MIPS: DSPr2: Added several nearest neighbor fast paths with a8 mask:

Performance numbers before/after on MIPS-74kc @ 1GHz:

lowlevel-blt-bench -n

Referent (before):
        over_8888_8_0565 =  L1:   9.62  L2:   8.85  M:  7.40 ( 39.27%)  HT:  5.67  VT:  5.61  R:  5.45  RT:  2.98 (  22Kops/s)
        over_0565_8_0565 =  L1:   7.90  L2:   7.49  M:  6.72 ( 26.75%)  HT:  5.24  VT:  5.20  R:  5.06  RT:  2.90 (  22Kops/s)

Optimized:
        over_8888_8_0565 =  L1:  18.51  L2:  16.82  M: 12.13 ( 64.43%)  HT: 10.06  VT:  9.88  R:  9.54  RT:  5.63 (  31Kops/s)
        over_0565_8_0565 =  L1:  14.82  L2:  13.94  M: 11.34 ( 45.20%)  HT:  9.45  VT:  9.35  R:  9.03  RT:  5.50 (  31Kops/s)

3 files changed, 214 insertions, 0 deletions

diff --git a/pixman/pixman-mips-dspr2-asm.S b/pixman/pixman-mips-dspr2-asm.S
index 791244a..ba22e62 100644
--- a/pixman/pixman-mips-dspr2-asm.S
+++ b/pixman/pixman-mips-dspr2-asm.S
@@ -2100,6 +2100,163 @@ LEAF_MIPS_DSPR2(pixman_composite_add_8888_8888_asm_mips)
 
 END(pixman_composite_add_8888_8888_asm_mips)
 
