adolforeyna/tactility
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update LVGL (master 9.2+) and related libraries (#130)

Browse Source 
- LVGL 9.2+ (master commit)
- esp_lvgl_port (master with my PR hotfix changes)
- espressif/esp_lcd_touch_gt911 1.1.1~2
This commit is contained in:
Ken Van Hoeylandt
2024-12-16 23:30:57 +01:00
committed by GitHub
parent e4b8519f67
commit 80b69b7f45
90 changed files with 7328 additions and 469 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Libraries/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_argb8888_esp32.S
+81
View File
@@ -0,0 +1,81 @@
/*
* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
// This is LVGL ARGB8888 simple fill for ESP32 processor
.section .text
.align 4
.global lv_color_blend_to_argb8888_esp
.type lv_color_blend_to_argb8888_esp,@function
// The function implements the following C code:
// void lv_color_blend_to_argb8888(_lv_draw_sw_blend_fill_dsc_t * dsc);
// Input params
//
// dsc - a2
// typedef struct {
// uint32_t opa; l32i 0
// void * dst_buf; l32i 4
// uint32_t dst_w; l32i 8
// uint32_t dst_h; l32i 12
// uint32_t dst_stride; l32i 16
// const void * src_buf; l32i 20
// uint32_t src_stride; l32i 24
// const lv_opa_t * mask_buf; l32i 28
// uint32_t mask_stride; l32i 32
// } asm_dsc_t;
lv_color_blend_to_argb8888_esp:
entry a1, 32
l32i.n a3, a2, 4 // a3 - dest_buff
l32i.n a4, a2, 8 // a4 - dest_w in uint32_t
l32i.n a5, a2, 12 // a5 - dest_h in uint32_t
l32i.n a6, a2, 16 // a6 - dest_stride in bytes
l32i.n a7, a2, 20 // a7 - src_buff (color)
l32i.n a8, a7, 0 // a8 - color as value
slli a11, a4, 2 // a11 - dest_w_bytes = sizeof(uint32_t) * dest_w
movi a7, 0xff000000 // oppactiy mask
or a10, a7, a8 // apply oppacity
srli a9, a4, 2 // a9 - loop_len = dest_w / 4
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
.outer_loop:
// Run main loop which sets 16 bytes in one loop run
loopnez a9, ._main_loop
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 8 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 12 // save 32 bits from a10 to dest_buff a3
addi.n a3, a3, 16 // increment dest_buff pointer by 16 bytes
._main_loop:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 8 bytes
bbci a11, 3, _mod_8_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_mod_8_check:
// Check modulo 4 of the dest_w_bytes, if - then set 4 bytes
bbci a11, 2, _mod_4_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_mod_4_check:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
Libraries/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_argb8888_esp32s3.S
+326
View File
@@ -0,0 +1,326 @@
/*
* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
// This is LVGL ARGB8888 simple fill for ESP32S3 processor
.section .text
.align 4
.global lv_color_blend_to_argb8888_esp
.type lv_color_blend_to_argb8888_esp,@function
// The function implements the following C code:
// void lv_color_blend_to_argb8888(_lv_draw_sw_blend_fill_dsc_t * dsc);
// Input params
//
// dsc - a2
// typedef struct {
// uint32_t opa; l32i 0
// void * dst_buf; l32i 4
// uint32_t dst_w; l32i 8
// uint32_t dst_h; l32i 12
// uint32_t dst_stride; l32i 16
// const void * src_buf; l32i 20
// uint32_t src_stride; l32i 24
// const lv_opa_t * mask_buf; l32i 28
// uint32_t mask_stride; l32i 32
// } asm_dsc_t;
lv_color_blend_to_argb8888_esp:
entry a1, 32
ee.zero.q q0 // dummy TIE instruction, to enable the TIE
l32i.n a3, a2, 4 // a3 - dest_buff
l32i.n a4, a2, 8 // a4 - dest_w in uint32_t
l32i.n a5, a2, 12 // a5 - dest_h in uint32_t
l32i.n a6, a2, 16 // a6 - dest_stride in bytes
l32i.n a7, a2, 20 // a7 - src_buff (color)
l32i.n a8, a7, 0 // a8 - color as value
slli a11, a4, 2 // a11 - dest_w_bytes = sizeof(uint32_t) * dest_w
movi a7, 0xff000000 // oppactiy mask
or a10, a7, a8 // apply oppacity
// Check for short lengths
// dest_w should be at least 8, othewise it's not worth using esp32s3 TIE
bgei a4, 8, _esp32s3_implementation // Branch if dest_w is greater than or equal to 8
j .lv_color_blend_to_argb8888_esp32_body // Jump to esp32 implementation
_esp32s3_implementation:
ee.movi.32.q q0, a10, 0 // fill q0 register from a10 by 32 bits
ee.movi.32.q q0, a10, 1
ee.movi.32.q q0, a10, 2
ee.movi.32.q q0, a10, 3
