ref: 46980237595c3065f15106b0c3483cdd57fd3153
parent: 4a2ea99d3dc6b6bbb43e4680392584dcf4c8882f
author: Martin Storsjö <martin@martin.st>
date: Fri Jun 21 19:12:12 EDT 2019
arm: mc: Move the blend functions up above put/prep This keeps the put/prep functions close to the 8tap/bilin functions that use them.
--- a/src/arm/32/mc.S
+++ b/src/arm/32/mc.S
@@ -215,230 +215,6 @@
bidir_fn mask
-// This has got the same signature as the put_8tap functions,
-// assumes that the caller has loaded the h argument into r5,
-// and assumes that r8 is set to (clz(w)-24).
-function put_neon
- adr r9, L(put_tbl)
- ldr r8, [r9, r8, lsl #2]
- add r9, r9, r8
- bx r9
-
- .align 2
-L(put_tbl):
- .word 1280f - L(put_tbl) + CONFIG_THUMB
- .word 640f - L(put_tbl) + CONFIG_THUMB
- .word 32f - L(put_tbl) + CONFIG_THUMB
- .word 160f - L(put_tbl) + CONFIG_THUMB
- .word 8f - L(put_tbl) + CONFIG_THUMB
- .word 4f - L(put_tbl) + CONFIG_THUMB
- .word 2f - L(put_tbl) + CONFIG_THUMB
-
-2:
- vld1.16 {d0[]}, [r2], r3- vld1.16 {d1[]}, [r2], r3- subs r5, r5, #2
- vst1.16 {d0[0]}, [r0, :16], r1- vst1.16 {d1[0]}, [r0, :16], r1- bgt 2b
- pop {r4-r11,pc}-4:
- vld1.32 {d0[]}, [r2], r3- vld1.32 {d1[]}, [r2], r3- subs r5, r5, #2
- vst1.32 {d0[0]}, [r0, :32], r1- vst1.32 {d1[0]}, [r0, :32], r1- bgt 4b
- pop {r4-r11,pc}-8:
- vld1.8 {d0}, [r2], r3- vld1.8 {d1}, [r2], r3- subs r5, r5, #2
- vst1.8 {d0}, [r0, :64], r1- vst1.8 {d1}, [r0, :64], r1- bgt 8b
- pop {r4-r11,pc}-160:
- add r8, r0, r1
- lsl r1, r1, #1
- add r9, r2, r3
- lsl r3, r3, #1
-16:
- vld1.8 {q0}, [r2], r3- vld1.8 {q1}, [r9], r3- subs r5, r5, #2
- vst1.8 {q0}, [r0, :128], r1- vst1.8 {q1}, [r8, :128], r1- bgt 16b
- pop {r4-r11,pc}-32:
- vld1.8 {q0, q1}, [r2], r3- subs r5, r5, #1
- vst1.8 {q0, q1}, [r0, :128], r1- bgt 32b
- pop {r4-r11,pc}-640:
- sub r1, r1, #32
- sub r3, r3, #32
-64:
- vld1.8 {q0, q1}, [r2]!- vst1.8 {q0, q1}, [r0, :128]!- vld1.8 {q2, q3}, [r2], r3- subs r5, r5, #1
- vst1.8 {q2, q3}, [r0, :128], r1- bgt 64b
- pop {r4-r11,pc}-1280:
- sub r1, r1, #96
- sub r3, r3, #96
-128:
- vld1.8 {q8, q9}, [r2]!- vst1.8 {q8, q9}, [r0, :128]!- vld1.8 {q10, q11}, [r2]!- vst1.8 {q10, q11}, [r0, :128]!- vld1.8 {q12, q13}, [r2]!- vst1.8 {q12, q13}, [r0, :128]!- vld1.8 {q14, q15}, [r2], r3- subs r5, r5, #1
- vst1.8 {q14, q15}, [r0, :128], r1- bgt 128b
- pop {r4-r11,pc}-endfunc
-
-
-// This has got the same signature as the put_8tap functions,
-// assumes that the caller has loaded the h argument into r4,
-// and assumes that r8 is set to (clz(w)-24), and r7 to w*2.
