aboutsummaryrefslogtreecommitdiff
path: root/pixman/pixman/pixman-mips-dspr2-asm.S
diff options
context:
space:
mode:
Diffstat (limited to 'pixman/pixman/pixman-mips-dspr2-asm.S')
-rw-r--r--pixman/pixman/pixman-mips-dspr2-asm.S424
1 files changed, 424 insertions, 0 deletions
diff --git a/pixman/pixman/pixman-mips-dspr2-asm.S b/pixman/pixman/pixman-mips-dspr2-asm.S
index b5cae1690..ba22e6254 100644
--- a/pixman/pixman/pixman-mips-dspr2-asm.S
+++ b/pixman/pixman/pixman-mips-dspr2-asm.S
@@ -310,6 +310,139 @@ LEAF_MIPS_DSPR2(pixman_composite_src_x888_8888_asm_mips)
END(pixman_composite_src_x888_8888_asm_mips)
+LEAF_MIPS_DSPR2(pixman_composite_src_n_8_8888_asm_mips)
+/*
+ * a0 - dst (a8r8g8b8)
+ * a1 - src (32bit constant)
+ * a2 - mask (a8)
+ * a3 - w
+ */
+
+
+ SAVE_REGS_ON_STACK 0, v0
+ li v0, 0x00ff00ff
+
+ beqz a3, 3f
+ nop
+ addiu t1, a3, -1
+ beqz t1, 2f
+ nop
+
+1:
+ /* a1 = source (32bit constant) */
+ lbu t0, 0(a2) /* t2 = mask (a8) */
+ lbu t1, 1(a2) /* t3 = mask (a8) */
+ addiu a2, a2, 2
+
+ MIPS_2xUN8x4_MUL_2xUN8 a1, a1, t0, t1, t2, t3, v0, t4, t5, t6, t7, t8, t9
+
+ sw t2, 0(a0)
+ sw t3, 4(a0)
+ addiu a3, a3, -2
+ addiu t2, a3, -1
+ bgtz t2, 1b
+ addiu a0, a0, 8
+
+ beqz a3, 3f
+ nop
+
+2:
+ lbu t0, 0(a2)
+ addiu a2, a2, 1
+
+ MIPS_UN8x4_MUL_UN8 a1, t0, t1, v0, t3, t4, t5
+
+ sw t1, 0(a0)
+ addiu a3, a3, -1
+ addiu a0, a0, 4
+
+3:
+ RESTORE_REGS_FROM_STACK 0, v0
+ j ra
+ nop
+
+END(pixman_composite_src_n_8_8888_asm_mips)
+
+LEAF_MIPS_DSPR2(pixman_composite_src_n_8_8_asm_mips)
+/*
+ * a0 - dst (a8)
+ * a1 - src (32bit constant)
+ * a2 - mask (a8)
+ * a3 - w
+ */
+
+ li t9, 0x00ff00ff
+ beqz a3, 3f
+ nop
+ srl t7, a3, 2 /* t7 = how many multiples of 4 dst pixels */
+ beqz t7, 1f /* branch if less than 4 src pixels */
+ nop
+
+ srl t8, a1, 24
+ replv.ph t8, t8
+
+0:
+ beqz t7, 1f
+ addiu t7, t7, -1
+ lbu t0, 0(a2)
+ lbu t1, 1(a2)
+ lbu t2, 2(a2)
+ lbu t3, 3(a2)
+
+ addiu a2, a2, 4
+
+ precr_sra.ph.w t1, t0, 0
+ precr_sra.ph.w t3, t2, 0
+ precr.qb.ph t0, t3, t1
+
+ muleu_s.ph.qbl t2, t0, t8
+ muleu_s.ph.qbr t3, t0, t8
+ shra_r.ph t4, t2, 8
+ shra_r.ph t5, t3, 8
+ and t4, t4, t9
+ and t5, t5, t9
+ addq.ph t2, t2, t4
+ addq.ph t3, t3, t5
+ shra_r.ph t2, t2, 8
+ shra_r.ph t3, t3, 8
+ precr.qb.ph t2, t2, t3
+
+ sb t2, 0(a0)
+ srl t2, t2, 8
+ sb t2, 1(a0)
+ srl t2, t2, 8
+ sb t2, 2(a0)
+ srl t2, t2, 8
+ sb t2, 3(a0)
+ addiu a3, a3, -4
+ b 0b
+ addiu a0, a0, 4
+
+1:
+ beqz a3, 3f
+ nop
+ srl t8, a1, 24
+2:
+ lbu t0, 0(a2)
+ addiu a2, a2, 1
+
+ mul t2, t0, t8
+ shra_r.ph t3, t2, 8
+ andi t3, t3, 0x00ff
+ addq.ph t2, t2, t3
+ shra_r.ph t2, t2, 8
+
+ sb t2, 0(a0)
+ addiu a3, a3, -1
+ bnez a3, 2b
+ addiu a0, a0, 1
+
+3:
+ j ra
+ nop
+
+END(pixman_composite_src_n_8_8_asm_mips)
