forked from enlightenment/efl
Evas: Neon: More .fpu neon flags
Also clean up some ugly code. SVN revision: 55314
This commit is contained in:
parent
ee03254a3e
commit
157bf3a573
|
@ -136,6 +136,7 @@ evas_common_copy_pixels_rev_neon(DATA32 *src, DATA32 *dst, int len)
|
|||
uint32_t *tmp = (void *)37;
|
||||
#define AP "evas_common_copy_rev_pixels_neon_"
|
||||
asm volatile (
|
||||
".fpu neon \n\t"
|
||||
// Can we do 32 byte?
|
||||
"andS %[tmp], %[d], $0x1f \n\t"
|
||||
"beq "AP"quadstart \n\t"
|
||||
|
@ -334,6 +335,7 @@ evas_common_copy_pixels_neon(DATA32 *src, DATA32 *dst, int len){
|
|||
e = dst + len;
|
||||
#define AP "evas_common_copy_pixels_neon_"
|
||||
asm volatile (
|
||||
".fpu neon \n\t"
|
||||
// Can we do 32 byte?
|
||||
"andS %[tmp], %[d], $0x1f \n\t"
|
||||
"beq "AP"quadstart \n\t"
|
||||
|
|
|
@ -102,6 +102,7 @@ evas_common_convert_rgba_to_32bpp_rgb_8888_rot_90 (DATA32 *src, DATA8 *dst, int
|
|||
} else {
|
||||
#define AP "convert_rgba32_rot_90_"
|
||||
asm volatile (
|
||||
".fpu neon \n\t"
|
||||
" mov %[s1], %[src] \n\t"
|
||||
" add %[s1], %[h],lsl #2 \n\t"
|
||||
" sub %[s1], #8 \n\t"
|
||||
|
|
|
@ -151,12 +151,8 @@ _op_blend_p_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
|
|||
"bhi "AP"octloopint\n\t"
|
||||
|
||||
AP "loopout: \n\t"
|
||||
//"sub %[tmp], %[d], #4\n\t"
|
||||
//"vmov.i16 d0, $0xff00 \n\t"
|
||||
//"vst1.32 d0[0], [%[tmp]] \n\t"
|
||||
|
||||
"cmp %[d], %[e]\n\t"
|
||||
"beq "AP"done\n\t"
|
||||
"cmp %[d], %[e] \n\t"
|
||||
"beq "AP"done \n\t"
|
||||
"sub %[tmp],%[e], %[d] \n\t"
|
||||
"cmp %[tmp],$0x04 \n\t"
|
||||
"ble "AP"singleloop2 \n\t"
|
||||
|
@ -312,12 +308,12 @@ _op_blend_pas_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
|
|||
"vshr.u32 q5, q5,$0x18 \n\t"
|
||||
|
||||
// Prepare to preload
|
||||
"add %[pl], %[s], #32\n\t"
|
||||
"add %[pl], %[s], #32 \n\t"
|
||||
|
||||
// Mulitply into all fields
|
||||
"vmul.u32 q4, q4, q8 \n\t"
|
||||
"vmul.u32 q5, q5, q8 \n\t"
|
||||
"pld [%[pl]]\n\t"
|
||||
"pld [%[pl]] \n\t"
|
||||
|
||||
// Multiply out
|
||||
"vmull.u8 q6, d8, d4 \n\t"
|
||||
|
@ -325,13 +321,13 @@ _op_blend_pas_dp_neon(DATA32 *s, DATA8 *m, DATA32 c, DATA32 *d, int l) {
|
|||
"vmull.u8 q2, d9, d5 \n\t"
|
||||
"vmull.u8 q3, d11, d7 \n\t"
|
||||
|
||||
"add %[pl], %[d], #32\n\t"
|
||||
"add %[pl], %[d], #32 \n\t"
|
||||
|
||||
"vqrshrn.u16 d8, q6, #8 \n\t"
|
||||
"vqrshrn.u16 d10, q7, #8 \n\t"
|
||||
"vqrshrn.u16 d9, q2, #8 \n\t"
|
||||
"vqrshrn.u16 d11, q3, #8 \n\t"
|
||||
"pld [%[pl]]\n\t"
|
||||
"pld [%[pl]] \n\t"
|
||||
|
||||
"cmp %[tmp], %[pl] \n\t"
|
||||
// Add to s
|
||||
|
|
|
@ -6,7 +6,7 @@ _op_copy_p_dp_neon(DATA32 *s, DATA8 *m __UNUSED__, DATA32 c __UNUSED__, DATA32 *
|
|||
uint32_t *e;
|
||||
e = d + l;
|
||||
//#ifdef NEON_INSTRINSICS_OK
|
||||
#if 1
|
||||
#if 0
|
||||
// odd this is faster than the below asm... :(
|
||||
e -= 15;
|
||||
uint32x4_t col1, col2, col3, col4;
|
||||
|
|
Loading…
Reference in New Issue