-function prep_neon
- adr r9, L(prep_tbl)
- ldr r8, [r9, r8, lsl #2]
- add r9, r9, r8
- bx r9
-
- .align 2
-L(prep_tbl):
- .word 1280f - L(prep_tbl) + CONFIG_THUMB
- .word 640f - L(prep_tbl) + CONFIG_THUMB
- .word 320f - L(prep_tbl) + CONFIG_THUMB
- .word 160f - L(prep_tbl) + CONFIG_THUMB
- .word 8f - L(prep_tbl) + CONFIG_THUMB
- .word 4f - L(prep_tbl) + CONFIG_THUMB
-
-4:
- vld1.32 {d0[]}, [r1], r2- vld1.32 {d2[]}, [r1], r2- subs r4, r4, #2
- vshll.u8 q0, d0, #4
- vshll.u8 q1, d2, #4
- vst1.16 {d1, d2}, [r0, :64]!- bgt 4b
- pop {r4-r11,pc}-8:
- vld1.8 {d0}, [r1], r2- vld1.8 {d2}, [r1], r2- subs r4, r4, #2
- vshll.u8 q0, d0, #4
- vshll.u8 q1, d2, #4
- vst1.16 {q0, q1}, [r0, :128]!- bgt 8b
- pop {r4-r11,pc}-160:
- add r9, r1, r2
- lsl r2, r2, #1
- add r8, r0, r7
- lsl r7, r7, #1
-16:
- vld1.8 {q2}, [r1], r2- vld1.8 {q3}, [r9], r2- subs r4, r4, #2
- vshll.u8 q0, d4, #4
- vshll.u8 q1, d5, #4
- vshll.u8 q2, d6, #4
- vshll.u8 q3, d7, #4
- vst1.16 {q0, q1}, [r0, :128], r7- vst1.16 {q2, q3}, [r8, :128], r7- bgt 16b
- pop {r4-r11,pc}-320:
- add r8, r0, r3
-32:
- vld1.8 {q0, q1}, [r1], r2- subs r4, r4, #2
- vshll.u8 q8, d0, #4
- vshll.u8 q9, d1, #4
- vld1.8 {q2, q3}, [r1], r2- vshll.u8 q10, d2, #4
- vshll.u8 q11, d3, #4
- vshll.u8 q12, d4, #4
- vst1.16 {q8, q9}, [r0, :128], r7- vshll.u8 q13, d5, #4
- vst1.16 {q10, q11}, [r8, :128], r7- vshll.u8 q14, d6, #4
- vst1.16 {q12, q13}, [r0, :128], r7- vshll.u8 q15, d7, #4
- vst1.16 {q14, q15}, [r8, :128], r7- bgt 32b
- pop {r4-r11,pc}-640:
- sub r2, r2, #32
- add r8, r0, #32
- mov r6, #64
-64:
- vld1.8 {q0, q1}, [r1]!- subs r4, r4, #1
- vshll.u8 q8, d0, #4
- vshll.u8 q9, d1, #4
- vld1.8 {q2, q3}, [r1], r2- vshll.u8 q10, d2, #4
- vshll.u8 q11, d3, #4
- vshll.u8 q12, d4, #4
- vst1.16 {q8, q9}, [r0, :128], r6- vshll.u8 q13, d5, #4
- vshll.u8 q14, d6, #4
- vst1.16 {q10, q11}, [r8, :128], r6- vshll.u8 q15, d7, #4
- vst1.16 {q12, q13}, [r0, :128], r6- vst1.16 {q14, q15}, [r8, :128], r6- bgt 64b
- pop {r4-r11,pc}-1280:
- sub r2, r2, #96
- add r8, r0, #32
- mov r6, #64
-128:
- vld1.8 {q0, q1}, [r1]!- vld1.8 {q2, q3}, [r1]!- vshll.u8 q10, d0, #4
- vshll.u8 q11, d1, #4
- vshll.u8 q12, d2, #4
- vshll.u8 q13, d3, #4
- vshll.u8 q14, d4, #4
- vshll.u8 q15, d5, #4
- vld1.8 {q8, q9}, [r1]!- vst1.16 {q10, q11}, [r0, :128], r6- vst1.16 {q12, q13}, [r8, :128], r6- vshll.u8 q0, d6, #4
- vshll.u8 q1, d7, #4
- vshll.u8 q2, d16, #4
- vshll.u8 q3, d17, #4
- vshll.u8 q8, d18, #4
- vshll.u8 q9, d19, #4
- vld1.8 {q10, q11}, [r1], r2- vst1.16 {q14, q15}, [r0, :128], r6- vst1.16 {q0, q1}, [r8, :128], r6- vshll.u8 q12, d20, #4
- vshll.u8 q13, d21, #4
- vshll.u8 q14, d22, #4
- vshll.u8 q15, d23, #4
- subs r4, r4, #1
- vst1.16 {q2, q3}, [r0, :128], r6- vst1.16 {q8, q9}, [r8, :128], r6- vst1.16 {q12, q13}, [r0, :128], r6- vst1.16 {q14, q15}, [r8, :128], r6- bgt 128b
- pop {r4-r11,pc}-endfunc
-
function blend_8bpc_neon, export=1
push {r4-r8,lr}ldr r4, [sp, #24]
@@ -854,6 +630,232 @@
bgt 32b
pop {r4-r8,pc}endfunc
+
+
+// This has got the same signature as the put_8tap functions,
+// assumes that the caller has loaded the h argument into r5,
+// and assumes that r8 is set to (clz(w)-24).