+
LEAF_MIPS_DSPR2(pixman_composite_over_n_8888_8888_ca_asm_mips)
/*
* a0 - dst (a8r8g8b8)
@@ -1209,6 +1342,140 @@ LEAF_MIPS_DSPR2(pixman_composite_over_8888_8888_asm_mips)
END(pixman_composite_over_8888_8888_asm_mips)
+LEAF_MIPS_DSPR2(pixman_composite_over_n_0565_asm_mips)
+/*
+ * a0 - dst (r5g6b5)
+ * a1 - src (32bit constant)
+ * a2 - w
+ */
+
+ beqz a2, 5f
+ nop
+
+ not t0, a1
+ srl t0, t0, 24
+ bgtz t0, 1f
+ nop
+ CONVERT_1x8888_TO_1x0565 a1, t1, t2, t3
+0:
+ sh t1, 0(a0)
+ addiu a2, a2, -1
+ bgtz a2, 0b
+ addiu a0, a0, 2
+ j ra
+ nop
+
+1:
+ SAVE_REGS_ON_STACK 0, s0, s1, s2
+ li t4, 0x00ff00ff
+ li t5, 0xf800f800
+ li t6, 0x07e007e0
+ li t7, 0x001F001F
+ addiu t1, a2, -1
+ beqz t1, 3f
+ nop
+2:
+ lhu t1, 0(a0) /* t1 = destination (r5g6b5) */
+ lhu t2, 2(a0) /* t2 = destination (r5g6b5) */
+
+ CONVERT_2x0565_TO_2x8888 t1, t2, t3, t8, t6, t7, t9, s0, s1, s2
+ MIPS_2xUN8x4_MUL_2xUN8 t3, t8, t0, t0, t1, t2, t4, t9, s0, s1, s2, t3, t8
+ addu_s.qb t1, t1, a1
+ addu_s.qb t2, t2, a1
+ CONVERT_2x8888_TO_2x0565 t1, t2, t3, t8, t5, t6, t7, s0, s1
+
+ sh t3, 0(a0)
+ sh t8, 2(a0)
+
+ addiu a2, a2, -2
+ addiu t1, a2, -1
+ bgtz t1, 2b
+ addiu a0, a0, 4
+3:
+ beqz a2, 4f
+ nop
+
+ lhu t1, 0(a0) /* t1 = destination (r5g6b5) */
+
+ CONVERT_1x0565_TO_1x8888 t1, t2, s0, s1
+ MIPS_UN8x4_MUL_UN8 t2, t0, t1, t4, s0, s1, s2
+ addu_s.qb t1, t1, a1
+ CONVERT_1x8888_TO_1x0565 t1, t2, s0, s1
+
+ sh t2, 0(a0)
+
+4:
+ RESTORE_REGS_FROM_STACK 0, s0, s1, s2
+5:
+ j ra
+ nop
+
+END(pixman_composite_over_n_0565_asm_mips)
+
+LEAF_MIPS_DSPR2(pixman_composite_over_n_8888_asm_mips)
+/*
+ * a0 - dst (a8r8g8b8)
+ * a1 - src (32bit constant)
+ * a2 - w
+ */
+
+ beqz a2, 5f
+ nop
+
+ not t0, a1
+ srl t0, t0, 24
+ bgtz t0, 1f
+ nop
+0:
+ sw a1, 0(a0)
+ addiu a2, a2, -1
+ bgtz a2, 0b
+ addiu a0, a0, 4
+ j ra
+ nop
+
+1:
+ SAVE_REGS_ON_STACK 0, s0, s1, s2
+ li t4, 0x00ff00ff
+ addiu t1, a2, -1
+ beqz t1, 3f
+ nop
+2:
+ lw t2, 0(a0) /* t2 = destination (a8r8g8b8) */
+ lw t3, 4(a0) /* t3 = destination (a8r8g8b8) */
+
+ MIPS_2xUN8x4_MUL_2xUN8 t2, t3, t0, t0, t7, t8, t4, t9, s0, s1, s2, t2, t3
+
+ addu_s.qb t7, t7, a1
+ addu_s.qb t8, t8, a1
+
+ sw t7, 0(a0)
+ sw t8, 4(a0)
+
+ addiu a2, a2, -2
+ addiu t1, a2, -1
+ bgtz t1, 2b
+ addiu a0, a0, 8
+3:
+ beqz a2, 4f
+ nop
+
+ lw t1, 0(a0) /* t1 = destination (a8r8g8b8) */
+
+ MIPS_UN8x4_MUL_UN8 t1, t0, t3, t4, t5, t6, t7
+
+ addu_s.qb t3, t3, a1
+
+ sw t3, 0(a0)
+
+4:
+ RESTORE_REGS_FROM_STACK 0, s0, s1, s2
+5:
+ j ra
+ nop
+
+END(pixman_composite_over_n_8888_asm_mips)
+
LEAF_MIPS_DSPR2(pixman_composite_add_8_8_8_asm_mips)
/*
* a0 - dst (a8)
@@ -1833,6 +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
generated by cgit v1.2.3 (git 2.25.1) at 2024-08-31 10:24:44 +0000