diff options
Diffstat (limited to 'pixman/pixman/pixman-mips-dspr2-asm.S')
| -rw-r--r-- | pixman/pixman/pixman-mips-dspr2-asm.S | 224 |
1 files changed, 224 insertions, 0 deletions
diff --git a/pixman/pixman/pixman-mips-dspr2-asm.S b/pixman/pixman/pixman-mips-dspr2-asm.S index 6a0fc1803..68ad33f7c 100644 --- a/pixman/pixman/pixman-mips-dspr2-asm.S +++ b/pixman/pixman/pixman-mips-dspr2-asm.S @@ -527,3 +527,227 @@ LEAF_MIPS_DSPR2(pixman_composite_over_n_8888_0565_ca_asm_mips) nop END(pixman_composite_over_n_8888_0565_ca_asm_mips) + +LEAF_MIPS_DSPR2(pixman_composite_over_n_8_8888_asm_mips) +/* + * a0 - dst (a8r8g8b8) + * a1 - src (32bit constant) + * a2 - mask (a8) + * a3 - w + */ + + SAVE_REGS_ON_STACK 4, s0, s1, s2, s3, s4 + beqz a3, 4f + nop + li t4, 0x00ff00ff + li t5, 0xff + addiu t0, a3, -1 + beqz t0, 3f /* last pixel */ + srl t6, a1, 24 /* t6 = srca */ + not s4, a1 + beq t5, t6, 2f /* if (srca == 0xff) */ + srl s4, s4, 24 +1: + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + lbu t1, 1(a2) /* t1 = mask */ + or t2, t0, t1 + beqz t2, 111f /* if (t0 == 0) && (t1 == 0) */ + addiu a2, a2, 2 + and t3, t0, t1 + + lw t2, 0(a0) /* t2 = dst */ + beq t3, t5, 11f /* if (t0 == 0xff) && (t1 == 0xff) */ + lw t3, 4(a0) /* t3 = dst */ + + MIPS_2xUN8x4_MUL_2xUN8 a1, a1, t0, t1, s0, s1, t4, t6, t7, t8, t9, s2, s3 + not s2, s0 + not s3, s1 + srl s2, s2, 24 + srl s3, s3, 24 + MIPS_2xUN8x4_MUL_2xUN8 t2, t3, s2, s3, t2, t3, t4, t0, t1, t6, t7, t8, t9 + addu_s.qb s2, t2, s0 + addu_s.qb s3, t3, s1 + sw s2, 0(a0) + b 111f + sw s3, 4(a0) +11: + MIPS_2xUN8x4_MUL_2xUN8 t2, t3, s4, s4, t2, t3, t4, t0, t1, t6, t7, t8, t9 + addu_s.qb s2, t2, a1 + addu_s.qb s3, t3, a1 + sw s2, 0(a0) + sw s3, 4(a0) + +111: + addiu a3, a3, -2 + addiu t0, a3, -1 + bgtz t0, 1b + addiu a0, a0, 8 + b 3f + nop +2: + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + lbu t1, 1(a2) /* t1 = mask */ + or t2, t0, t1 + beqz t2, 222f /* if (t0 == 0) && (t1 == 0) */ + addiu a2, a2, 2 + and t3, t0, t1 + beq t3, t5, 22f /* if (t0 == 0xff) && (t1 == 0xff) */ + nop + lw t2, 0(a0) /* t2 = dst */ + lw t3, 4(a0) /* t3 = dst */ + + OVER_2x8888_2x8_2x8888 a1, a1, t0, t1, t2, t3, \ + t6, t7, t4, t8, t9, s0, s1, s2, s3 + sw t6, 0(a0) + b 222f + sw t7, 4(a0) +22: + sw a1, 0(a0) + sw a1, 4(a0) +222: + addiu a3, a3, -2 + addiu t0, a3, -1 + bgtz t0, 2b + addiu a0, a0, 8 +3: + blez a3, 4f + nop + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + beqz t0, 4f /* if (t0 == 0) */ + addiu a2, a2, 1 + move t3, a1 + beq t0, t5, 31f /* if (t0 == 0xff) */ + lw t1, 0(a0) /* t1 = dst */ + + MIPS_UN8x4_MUL_UN8 a1, t0, t3, t4, t6, t7, t8 +31: + not t2, t3 + srl t2, t2, 24 + MIPS_UN8x4_MUL_UN8 t1, t2, t1, t4, t6, t7, t8 + addu_s.qb t2, t1, t3 + sw t2, 0(a0) +4: + RESTORE_REGS_FROM_STACK 4, s0, s1, s2, s3, s4 + j ra + nop + +END(pixman_composite_over_n_8_8888_asm_mips) + +LEAF_MIPS_DSPR2(pixman_composite_over_n_8_0565_asm_mips) +/* + * a0 - dst (r5g6b5) + * a1 - src (32bit constant) + * a2 - mask (a8) + * a3 - w + */ + SAVE_REGS_ON_STACK 24, v0, s0, s1, s2, s3, s4, s5, s6, s7, s8 + beqz a3, 4f + nop + li t4, 0x00ff00ff + li t5, 0xff + li t6, 0xf800f800 + li t7, 0x07e007e0 + li t8, 0x001F001F + addiu t1, a3, -1 + beqz t1, 3f /* last pixel */ + srl t0, a1, 24 /* t0 = srca */ + not v0, a1 + beq t0, t5, 2f /* if (srca == 0xff) */ + srl v0, v0, 24 +1: + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + lbu t1, 1(a2) /* t1 = mask */ + or t2, t0, t1 + beqz t2, 111f /* if (t0 == 0) && (t1 == 0) */ + addiu a2, a2, 2 + lhu t2, 0(a0) /* t2 = dst */ + lhu t3, 2(a0) /* t3 = dst */ + CONVERT_2x0565_TO_2x8888 t2, t3, s0, s1, t7, t8, t9, s2, s3, s4 + and t9, t0, t1 + beq t9, t5, 11f /* if (t0 == 0xff) && (t1 == 0xff) */ + nop + + MIPS_2xUN8x4_MUL_2xUN8 a1, a1, t0, t1, s2, s3, t4, t9, s4, s5, s6, s7, s8 + not s4, s2 + not s5, s3 + srl s4, s4, 24 + srl s5, s5, 24 + MIPS_2xUN8x4_MUL_2xUN8 s0, s1, s4, s5, s0, s1, t4, t9, t0, t1, s6, s7, s8 + addu_s.qb s4, s2, s0 + addu_s.qb s5, s3, s1 + CONVERT_2x8888_TO_2x0565 s4, s5, t2, t3, t6, t7, t8, s0, s1 + sh t2, 0(a0) + b 111f + sh t3, 2(a0) +11: + MIPS_2xUN8x4_MUL_2xUN8 s0, s1, v0, v0, s0, s1, t4, t9, t0, t1, s6, s7, s8 + addu_s.qb s4, a1, s0 + addu_s.qb s5, a1, s1 + CONVERT_2x8888_TO_2x0565 s4, s5, t2, t3, t6, t7, t8, s0, s1 + sh t2, 0(a0) + sh t3, 2(a0) +111: + addiu a3, a3, -2 + addiu t0, a3, -1 + bgtz t0, 1b + addiu a0, a0, 4 + b 3f + nop +2: + CONVERT_1x8888_TO_1x0565 a1, s0, s1, s2 +21: + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + lbu t1, 1(a2) /* t1 = mask */ + or t2, t0, t1 + beqz t2, 222f /* if (t0 == 0) && (t1 == 0) */ + addiu a2, a2, 2 + and t9, t0, t1 + move s2, s0 + beq t9, t5, 22f /* if (t0 == 0xff) && (t2 == 0xff) */ + move s3, s0 + lhu t2, 0(a0) /* t2 = dst */ + lhu t3, 2(a0) /* t3 = dst */ + + CONVERT_2x0565_TO_2x8888 t2, t3, s2, s3, t7, t8, s4, s5, s6, s7 + OVER_2x8888_2x8_2x8888 a1, a1, t0, t1, s2, s3, \ + t2, t3, t4, t9, s4, s5, s6, s7, s8 + CONVERT_2x8888_TO_2x0565 t2, t3, s2, s3, t6, t7, t8, s4, s5 +22: + sh s2, 0(a0) + sh s3, 2(a0) +222: + addiu a3, a3, -2 + addiu t0, a3, -1 + bgtz t0, 21b + addiu a0, a0, 4 +3: + blez a3, 4f + nop + /* a1 = src */ + lbu t0, 0(a2) /* t0 = mask */ + beqz t0, 4f /* if (t0 == 0) */ + nop + lhu t1, 0(a0) /* t1 = dst */ + CONVERT_1x0565_TO_1x8888 t1, t2, t3, t7 + beq t0, t5, 31f /* if (t0 == 0xff) */ + move t3, a1 + + MIPS_UN8x4_MUL_UN8 a1, t0, t3, t4, t7, t8, t9 +31: + not t6, t3 + srl t6, t6, 24 + MIPS_UN8x4_MUL_UN8 t2, t6, t2, t4, t7, t8, t9 + addu_s.qb t1, t2, t3 + CONVERT_1x8888_TO_1x0565 t1, t2, t3, t7 + sh t2, 0(a0) +4: + RESTORE_REGS_FROM_STACK 24, v0, s0, s1, s2, s3, s4, s5, s6, s7, s8 + j ra + nop + +END(pixman_composite_over_n_8_0565_asm_mips) |