+function put_neon
+ adr r9, L(put_tbl)
+ ldr r8, [r9, r8, lsl #2]
+ add r9, r9, r8
+ bx r9
+
+ .align 2
+L(put_tbl):
+ .word 1280f - L(put_tbl) + CONFIG_THUMB
+ .word 640f - L(put_tbl) + CONFIG_THUMB
+ .word 32f - L(put_tbl) + CONFIG_THUMB
+ .word 160f - L(put_tbl) + CONFIG_THUMB
+ .word 8f - L(put_tbl) + CONFIG_THUMB
+ .word 4f - L(put_tbl) + CONFIG_THUMB
+ .word 2f - L(put_tbl) + CONFIG_THUMB
+
+2:
+ vld1.16 {d0[]}, [r2], r3+ vld1.16 {d1[]}, [r2], r3+ subs r5, r5, #2
+ vst1.16 {d0[0]}, [r0, :16], r1+ vst1.16 {d1[0]}, [r0, :16], r1+ bgt 2b
+ pop {r4-r11,pc}+4:
+ vld1.32 {d0[]}, [r2], r3+ vld1.32 {d1[]}, [r2], r3+ subs r5, r5, #2
+ vst1.32 {d0[0]}, [r0, :32], r1+ vst1.32 {d1[0]}, [r0, :32], r1+ bgt 4b
+ pop {r4-r11,pc}+8:
+ vld1.8 {d0}, [r2], r3+ vld1.8 {d1}, [r2], r3+ subs r5, r5, #2
+ vst1.8 {d0}, [r0, :64], r1+ vst1.8 {d1}, [r0, :64], r1+ bgt 8b
+ pop {r4-r11,pc}+160:
+ add r8, r0, r1
+ lsl r1, r1, #1
+ add r9, r2, r3
+ lsl r3, r3, #1
+16:
+ vld1.8 {q0}, [r2], r3+ vld1.8 {q1}, [r9], r3+ subs r5, r5, #2
+ vst1.8 {q0}, [r0, :128], r1+ vst1.8 {q1}, [r8, :128], r1+ bgt 16b
+ pop {r4-r11,pc}+32:
+ vld1.8 {q0, q1}, [r2], r3+ subs r5, r5, #1
+ vst1.8 {q0, q1}, [r0, :128], r1+ bgt 32b
+ pop {r4-r11,pc}+640:
+ sub r1, r1, #32
+ sub r3, r3, #32
+64:
+ vld1.8 {q0, q1}, [r2]!+ vst1.8 {q0, q1}, [r0, :128]!+ vld1.8 {q2, q3}, [r2], r3+ subs r5, r5, #1
+ vst1.8 {q2, q3}, [r0, :128], r1+ bgt 64b
+ pop {r4-r11,pc}+1280:
+ sub r1, r1, #96
+ sub r3, r3, #96
+128:
+ vld1.8 {q8, q9}, [r2]!+ vst1.8 {q8, q9}, [r0, :128]!+ vld1.8 {q10, q11}, [r2]!+ vst1.8 {q10, q11}, [r0, :128]!+ vld1.8 {q12, q13}, [r2]!+ vst1.8 {q12, q13}, [r0, :128]!+ vld1.8 {q14, q15}, [r2], r3+ subs r5, r5, #1
+ vst1.8 {q14, q15}, [r0, :128], r1+ bgt 128b
+ pop {r4-r11,pc}+endfunc
+
+
+// This has got the same signature as the put_8tap functions,
+// assumes that the caller has loaded the h argument into r4,
+// and assumes that r8 is set to (clz(w)-24), and r7 to w*2.
