1
0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00

aarch64: Use ret x<n> instead of br x<n> where possible

Change AArch64 assembly code to use:
  ret     x<n>
instead of:
  br      x<n>

"ret x<n>" is already used in a lot of places so this patch makes it
consistent across the code base. This does not change behavior or
performance.

In addition, this change reduces the number of landing pads needed in
a subsequent patch to support the Armv8.5-A Branch Target
Identification (BTI) security feature.

Signed-off-by: Jonathan Wright <jonathan.wright@arm.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
This commit is contained in:
Jonathan Wright 2020-09-28 13:35:51 +01:00 committed by Martin Storsjö
parent 20c66fe2f9
commit 6f04cf54f5
5 changed files with 73 additions and 73 deletions

View File

@ -58,7 +58,7 @@ endconst
.endm
.macro idct_end
br x10
ret x10
.endm
.macro smull1 a, b, c

View File

@ -1040,7 +1040,7 @@ function \txfm\()16_1d_4x16_pass1_neon
.irp i, 16, 20, 24, 28, 17, 21, 25, 29, 18, 22, 26, 30, 19, 23, 27, 31
store \i, x0, #16
.endr
br x14
ret x14
1:
// Special case: For the last input column (x1 == 12),
// which would be stored as the last row in the temp buffer,
@ -1068,7 +1068,7 @@ function \txfm\()16_1d_4x16_pass1_neon
mov v29.16b, v17.16b
mov v30.16b, v18.16b
mov v31.16b, v19.16b
br x14
ret x14
endfunc
// Read a vertical 4x16 slice out of a 16x16 matrix, do a transform on it,
@ -1098,7 +1098,7 @@ function \txfm\()16_1d_4x16_pass2_neon
load_add_store v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
load_add_store v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
br x14
ret x14
endfunc
.endm
@ -1208,7 +1208,7 @@ function vp9_\txfm1\()_\txfm2\()_16x16_add_16_neon
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
.endif
br x15
ret x15
endfunc
function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_10_neon, export=1
@ -1264,7 +1264,7 @@ function idct16_1d_4x16_pass1_quarter_neon
st1 {v23.4s}, [x0], #16
st1 {v27.4s}, [x0], #16
st1 {v31.4s}, [x0], #16
br x14
ret x14
endfunc
function idct16_1d_4x16_pass2_quarter_neon
@ -1286,7 +1286,7 @@ function idct16_1d_4x16_pass2_quarter_neon
load_add_store v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
load_add_store v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
br x14
ret x14
endfunc
function idct16_1d_4x16_pass1_half_neon
@ -1313,7 +1313,7 @@ function idct16_1d_4x16_pass1_half_neon
.irp i, 16, 20, 24, 28, 17, 21, 25, 29, 18, 22, 26, 30, 19, 23, 27, 31
store \i, x0, #16
.endr
br x14
ret x14
1:
// Special case: For the second input column (r1 == 4),
// which would be stored as the second row in the temp buffer,
@ -1341,7 +1341,7 @@ function idct16_1d_4x16_pass1_half_neon
mov v21.16b, v17.16b
mov v22.16b, v18.16b
mov v23.16b, v19.16b
br x14
ret x14
endfunc
function idct16_1d_4x16_pass2_half_neon
@ -1364,7 +1364,7 @@ function idct16_1d_4x16_pass2_half_neon
load_add_store v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
load_add_store v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
br x14
ret x14
endfunc
.macro idct16_partial size
@ -1390,7 +1390,7 @@ function idct16x16_\size\()_add_16_neon
add sp, sp, #1024
ldp d8, d9, [sp], 0x10
br x15
ret x15
endfunc
.endm
@ -1729,7 +1729,7 @@ function idct32_1d_4x32_pass1\suffix\()_neon
store_rev v29.4s, v25.4s, v21.4s, v17.4s, v29.16b, v25.16b
store_rev v28.4s, v24.4s, v20.4s, v16.4s, v28.16b, v24.16b
.purgem store_rev
br x14
ret x14
endfunc
// This is mostly the same as 4x32_pass1, but without the transpose,
@ -1849,7 +1849,7 @@ function idct32_1d_4x32_pass2\suffix\()_neon
load_acc_store v24.4s, v25.4s, v26.4s, v27.4s, 1
load_acc_store v28.4s, v29.4s, v30.4s, v31.4s, 1
.purgem load_acc_store
br x14
ret x14
endfunc
.endm
@ -1943,7 +1943,7 @@ function vp9_idct_idct_32x32_add_16_neon
ldp d10, d11, [sp], 0x10
ldp d8, d9, [sp], 0x10
br x15
ret x15
endfunc
function ff_vp9_idct_idct_32x32_add_10_neon, export=1
@ -2009,7 +2009,7 @@ function idct32x32_\size\()_add_16_neon
ldp d10, d11, [sp], 0x10
ldp d8, d9, [sp], 0x10
br x15
ret x15
endfunc
.endm

