X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=frontend%2Farm_utils.s;h=d13192303c079d58f76397b0a674325f185fccf8;hp=d74c8b33a1d71e8f5d9d4e48abe4f543b97346e3;hb=1ab7621a76d1ef82fde77322c12d4076889bed01;hpb=1972732abfea710d3d7b61180971580d9e5081ce

diff --git a/frontend/arm_utils.s b/frontend/arm_utils.s
index d74c8b33..d1319230 100644
--- a/frontend/arm_utils.s
+++ b/frontend/arm_utils.s
@@ -1,7 +1,10 @@
 /*
  * (C) Gražvydas "notaz" Ignotas, 2010
  *
- * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * This work is licensed under the terms of any of these licenses
+ * (at your option):
+ *  - GNU GPL, version 2 or later.
+ *  - GNU LGPL, version 2.1 or later.
  * See the COPYING file in the top-level directory.
  */
 
@@ -10,39 +13,62 @@
 
 .global bgr555_to_rgb565
 bgr555_to_rgb565:
-    mov         r3, #0x03e0
+    mov         r3, #0x07c0
     vdup.16     q15, r3
-    mov         r2, r2, lsr #6
+    subs        r2, r2, #64
+    blt         btr16_end64
 0:
     vldmia      r1!, {q0-q3}
-    vshr.u16    q4, q0, #10
-    vshr.u16    q5, q1, #10
-    vshr.u16    q6, q2, #10
-    vshr.u16    q7, q3, #10
-    vshl.u16    q8, q0, #11
-    vshl.u16    q9, q1, #11
-    vshl.u16    q10, q2, #11
-    vshl.u16    q11, q3, #11
-    vand        q0, q0, q15
-    vand        q1, q1, q15
-    vand        q2, q2, q15
-    vand        q3, q3, q15
+    vshl.u16    q4, q0, #11
+    vshl.u16    q5, q1, #11
+    vshl.u16    q6, q2, #11
+    vshl.u16    q7, q3, #11
+    vsri.u16    q4, q0, #10
+    vsri.u16    q5, q1, #10
+    vsri.u16    q6, q2, #10
+    vsri.u16    q7, q3, #10
     vshl.u16    q0, q0, #1
     vshl.u16    q1, q1, #1
     vshl.u16    q2, q2, #1
     vshl.u16    q3, q3, #1
-    vorr        q0, q0, q4
-    vorr        q1, q1, q5
-    vorr        q2, q2, q6
-    vorr        q3, q3, q7
-    vorr        q0, q0, q8
-    vorr        q1, q1, q9
-    vorr        q2, q2, q10
-    vorr        q3, q3, q11
-    vstmia      r0!, {q0-q3}
-    subs        r2, r2, #1
-    bne         0b
+    vbit        q4, q0, q15
+    vbit        q5, q1, q15
+    vbit        q6, q2, q15
+    vbit        q7, q3, q15
+    vstmia      r0!, {q4-q7}
+    subs        r2, r2, #64
+    bge         0b
+
+btr16_end64:
+    adds        r2, r2, #64
+    bxeq        lr
+    subs        r2, r2, #16
+    blt         btr16_end16
+
+    @ handle the remainder (reasonably rare)
+0:
+    vld1.16     {q0}, [r1]!
+    vshl.u16    q1, q0, #11
+    vshl.u16    q2, q0, #1
+    vsri.u16    q1, q0, #10
+    vbit        q1, q2, q15
+    subs        r2, r2, #16
+    vst1.16     {q1}, [r0]!
+    bge         0b
+
+btr16_end16:
+    adds        r2, r2, #16
+    bxeq        lr
+    subs        r2, r2, #8
+    bxlt        lr
 
+    @ very rare
+    vld1.16     d0, [r1]!
+    vshl.u16    d1, d0, #11
+    vshl.u16    d2, d0, #1
+    vsri.u16    d1, d0, #10
+    vbit        d1, d2, d30
+    vst1.16     d1, [r0]!
     bx          lr
 
 
@@ -66,4 +92,35 @@ bgr888_to_rgb888:
     bx          lr
 
 
+.global bgr888_to_rgb565
+bgr888_to_rgb565:
+    @ 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:
+    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     d0, d3, #3      @ b
+    vshr.u8     d4, d7, #3
+    vzip.8      d0, d1          @ rb
+    vzip.8      d4, d5
+    vbit        q0, q8, q15
+    vbit        q2, q9, q15
+
+    vstmia      r0!, {d0,d1}
+    vstmia      r0!, {d4,d5}
+    subs        r2, r2, #1
+    bne         0b
+
+    bx          lr
+
+
 @ vim:filetype=armasm