+LEAF_MIPS_DSPR2(pixman_scaled_nearest_scanline_8888_8_0565_OVER_asm_mips)
+/*
+ * a0     - dst  (r5g6b5)
+ * a1     - src  (a8r8g8b8)
+ * a2     - mask (a8)
+ * a3     - w
+ * 16(sp) - vx
+ * 20(sp) - unit_x
+ */
+    beqz     a3, 4f
+     nop
+
+    SAVE_REGS_ON_STACK 20, v0, v1, s0, s1, s2, s3, s4, s5
+    lw       v0, 36(sp) /* v0 = vx */
+    lw       v1, 40(sp) /* v1 = unit_x */
+    li       t6, 0x00ff00ff
+    li       t7, 0xf800f800
+    li       t8, 0x07e007e0
+    li       t9, 0x001F001F
+
+    addiu    t1, a3, -1
+    beqz     t1, 2f
+     nop
+1:
+    sra      t0, v0, 16 /* t0 = vx >> 16 */
+    sll      t0, t0, 2  /* t0 = t0 * 4      (a8r8g8b8) */
+    addu     t0, a1, t0
+    lw       t0, 0(t0)  /* t0 = source      (a8r8g8b8) */
+    addu     v0, v0, v1 /* v0 = vx + unit_x */
+    sra      t1, v0, 16 /* t1 = vx >> 16 */
+    sll      t1, t1, 2  /* t1 = t1 * 4      (a8r8g8b8) */
+    addu     t1, a1, t1
+    lw       t1, 0(t1)  /* t1 = source      (a8r8g8b8) */
+    addu     v0, v0, v1 /* v0 = vx + unit_x */
+    lbu      t2, 0(a2)  /* t2 = mask        (a8) */
+    lbu      t3, 1(a2)  /* t3 = mask        (a8) */
+    lhu      t4, 0(a0)  /* t4 = destination (r5g6b5) */
+    lhu      t5, 2(a0)  /* t5 = destination (r5g6b5) */
+    addiu    a2, a2, 2
+
+    CONVERT_2x0565_TO_2x8888 t4, t5, s0, s1, t8, t9, s2, s3, s4, s5
+    OVER_2x8888_2x8_2x8888   t0, t1, \
+                             t2, t3, \
+                             s0, s1, \
+                             t4, t5, \
+                             t6, s2, s3, s4, s5, t2, t3
+    CONVERT_2x8888_TO_2x0565 t4, t5, s0, s1, t7, t8, t9, s2, s3
+
+    sh       s0, 0(a0)
+    sh       s1, 2(a0)
+    addiu    a3, a3, -2
+    addiu    t1, a3, -1
+    bgtz     t1, 1b
+     addiu   a0, a0, 4
+2:
+    beqz     a3, 3f
+     nop
+    sra      t0, v0, 16 /* t0 = vx >> 16 */
+    sll      t0, t0, 2  /* t0 = t0 * 4      (a8r8g8b8) */
+    addu     t0, a1, t0
+    lw       t0, 0(t0)  /* t0 = source      (a8r8g8b8) */
+    lbu      t1, 0(a2)  /* t1 = mask        (a8) */
+    lhu      t2, 0(a0)  /* t2 = destination (r5g6b5) */
+
+    CONVERT_1x0565_TO_1x8888 t2, t3, t4, t5
+    OVER_8888_8_8888         t0, t1, t3, t2, t6, t4, t5, t7, t8
+    CONVERT_1x8888_TO_1x0565 t2, t3, t4, t5
+
+    sh       t3, 0(a0)
+3:
+    RESTORE_REGS_FROM_STACK 20, v0, v1, s0, s1, s2, s3, s4, s5
+4:
+    j        ra
+     nop
+
+END(pixman_scaled_nearest_scanline_8888_8_0565_OVER_asm_mips)
+
+LEAF_MIPS_DSPR2(pixman_scaled_nearest_scanline_0565_8_0565_OVER_asm_mips)
+/*
+ * a0     - dst  (r5g6b5)
+ * a1     - src  (r5g6b5)
+ * a2     - mask (a8)
+ * a3     - w
+ * 16(sp) - vx
+ * 20(sp) - unit_x
+ */
+
+    beqz     a3, 4f
+     nop
+    SAVE_REGS_ON_STACK 20, v0, v1, s0, s1, s2, s3, s4, s5