View File

@ -787,7 +787,7 @@ function \txfm\()16_1d_8x16_pass1_neon
.irp i, 16, 24, 17, 25, 18, 26, 19, 27, 20, 28, 21, 29, 22, 30, 23, 31
store \i, x0, #16
.endr
br x14
ret x14
1:
// Special case: For the last input column (x1 == 8),
// which would be stored as the last row in the temp buffer,
@ -806,7 +806,7 @@ function \txfm\()16_1d_8x16_pass1_neon
mov v29.16b, v21.16b
mov v30.16b, v22.16b
mov v31.16b, v23.16b
br x14
ret x14
endfunc
// Read a vertical 8x16 slice out of a 16x16 matrix, do a transform on it,
@ -834,7 +834,7 @@ function \txfm\()16_1d_8x16_pass2_neon
load_add_store v16.8h, v17.8h, v18.8h, v19.8h, v20.8h, v21.8h, v22.8h, v23.8h, v16.8b, v17.8b
load_add_store v24.8h, v25.8h, v26.8h, v27.8h, v28.8h, v29.8h, v30.8h, v31.8h, v16.8b, v17.8b
br x14
ret x14
endfunc
.endm
@ -925,7 +925,7 @@ function ff_vp9_\txfm1\()_\txfm2\()_16x16_add_neon, export=1
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
.endif
br x15
ret x15
endfunc
.endm
@ -960,7 +960,7 @@ function idct16_1d_8x16_pass1_quarter_neon
.irp i, 24, 25, 26, 27
store \i, x0, x9
.endr
br x14
ret x14
endfunc
function idct16_1d_8x16_pass2_quarter_neon
@ -978,7 +978,7 @@ function idct16_1d_8x16_pass2_quarter_neon
load_add_store v16.8h, v17.8h, v18.8h, v19.8h, v20.8h, v21.8h, v22.8h, v23.8h, v16.8b, v17.8b
load_add_store v24.8h, v25.8h, v26.8h, v27.8h, v28.8h, v29.8h, v30.8h, v31.8h, v16.8b, v17.8b
br x14
ret x14
endfunc
function idct16_1d_8x16_pass1_half_neon
@ -1003,7 +1003,7 @@ function idct16_1d_8x16_pass1_half_neon
.irp i, 24, 25, 26, 27, 28, 29, 30, 31
store \i, x0, x9
.endr
br x14
ret x14
endfunc
function idct16_1d_8x16_pass2_half_neon
@ -1021,7 +1021,7 @@ function idct16_1d_8x16_pass2_half_neon
load_add_store v16.8h, v17.8h, v18.8h, v19.8h, v20.8h, v21.8h, v22.8h, v23.8h, v16.8b, v17.8b
load_add_store v24.8h, v25.8h, v26.8h, v27.8h, v28.8h, v29.8h, v30.8h, v31.8h, v16.8b, v17.8b
br x14
ret x14
endfunc
.macro idct16_partial size
@ -1038,7 +1038,7 @@ function idct16x16_\size\()_add_neon
.endr
add sp, sp, #512
br x15
ret x15
endfunc
.endm
@ -1349,7 +1349,7 @@ function idct32_1d_8x32_pass1\suffix\()_neon
store_rev v25.8h, v17.8h
store_rev v24.8h, v16.8h
.purgem store_rev
br x14
ret x14
endfunc
// This is mostly the same as 8x32_pass1, but without the transpose,
@ -1466,7 +1466,7 @@ function idct32_1d_8x32_pass2\suffix\()_neon
load_acc_store v24.8h, v25.8h, v26.8h, v27.8h, 1
load_acc_store v28.8h, v29.8h, v30.8h, v31.8h, 1
.purgem load_acc_store
br x14
ret x14
endfunc
.endm
@ -1547,7 +1547,7 @@ function ff_vp9_idct_idct_32x32_add_neon, export=1
ldp d8, d9, [sp], 0x10
ldp d10, d11, [sp], 0x10
br x15
ret x15
endfunc
.macro idct32_partial size
@ -1572,7 +1572,7 @@ function idct32x32_\size\()_add_neon
ldp d8, d9, [sp], 0x10
ldp d10, d11, [sp], 0x10
br x15
ret x15
endfunc
.endm

