X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=frontend%2Fcspace_neon.s;h=b458f06a760fea5355d98e3b9b76fca3dddf8263;hp=d13192303c079d58f76397b0a674325f185fccf8;hb=a48745850385e49bf1ca9f8c4e98d925032a8a06;hpb=a80ae4a0353fce94df700ec84222d3c56c3d813a

diff --git a/frontend/cspace_neon.s b/frontend/cspace_neon.s
index d1319230..b458f06a 100644
--- a/frontend/cspace_neon.s
+++ b/frontend/cspace_neon.s
@@ -13,11 +13,13 @@
 
 .global bgr555_to_rgb565
 bgr555_to_rgb565:
+    pld         [r1]
     mov         r3, #0x07c0
     vdup.16     q15, r3
     subs        r2, r2, #64
     blt         btr16_end64
 0:
+    pld         [r1, #64*2]
     vldmia      r1!, {q0-q3}
     vshl.u16    q4, q0, #11
     vshl.u16    q5, q1, #11
@@ -74,12 +76,14 @@ btr16_end16:
 
 .global bgr888_to_rgb888
 bgr888_to_rgb888:
+    pld         [r1]
     @ r2 /= 48
     mov         r2, r2, lsr #4
     movw        r3, #0x5556
     movt        r3, #0x5555
     umull       r12,r2, r3, r2
 0:
+    pld         [r1, #48*3]
     vld3.8      {d0-d2}, [r1, :64]!
     vld3.8      {d3-d5}, [r1, :64]!
     vswp        d0, d2
@@ -94,6 +98,7 @@ bgr888_to_rgb888:
 
 .global bgr888_to_rgb565
 bgr888_to_rgb565:
+    pld         [r1]
     @ r2 /= 48
     mov         r2, r2, lsr #4
     movw        r3, #0x5556
@@ -103,6 +108,7 @@ bgr888_to_rgb565:
     mov         r3, #0x07e0
     vdup.16     q15, r3
 0:
+    pld         [r1, #48*3]
     vld3.8      {d1-d3}, [r1, :64]!
     vld3.8      {d5-d7}, [r1, :64]!
 
@@ -123,4 +129,37 @@ bgr888_to_rgb565:
     bx          lr
 
 
+.global rgb888_to_rgb565
+rgb888_to_rgb565:
+    pld         [r1]
+    @ r2 /= 48
+    mov         r2, r2, lsr #4
+    movw        r3, #0x5556
+    movt        r3, #0x5555
+    umull       r12,r2, r3, r2
+
+    mov         r3, #0x07e0
+    vdup.16     q15, r3
+0:
+    pld         [r1, #48*3]
+    vld3.8      {d1-d3}, [r1, :64]!
+    vld3.8      {d5-d7}, [r1, :64]!
+
+    vshll.u8    q8, d2, #3      @ g
+    vshll.u8    q9, d6, #3
+    vshr.u8     d2, d1, #3      @ b
+    vshr.u8     d6, d5, #3
+    vzip.8      d2, d3          @ rb
+    vzip.8      d6, d7
+    vbit        q1, q8, q15
+    vbit        q3, q9, q15
+
+    vstmia      r0!, {d2,d3}
+    vstmia      r0!, {d6,d7}
+    subs        r2, r2, #1
+    bne         0b
+
+    bx          lr
+
+
 @ vim:filetype=armasm