+    lw       v0, 36(sp) /* v0 = vx */
+    lw       v1, 40(sp) /* v1 = unit_x */
+    li       t4, 0xf800f800
+    li       t5, 0x07e007e0
+    li       t6, 0x001F001F
+    li       t7, 0x00ff00ff
+
+    addiu    t1, a3, -1
+    beqz     t1, 2f
+     nop
+1:
+    sra      t0, v0, 16 /* t0 = vx >> 16 */
+    sll      t0, t0, 1  /* t0 = t0 * 2      (r5g6b5) */
+    addu     t0, a1, t0
+    lhu      t0, 0(t0)  /* t0 = source      (r5g6b5) */
+    addu     v0, v0, v1 /* v0 = vx + unit_x */
+    sra      t1, v0, 16 /* t1 = vx >> 16 */
+    sll      t1, t1, 1  /* t1 = t1 * 2      (r5g6b5) */
+    addu     t1, a1, t1
+    lhu      t1, 0(t1)  /* t1 = source      (r5g6b5) */
+    addu     v0, v0, v1 /* v0 = vx + unit_x */
+    lbu      t2, 0(a2)  /* t2 = mask        (a8) */
+    lbu      t3, 1(a2)  /* t3 = mask        (a8) */
+    lhu      t8, 0(a0)  /* t8 = destination (r5g6b5) */
+    lhu      t9, 2(a0)  /* t9 = destination (r5g6b5) */
+    addiu    a2, a2, 2
+
+    CONVERT_2x0565_TO_2x8888 t0, t1, s0, s1, t5, t6, s2, s3, s4, s5
+    CONVERT_2x0565_TO_2x8888 t8, t9, s2, s3, t5, t6, s4, s5, t0, t1
+    OVER_2x8888_2x8_2x8888   s0, s1, \
+                             t2, t3, \
+                             s2, s3, \
+                             t0, t1, \
+                             t7, t8, t9, s4, s5, s0, s1
+    CONVERT_2x8888_TO_2x0565 t0, t1, s0, s1, t4, t5, t6, s2, s3
+
+    sh       s0, 0(a0)
+    sh       s1, 2(a0)
+    addiu    a3, a3, -2
+    addiu    t1, a3, -1
+    bgtz     t1, 1b
+     addiu   a0, a0, 4
+2:
+    beqz     a3, 3f
+     nop
+    sra      t0, v0, 16 /* t0 = vx >> 16 */
+    sll      t0, t0, 1  /* t0 = t0 * 2      (r5g6b5) */
+    addu     t0, a1, t0
+
+    lhu      t0, 0(t0)  /* t0 = source      (r5g6b5) */
+    lbu      t1, 0(a2)  /* t1 = mask        (a8) */
+    lhu      t2, 0(a0)  /* t2 = destination (r5g6b5) */
+
+    CONVERT_1x0565_TO_1x8888 t0, t3, t4, t5
+    CONVERT_1x0565_TO_1x8888 t2, t4, t5, t6
+    OVER_8888_8_8888         t3, t1, t4, t0, t7, t2, t5, t6, t8
+    CONVERT_1x8888_TO_1x0565 t0, t3, t4, t5
+
+    sh       t3, 0(a0)
+3:
+    RESTORE_REGS_FROM_STACK 20, v0, v1, s0, s1, s2, s3, s4, s5
+4:
+    j        ra
+     nop
+
+END(pixman_scaled_nearest_scanline_0565_8_0565_OVER_asm_mips)
+
 LEAF_MIPS_DSPR2(pixman_scaled_bilinear_scanline_8888_8888_SRC_asm_mips)
 /*
  * a0     - *dst
diff --git a/pixman/pixman-mips-dspr2.c b/pixman/pixman-mips-dspr2.c
index afe4ec6..11f1254 100644
--- a/pixman/pixman-mips-dspr2.c
+++ b/pixman/pixman-mips-dspr2.c
@@ -116,6 +116,11 @@ PIXMAN_MIPS_BIND_SCALED_BILINEAR_SRC_DST (SKIP_ZERO_SRC, 8888_8888, OVER,
 PIXMAN_MIPS_BIND_SCALED_BILINEAR_SRC_DST (SKIP_ZERO_SRC, 8888_8888, ADD,
                                           uint32_t, uint32_t)
 