View File

@ -57,7 +57,7 @@
mov x12, v4.d[1]
adds x11, x11, x12
b.ne 1f
br x10
ret x10
1:
.if \wd >= 8
@ -193,7 +193,7 @@
b.eq 6f
.else
b.ne 1f
br x13
ret x13
1:
.endif
@ -252,7 +252,7 @@
b.ne 1f
// If no pixels needed flat8in nor flat8out, jump to a
// writeout of the inner 4 pixels
br x14
ret x14
1:
mov x11, v7.d[0]
@ -260,7 +260,7 @@
adds x11, x11, x12
b.ne 1f
// If no pixels need flat8out, jump to a writeout of the inner 6 pixels
br x15
ret x15
1:
// flat8out
@ -434,7 +434,7 @@ function ff_\func\()_\bpp\()_neon, export=1
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x16
ret x16
.else
b \func\()_16_neon
.endif
@ -474,7 +474,7 @@ function ff_\func\()_\suffix\()_\bpp\()_neon, export=1
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
.endif
br x16
ret x16
endfunc
.endm
@ -508,7 +508,7 @@ function ff_vp9_loop_filter_\dir\()_\wd1\()\wd2\()_16_\bpp\()_neon, export=1
lsl w3, w14, #\bpp - 8
lsl w4, w15, #\bpp - 8
bl vp9_loop_filter_\dir\()_\wd2\()_8_16_neon
br x16
ret x16
endfunc
.endm
@ -541,7 +541,7 @@ function vp9_loop_filter_v_4_8_16_neon
st1 {v25.8h}, [x0], x1
sub x0, x0, x1, lsl #1
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_v_4_8
@ -589,7 +589,7 @@ function vp9_loop_filter_h_4_8_16_neon
sub x0, x0, x1, lsl #3
add x0, x0, #4
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_h_4_8
@ -620,7 +620,7 @@ function vp9_loop_filter_v_8_8_16_neon
sub x0, x0, x1, lsl #1
sub x0, x0, x1
br x10
ret x10
6:
sub x9, x0, x1, lsl #1
st1 {v22.8h}, [x9], x1
@ -628,7 +628,7 @@ function vp9_loop_filter_v_8_8_16_neon
st1 {v23.8h}, [x9], x1
st1 {v25.8h}, [x0], x1
sub x0, x0, x1, lsl #1
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_v_8_8
@ -671,7 +671,7 @@ function vp9_loop_filter_h_8_8_16_neon
sub x0, x0, x1, lsl #3
add x0, x0, #8
br x10
ret x10
6:
// If we didn't need to do the flat8in part, we use the same writeback
// as in loop_filter_h_4_8.
@ -688,7 +688,7 @@ function vp9_loop_filter_h_8_8_16_neon
st1 {v25.d}[1], [x0], x1
sub x0, x0, x1, lsl #3
add x0, x0, #4
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_h_8_8
@ -743,7 +743,7 @@ function vp9_loop_filter_v_16_8_16_neon
sub x0, x0, x1, lsl #3
add x0, x0, x1
br x10
ret x10
8:
add x9, x9, x1, lsl #2
// If we didn't do the flat8out part, the output is left in the
@ -756,7 +756,7 @@ function vp9_loop_filter_v_16_8_16_neon
st1 {v26.8h}, [x0], x1
sub x0, x0, x1, lsl #1
sub x0, x0, x1
br x10
ret x10
7:
sub x9, x0, x1, lsl #1
st1 {v22.8h}, [x9], x1
@ -764,7 +764,7 @@ function vp9_loop_filter_v_16_8_16_neon
st1 {v23.8h}, [x9], x1
st1 {v25.8h}, [x0], x1
sub x0, x0, x1, lsl #1
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_v_16_8, push=1
@ -821,7 +821,7 @@ function vp9_loop_filter_h_16_8_16_neon
st1 {v31.8h}, [x0], x1
sub x0, x0, x1, lsl #3
br x10
ret x10
8:
// The same writeback as in loop_filter_h_8_8
sub x9, x0, #8
@ -838,7 +838,7 @@ function vp9_loop_filter_h_16_8_16_neon
st1 {v27.8h}, [x0], x1
sub x0, x0, x1, lsl #3
add x0, x0, #8
br x10
ret x10
7:
// The same writeback as in loop_filter_h_4_8
sub x9, x0, #4
@ -854,7 +854,7 @@ function vp9_loop_filter_h_16_8_16_neon
st1 {v25.d}[1], [x0], x1
sub x0, x0, x1, lsl #3
add x0, x0, #4
br x10
ret x10
endfunc
bpp_frontends vp9_loop_filter_h_16_8, push=1

