r0]
+    pld        [r0, #64+32]
+2:
 .if \bgr2rgb
     vswp       d4, d6		@ BGR->RGB
 .endif
@@ -65,6 +96,35 @@
     bxeq       lr
     nop
     b          0b
+
+3:
+    @ unaligned ending nastiness :(
+    add        r3,  r0, #8*4
+    add        r12, r1, #8*4
+    lsr        r3,  #12
+    lsr        r12, #12
+    cmp        r3,  r0, lsr #12 @ are we crossing
+    cmpeq      r12, r1, lsr #12 @ the page boundary?
+    beq        1b               @ nope, overreading is safe
+
+    @ _wb_'s bad luck, do some slow stuff here
+    push       {r0-r2, lr}
+    vpush      {q11, q12}
+    sub        sp, #8*4*2
+    lsl        r2, #2
+    mov        r1, r0
+    mov        r0, sp
+    bl         memcpy
+    ldr        r2, [sp, #8*4*2 + 16*2 + 8] @ stacked r2
+    add        r0, sp, #8*4
+    ldr        r1, [sp, #8*4*2 + 16*2 + 4]
+    lsl        r2, #2
+    bl         memcpy
+    vld4.8     {d0-d3}, [sp]!
+    vld4.8     {d4-d7}, [sp]!
+    vpop       {q11, q12}
+    pop        {r0-r2, lr}
+    b          2b
 .endm
 
 
@@ -96,28 +156,119 @@ do_argb_finish:
     bx         lr
 
 
-.global neon_ARGBtoXRGB
-neon_ARGBtoXRGB:
+@ void *dst, const void *src, int count, uint global_alpha
+.macro do_argb_to_rgb565_alpha bgr2rgb global_alpha
+    mov        r12, #0xff
+.if \global_alpha
+    vdup.16    q11, r3
+.endif
+    vdup.i16   q12, r12
+0:
+    cmp        r2, #8
+    blt        3f
+1:
+    vld4.8     {d4-d7}, [r1]!
+    pld        [r1, #64*2]
+    vld2.8     {d1-d2}, [r0]
+    pld        [r0, #64+32]
+.if \bgr2rgb
+    vswp       d4, d6		@ BGR->RGB
+.endif
+.if !\global_alpha
+    vmovl.u8   q11, d7
+.endif
+    vshl.i8    d0, d1, #3
+    vshr.u8    d1, d1, #3
+    vsri.i8    d0, d0, #5       @ B
+    vsli.i8    d1, d2, #5
+    vsri.i8    d2, d2, #5       @ R
+    vsri.i8    d1, d1, #6       @ G
+    @ d = (((s-d)*a+255)>>8)+d
+    vsubl.u8   q8, d4, d0
+    vsubl.u8   q9, d5, d1
+    vsubl.u8   q10,d6, d2
+    vmul.s16   q8, q8, q11
+    vmul.s16   q9, q9, q11
+    vmul.s16   q10,q10,q11
+    vaddhn.i16 d4, q8, q12
+    vaddhn.i16 d5, q9, q12
+    vaddhn.i16 d6, q10,q12
+    vadd.i8    q2, q0
+    vadd.i8    d2, d6           @ rrrr rrrr
+    vshr.u8    d0, d5, #2
+    vshr.u8    d1, d4, #3       @ 000b bbbb
+    vsri.i8    d2, d5, #5       @ rrrr rggg
+    vsli.i8    d1, d0, #5       @ gggb bbbb
+    subs       r2, r2, #8
+    blt        do_rgb565_finish
+    vst2.8     {d1-d2}, [r0]!
+    bxeq       lr
+    nop
+    b          0b
+
+3:
+    @ unaligned ending nastiness :(
+    add        r3,  r0, #8*2
+    add        r12, r1, #8*4
+    lsr        r3,  #12
+    lsr        r12, #12
+    cmp        r3,  r0, lsr #12 @ are we crossing
+    cmpeq      r12, r1, lsr #12 @ the page boundary?
+    beq        1b               @ nope, overreading is safe
+
+    nop
+    bx         lr               @ abandon ship! (until someone complains)
+.endm
+
+
+do_rgb565_finish:
+    vzip.8     d1, d2
+    add        r2, r2, #8
+
+    vst1.16    d1[0], [r0]!
+    cmp        r2, #1
+    bxle       lr
+    vst1.16    d1[1], [r0]!
+    cmp        r2, #2
+    bxle       lr
+    vst1.16    d1[2], [r0]!
+    cmp        r2, #3
+    bxle       lr
+    vst1.16    d1[3], [r0]!
+    cmp        r2, #4
+    bxle       lr
+    vst1.16    d2[0], [r0]!
+    cmp        r2, #5
+    bxle       lr
+    vst1.16    d2[1], [r0]!
+    cmp        r2, #6
+    bxle       lr
+    vst1.16    d2[2], [r0]!
+    bx         lr
+
+
+func(neon_ARGBtoXRGB):
     do_argb 0
 
-.global neon_ABGRtoXRGB
-neon_ABGRtoXRGB:
+func(neon_ABGRtoXRGB):
     do_argb 1
 
-.global neon_ARGBtoXRGBalpha
-neon_ARGBtoXRGBalpha:
+func(neon_ARGBtoXRGBalpha):
     do_argb_alpha 0, 0
 
-.global neon_ABGRtoXRGBalpha
-neon_ABGRtoXRGBalpha:
+func(neon_ABGRtoXRGBalpha):
     do_argb_alpha 1, 0
 
-.global neon_ARGBtoXRGBalphaS
-neon_ARGBtoXRGBalphaS:
+func(neon_ARGBtoXRGBalphaS):
     do_argb_alpha 0, 1
 
-.global neon_ABGRtoXRGBalphaS
-neon_ABGRtoXRGBalphaS:
+func(neon_ABGRtoXRGBalphaS):
     do_argb_alpha 1, 1
 
+func(neon_ARGBtoRGB565alpha):
+    do_argb_to_rgb565_alpha 0, 0
+
+func(neon_ABGRtoRGB565alpha):
+    do_argb_to_rgb565_alpha 1, 0
+
 @ vim:filetype=armasm