fix part of [2017702] OpenGL bugs with alpha values of 1.0 in the sou…

…rce during blending into 8888 buffers

when ONE / ONE_MINUS_SRC_ALPHA blending mode was used, the code wasn't saturating the color component.
the reason was that this mode is used for premltiplied alpha blending, however, if used with a non
premultiplied source, the color component would wrap.

unfortunately, this costs 6 extra cycles per pixels, however... "correctness" prevails.

this should not impact the UI since it's using h/w acceleration most of the time it also doesn't
impact games which should be using h/w GL. This change will slow the emulator down a bit.

libpixelflinger/t32cb16blend.S

@@ -21,69 +21,100 @@
 
 	.global scanline_t32cb16blend_arm
 
-// uses r6, r7, lr
 
-.macro pixel,   DREG, SRC, FB, OFFSET
-
-    // SRC = AARRGGBB
+/*
+ * .macro pixel
+ *
+ * \DREG is a 32-bit register containing *two* original destination RGB565 
+ *       pixels, with the even one in the low-16 bits, and the odd one in the
+ *       high 16 bits.
+ *
+ * \SRC is a 32-bit 0xAABBGGRR pixel value, with pre-multiplied colors.
+ *
+ * \FB is a target register that will contain the blended pixel values.
+ *
+ * \ODD is either 0 or 1 and indicates if we're blending the lower or 
+ *      upper 16-bit pixels in DREG into FB
+ *
+ *
+ * clobbered: r6, r7, lr
+ *
+ */
+
+.macro pixel,   DREG, SRC, FB, ODD
+
+    // SRC = 0xAABBGGRR
     mov     r7, \SRC, lsr #24           // sA
     add     r7, r7, r7, lsr #7          // sA + (sA >> 7)
     rsb     r7, r7, #0x100              // sA = 0x100 - (sA+(sA>>7))
 
 1:
 
-.if \OFFSET
+.if \ODD
 
     // red
-    mov     lr, \DREG, lsr #(\OFFSET + 6 + 5)
+    mov     lr, \DREG, lsr #(16 + 11)
     smulbb  lr, r7, lr
     mov     r6, \SRC, lsr #3
     and     r6, r6, #0x1F
     add     lr, r6, lr, lsr #8
-    orr     \FB, lr, lsl #(\OFFSET + 11)
+    cmp     lr, #0x1F
+    orrhs   \FB, \FB, #(0x1F<<(16 + 11))
+    orrlo   \FB, \FB, lr, lsl #(16 + 11)
 
         // green
-        and     r6, \DREG, #(0x3F<<(\OFFSET + 5))
+        and     r6, \DREG, #(0x3F<<(16 + 5))
         smulbt  r6, r7, r6
         mov     lr, \SRC, lsr #(8+2)
         and     lr, lr, #0x3F
         add     r6, lr, r6, lsr #(5+8)
-        orr     \FB, \FB, r6, lsl #(\OFFSET + 5)
+        cmp     r6, #0x3F
+        orrhs   \FB, \FB, #(0x3F<<(16 + 5))
+        orrlo   \FB, \FB, r6, lsl #(16 + 5)
 
             // blue
-            and     lr, \DREG, #(0x1F << \OFFSET)
+            and     lr, \DREG, #(0x1F << 16)
             smulbt  lr, r7, lr
             mov     r6, \SRC, lsr #(8+8+3)
             and     r6, r6, #0x1F
             add     lr, r6, lr, lsr #8
-            orr     \FB, \FB, lr, lsl #\OFFSET
+            cmp     lr, #0x1F
+            orrhs   \FB, \FB, #(0x1F << 16)
+            orrlo   \FB, \FB, lr, lsl #16
 
 .else
 
     // red
-    mov     lr, \DREG, lsr #(6+5)
+    mov     lr, \DREG, lsr #11
     and     lr, lr, #0x1F
     smulbb  lr, r7, lr
     mov     r6, \SRC, lsr #3
     and     r6, r6, #0x1F
     add     lr, r6, lr, lsr #8
-    mov     \FB, lr, lsl #11
+    cmp     lr, #0x1F
+    movhs   \FB, #(0x1F<<11)
+    movlo   \FB, lr, lsl #11
+
 
         // green
         and     r6, \DREG, #(0x3F<<5)
         smulbb  r6, r7, r6
         mov     lr, \SRC, lsr #(8+2)
         and     lr, lr, #0x3F
         add     r6, lr, r6, lsr #(5+8)
-        orr     \FB, \FB, r6, lsl #5
+        cmp     r6, #0x3F
+        orrhs   \FB, \FB, #(0x3F<<5)
+        orrlo   \FB, \FB, r6, lsl #5
 
             // blue
             and     lr, \DREG, #0x1F
             smulbb  lr, r7, lr
             mov     r6, \SRC, lsr #(8+8+3)
             and     r6, r6, #0x1F
             add     lr, r6, lr, lsr #8
-            orr     \FB, \FB, lr
+            cmp     lr, #0x1F
+            orrhs   \FB, \FB, #0x1F
+            orrlo   \FB, \FB, lr
 
 .endif
 
@@ -128,7 +159,7 @@ aligned:
     subs    r2, r2, #2
     blo     9f
 
-    // The main loop is unrolled twice and process 4 pixels
+    // The main loop is unrolled twice and processes 4 pixels
 8:  ldmia   r1!, {r4, r5}
     // stream the source
     pld     [r1, #32]
@@ -142,7 +173,7 @@ aligned:
     // stream the destination
     pld     [r0, #32]
     pixel   r3, r4, r12, 0
-    pixel   r3, r5, r12, 16
+    pixel   r3, r5, r12, 1
     // effectively, we're getting write-combining by virtue of the
     // cpu's write-back cache.
     str     r12, [r0, #-4]