View File

@ -399,7 +399,7 @@
.endif
// If no pixels needed flat8in nor flat8out, jump to a
// writeout of the inner 4 pixels
br x14
ret x14
1:
mov x5, v7.d[0]
@ -411,7 +411,7 @@
cbnz x5, 1f
.endif
// If no pixels need flat8out, jump to a writeout of the inner 6 pixels
br x15
ret x15
1:
// flat8out
@ -532,32 +532,32 @@ function vp9_loop_filter_4
loop_filter 4, .8b, 0, v16, v17, v18, v19, v28, v29, v30, v31
ret
9:
br x10
ret x10
endfunc
function vp9_loop_filter_4_16b_mix_44
loop_filter 4, .16b, 44, v16, v17, v18, v19, v28, v29, v30, v31
ret
9:
br x10
ret x10
endfunc
function vp9_loop_filter_8
loop_filter 8, .8b, 0, v16, v17, v18, v19, v28, v29, v30, v31
ret
6:
br x13
ret x13
9:
br x10
ret x10
endfunc
function vp9_loop_filter_8_16b_mix
loop_filter 8, .16b, 88, v16, v17, v18, v19, v28, v29, v30, v31
ret
6:
br x13
ret x13
9:
br x10
ret x10
endfunc
function vp9_loop_filter_16
@ -568,7 +568,7 @@ function vp9_loop_filter_16
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
endfunc
function vp9_loop_filter_16_16b
@ -579,7 +579,7 @@ function vp9_loop_filter_16_16b
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
endfunc
.macro loop_filter_4
@ -648,7 +648,7 @@ function ff_vp9_loop_filter_v_4_8_neon, export=1
st1 {v23.8b}, [x9], x1
st1 {v25.8b}, [x0], x1
br x10
ret x10
endfunc
function ff_vp9_loop_filter_v_44_16_neon, export=1
@ -672,7 +672,7 @@ function ff_vp9_loop_filter_v_44_16_neon, export=1
st1 {v23.16b}, [x9], x1
st1 {v25.16b}, [x0], x1
br x10
ret x10
endfunc
function ff_vp9_loop_filter_h_4_8_neon, export=1
@ -714,7 +714,7 @@ function ff_vp9_loop_filter_h_4_8_neon, export=1
st1 {v25.s}[0], [x9], x1
st1 {v25.s}[1], [x0], x1
br x10
ret x10
endfunc
function ff_vp9_loop_filter_h_44_16_neon, export=1
@ -766,7 +766,7 @@ function ff_vp9_loop_filter_h_44_16_neon, export=1
st1 {v25.s}[1], [x9], x1
st1 {v25.s}[3], [x0], x1
br x10
ret x10
endfunc
function ff_vp9_loop_filter_v_8_8_neon, export=1
@ -793,14 +793,14 @@ function ff_vp9_loop_filter_v_8_8_neon, export=1
st1 {v23.8b}, [x9], x1
st1 {v26.8b}, [x0], x1