+PIXMAN_MIPS_BIND_SCALED_NEAREST_SRC_A8_DST (SKIP_ZERO_SRC, 8888_8_0565,
+                                            OVER, uint32_t, uint16_t)
+PIXMAN_MIPS_BIND_SCALED_NEAREST_SRC_A8_DST (SKIP_ZERO_SRC, 0565_8_0565,
+                                            OVER, uint16_t, uint16_t)
+
 PIXMAN_MIPS_BIND_SCALED_BILINEAR_SRC_A8_DST (0, 8888_8_8888, SRC,
                                              uint32_t, uint32_t)
 PIXMAN_MIPS_BIND_SCALED_BILINEAR_SRC_A8_DST (0, 8888_8_0565, SRC,
@@ -320,6 +325,12 @@ static const pixman_fast_path_t mips_dspr2_fast_paths[] =
     PIXMAN_STD_FAST_PATH (ADD,  a8r8g8b8, null,     a8r8g8b8, mips_composite_add_8888_8888),
     PIXMAN_STD_FAST_PATH (ADD,  a8b8g8r8, null,     a8b8g8r8, mips_composite_add_8888_8888),
 
+    PIXMAN_MIPS_SIMPLE_NEAREST_A8_MASK_FAST_PATH (OVER, a8r8g8b8, r5g6b5, mips_8888_8_0565),
+    PIXMAN_MIPS_SIMPLE_NEAREST_A8_MASK_FAST_PATH (OVER, a8b8g8r8, b5g6r5, mips_8888_8_0565),
+
+    PIXMAN_MIPS_SIMPLE_NEAREST_A8_MASK_FAST_PATH (OVER, r5g6b5, r5g6b5, mips_0565_8_0565),
+    PIXMAN_MIPS_SIMPLE_NEAREST_A8_MASK_FAST_PATH (OVER, b5g6r5, b5g6r5, mips_0565_8_0565),
+
     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8, mips_8888_8888),
     SIMPLE_BILINEAR_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8, mips_8888_8888),
     SIMPLE_BILINEAR_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8, mips_8888_8888),
diff --git a/pixman/pixman-mips-dspr2.h b/pixman/pixman-mips-dspr2.h
index 186a501..4ac9ff9 100644
--- a/pixman/pixman-mips-dspr2.h
+++ b/pixman/pixman-mips-dspr2.h
@@ -246,6 +246,52 @@ mips_composite_##name (pixman_implementation_t *imp,                     \
     }                                                                    \
 }
 
+/*****************************************************************************/
+
+#define PIXMAN_MIPS_BIND_SCALED_NEAREST_SRC_A8_DST(flags, name, op,           \
+                                                  src_type, dst_type)         \
+void                                                                          \
+pixman_scaled_nearest_scanline_##name##_##op##_asm_mips (                     \
+                                                   dst_type *       dst,      \
+                                                   const src_type * src,      \
+                                                   const uint8_t *  mask,     \
+                                                   int32_t          w,        \
+                                                   pixman_fixed_t   vx,       \
+                                                   pixman_fixed_t   unit_x);  \
+                                                                              \
+static force_inline void                                                      \
+scaled_nearest_scanline_mips_##name##_##op (const uint8_t *  mask,            \
+                                            dst_type *       pd,              \
+                                            const src_type * ps,              \
+                                            int32_t          w,               \
+                                            pixman_fixed_t   vx,              \
+                                            pixman_fixed_t   unit_x,          \
+                                            pixman_fixed_t   max_vx,          \
+                                            pixman_bool_t    zero_src)        \
+{                                                                             \
+    if ((flags & SKIP_ZERO_SRC) && zero_src)                                  \
+        return;                                                               \
+    pixman_scaled_nearest_scanline_##name##_##op##_asm_mips (pd, ps,          \
+                                                             mask, w,         \
+                                                             vx, unit_x);     \
+}                                                                             \
+                                                                              \
+FAST_NEAREST_MAINLOOP_COMMON (mips_##name##_cover_##op,                       \
+                              scaled_nearest_scanline_mips_##name##_##op,     \
+                              src_type, uint8_t, dst_type, COVER, TRUE, FALSE)\
+FAST_NEAREST_MAINLOOP_COMMON (mips_##name##_none_##op,                        \
+                              scaled_nearest_scanline_mips_##name##_##op,     \
+                              src_type, uint8_t, dst_type, NONE, TRUE, FALSE) \
+FAST_NEAREST_MAINLOOP_COMMON (mips_##name##_pad_##op,                         \
+                              scaled_nearest_scanline_mips_##name##_##op,     \
+                              src_type, uint8_t, dst_type, PAD, TRUE, FALSE)
+
+/* Provide entries for the fast path table */
+#define PIXMAN_MIPS_SIMPLE_NEAREST_A8_MASK_FAST_PATH(op,s,d,func)             \
+    SIMPLE_NEAREST_A8_MASK_FAST_PATH_COVER (op,s,d,func),                     \
+    SIMPLE_NEAREST_A8_MASK_FAST_PATH_NONE (op,s,d,func),                      \
+    SIMPLE_NEAREST_A8_MASK_FAST_PATH_PAD (op,s,d,func)
+
 /****************************************************************************/
 
 #define PIXMAN_MIPS_BIND_SCALED_BILINEAR_SRC_DST(flags, name, op,            \