+function prep_neon
+ adr r9, L(prep_tbl)
+ ldr r8, [r9, r8, lsl #2]
+ add r9, r9, r8
+ bx r9
+
+ .align 2
+L(prep_tbl):
+ .word 1280f - L(prep_tbl) + CONFIG_THUMB
+ .word 640f - L(prep_tbl) + CONFIG_THUMB
+ .word 320f - L(prep_tbl) + CONFIG_THUMB
+ .word 160f - L(prep_tbl) + CONFIG_THUMB
+ .word 8f - L(prep_tbl) + CONFIG_THUMB
+ .word 4f - L(prep_tbl) + CONFIG_THUMB
+
+4:
+ vld1.32 {d0[]}, [r1], r2+ vld1.32 {d2[]}, [r1], r2+ subs r4, r4, #2
+ vshll.u8 q0, d0, #4
+ vshll.u8 q1, d2, #4
+ vst1.16 {d1, d2}, [r0, :64]!+ bgt 4b
+ pop {r4-r11,pc}+8:
+ vld1.8 {d0}, [r1], r2+ vld1.8 {d2}, [r1], r2+ subs r4, r4, #2
+ vshll.u8 q0, d0, #4
+ vshll.u8 q1, d2, #4
+ vst1.16 {q0, q1}, [r0, :128]!+ bgt 8b
+ pop {r4-r11,pc}+160:
+ add r9, r1, r2
+ lsl r2, r2, #1
+ add r8, r0, r7
+ lsl r7, r7, #1
+16:
+ vld1.8 {q2}, [r1], r2+ vld1.8 {q3}, [r9], r2+ subs r4, r4, #2
+ vshll.u8 q0, d4, #4
+ vshll.u8 q1, d5, #4
+ vshll.u8 q2, d6, #4
+ vshll.u8 q3, d7, #4
+ vst1.16 {q0, q1}, [r0, :128], r7+ vst1.16 {q2, q3}, [r8, :128], r7+ bgt 16b
+ pop {r4-r11,pc}+320:
+ add r8, r0, r3
+32:
+ vld1.8 {q0, q1}, [r1], r2+ subs r4, r4, #2
+ vshll.u8 q8, d0, #4
+ vshll.u8 q9, d1, #4
+ vld1.8 {q2, q3}, [r1], r2+ vshll.u8 q10, d2, #4
+ vshll.u8 q11, d3, #4
+ vshll.u8 q12, d4, #4
+ vst1.16 {q8, q9}, [r0, :128], r7+ vshll.u8 q13, d5, #4
+ vst1.16 {q10, q11}, [r8, :128], r7+ vshll.u8 q14, d6, #4
+ vst1.16 {q12, q13}, [r0, :128], r7+ vshll.u8 q15, d7, #4
+ vst1.16 {q14, q15}, [r8, :128], r7+ bgt 32b
+ pop {r4-r11,pc}+640:
+ sub r2, r2, #32
+ add r8, r0, #32
+ mov r6, #64
+64:
+ vld1.8 {q0, q1}, [r1]!+ subs r4, r4, #1
+ vshll.u8 q8, d0, #4
+ vshll.u8 q9, d1, #4
+ vld1.8 {q2, q3}, [r1], r2+ vshll.u8 q10, d2, #4
+ vshll.u8 q11, d3, #4
+ vshll.u8 q12, d4, #4
+ vst1.16 {q8, q9}, [r0, :128], r6+ vshll.u8 q13, d5, #4
+ vshll.u8 q14, d6, #4
+ vst1.16 {q10, q11}, [r8, :128], r6+ vshll.u8 q15, d7, #4
+ vst1.16 {q12, q13}, [r0, :128], r6+ vst1.16 {q14, q15}, [r8, :128], r6+ bgt 64b
+ pop {r4-r11,pc}+1280:
+ sub r2, r2, #96
+ add r8, r0, #32
+ mov r6, #64
+128:
+ vld1.8 {q0, q1}, [r1]!+ vld1.8 {q2, q3}, [r1]!+ vshll.u8 q10, d0, #4
+ vshll.u8 q11, d1, #4
+ vshll.u8 q12, d2, #4
+ vshll.u8 q13, d3, #4
+ vshll.u8 q14, d4, #4
+ vshll.u8 q15, d5, #4
+ vld1.8 {q8, q9}, [r1]!+ vst1.16 {q10, q11}, [r0, :128], r6+ vst1.16 {q12, q13}, [r8, :128], r6+ vshll.u8 q0, d6, #4
+ vshll.u8 q1, d7, #4
+ vshll.u8 q2, d16, #4
+ vshll.u8 q3, d17, #4
+ vshll.u8 q8, d18, #4
+ vshll.u8 q9, d19, #4
+ vld1.8 {q10, q11}, [r1], r2+ vst1.16 {q14, q15}, [r0, :128], r6+ vst1.16 {q0, q1}, [r8, :128], r6+ vshll.u8 q12, d20, #4
+ vshll.u8 q13, d21, #4
+ vshll.u8 q14, d22, #4
+ vshll.u8 q15, d23, #4
+ subs r4, r4, #1
+ vst1.16 {q2, q3}, [r0, :128], r6+ vst1.16 {q8, q9}, [r8, :128], r6+ vst1.16 {q12, q13}, [r0, :128], r6+ vst1.16 {q14, q15}, [r8, :128], r6+ bgt 128b
+ pop {r4-r11,pc}+endfunc
+
.macro load_slice s0, s1, strd, wd, d0, d1, d2, d3, d4, d5, d6
vld1.\wd {\d0[]}, [\s0], \strd