br x10
ret x10
6:
sub x9, x0, x1, lsl #1
st1 {v22.8b}, [x9], x1
st1 {v24.8b}, [x0], x1
st1 {v23.8b}, [x9], x1
st1 {v25.8b}, [x0], x1
br x10
ret x10
endfunc
.macro mix_v_16 mix
@ -828,14 +828,14 @@ function ff_vp9_loop_filter_v_\mix\()_16_neon, export=1
st1 {v23.16b}, [x9], x1
st1 {v26.16b}, [x0], x1
br x10
ret x10
6:
sub x9, x0, x1, lsl #1
st1 {v22.16b}, [x9], x1
st1 {v24.16b}, [x0], x1
st1 {v23.16b}, [x9], x1
st1 {v25.16b}, [x0], x1
br x10
ret x10
endfunc
.endm
@ -876,7 +876,7 @@ function ff_vp9_loop_filter_h_8_8_neon, export=1
st1 {v23.8b}, [x9], x1
st1 {v27.8b}, [x0], x1
br x10
ret x10
6:
// If we didn't need to do the flat8in part, we use the same writeback
// as in loop_filter_h_4_8.
@ -891,7 +891,7 @@ function ff_vp9_loop_filter_h_8_8_neon, export=1
st1 {v24.s}[1], [x0], x1
st1 {v25.s}[0], [x9], x1
st1 {v25.s}[1], [x0], x1
br x10
ret x10
endfunc
.macro mix_h_16 mix
@ -942,7 +942,7 @@ function ff_vp9_loop_filter_h_\mix\()_16_neon, export=1
st1 {v27.8b}, [x9], x1
st1 {v27.d}[1], [x0], x1
br x10
ret x10
6:
add x9, x9, #2
add x0, x0, #2
@ -963,7 +963,7 @@ function ff_vp9_loop_filter_h_\mix\()_16_neon, export=1
st1 {v24.s}[3], [x0], x1
st1 {v25.s}[1], [x9], x1
st1 {v25.s}[3], [x0], x1
br x10
ret x10
endfunc
.endm
@ -1022,7 +1022,7 @@ function ff_vp9_loop_filter_v_16_8_neon, export=1
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
8:
add x9, x9, x1, lsl #2
// If we didn't do the flat8out part, the output is left in the
@ -1091,7 +1091,7 @@ function ff_vp9_loop_filter_v_16_16_neon, export=1
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
8:
add x9, x9, x1, lsl #2
st1 {v21.16b}, [x9], x1
@ -1168,7 +1168,7 @@ function ff_vp9_loop_filter_h_16_8_neon, export=1
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
8:
// The same writeback as in loop_filter_h_8_8
sub x9, x0, #4
@ -1287,7 +1287,7 @@ function ff_vp9_loop_filter_h_16_16_neon, export=1
ldp d10, d11, [sp], 0x10
ldp d12, d13, [sp], 0x10
ldp d14, d15, [sp], 0x10
br x10
ret x10
8:
sub x9, x0, #4
add x0, x9, x1, lsl #3