// Check dest_buff alignment
movi.n a7, 0xf // 0xf alignment mask (16-byte alignment)
and a15, a7, a3 // 16-byte alignment mask AND dest_buff pointer
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
// Check dest_stride alignment
and a15, a7, a6 // 16-byte alignment mask AND dest_stride
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
// Check dest_w_bytes alignment
and a15, a7, a11 // 16-byte alignment mask AND dest_w_bytes
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
//**********************************************************************************************************************
// all aligned, the most ideal case
// dest_buff (a3) - 16-byte aligned
// dest_stride (a6) - 16-byte multiple
// dest_w (a4) - 16-byte multiple
srli a9, a4, 2 // a9 - loop_len = dest_w / 4
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
.outer_loop_aligned:
loopnez a9, ._main_loop_aligned // 16 bytes (4 argb8888) in one loop
ee.vst.128.ip q0, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_aligned:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
_unaligned_by_4byte:
// Check dest_buff alignment
movi.n a7, 0x3 // 0x3 alignment mask (4-byte alignment)
and a15, a7, a3 // 4-byte alignment mask AND dest_buff pointer
bnez a15, _unaligned_by_1byte // branch if a15 not equals to zero
// Check dest_stride alignment
and a15, a7, a6 // 4-byte alignment mask AND dest_stride pointer
bnez a15, _unaligned_by_1byte // branch if a15 not equals to zero
//**********************************************************************************************************************
// either dest_buff or dest_stride is not 16-byte aligned
// dest_w is always 4-byte multiple
// all of the following are 4-byte aligned
// dest_buff (a3) - 16-byte, or 4-byte aligned
// dest_stride (a6) - 16-byte, or 4-byte multiple
// dest_w (a4) - 4-byte multiple
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
movi.n a7, 0xf // 0xf alignment mask
.outer_loop_aligned_by_4byte:
// alignment check
and a15, a7, a3 // 0xf (alignment mask) AND dest_buff pointer
mov a12, a11 // a12 - local_dest_w_bytes = dest_w_bytes
beqz a15, _dest_buff_aligned_by_4byte // branch if a15 equals to zero
movi.n a14, 16 // a14 - 16
sub a15, a14, a15 // a15 = 16 - unalignment (lower 4 bits of dest_buff address)
sub a12, a12, a15 // local_dest_w_bytes = len - (16 - unalignment)
// keep setting until dest_buff is aligned
// Check modulo 8 of the unalignment, if - then set 8 bytes
bbci a15, 3, _aligning_mod_8_check_4byte // branch if 3-rd bit of unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 4 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_aligning_mod_8_check_4byte:
// Check modulo 4 of the unalignment, if - then set 4 bytes
bbci a15, 2, _aligning_mod_4_check_4byte // branch if 2-nd bit unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligning_mod_4_check_4byte:
_dest_buff_aligned_by_4byte:
// Calculate main loop_len
srli a9, a12, 4 // a9 - loop_len = local_dest_w_bytes / 16
// Main loop
loopnez a9, ._main_loop_unaligned_by_4byte // 16 bytes (4 argb8888) in one loop
ee.vst.128.ip q0, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_unaligned_by_4byte:
// Check modulo 8 of the dest_w, if - then set 8 bytes
bbci a12, 3, _aligned_mod_8_check_4byte // branch if 3-rd bit of local_dest_w_bytes a12 is clear
ee.vst.l.64.ip q0, a3, 8 // save lower 64 bits from q0 to dest_buff a3, increase dest_buff pointer by 8 bytes
_aligned_mod_8_check_4byte:
// Check modulo 4 of the dest_w, if - then set 4 bytes
bbci a12, 2, _aligned_mod_4_check_4byte // branch if 2-nd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligned_mod_4_check_4byte:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned_by_4byte
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
_unaligned_by_1byte:
//**********************************************************************************************************************
// either dest_buff or dest_stride is not 4-byte aligned
// dest_w is always 4-byte multiple
// dest_buff (a3) - 4-byte, or 1-byte aligned
// dest_stride (a6) - 4-byte, or 1-byte multiple
// dest_w (a4) - 4-byte multiple
ee.zero.q q1 // clear q1
ee.orq q1, q1, q0 // copy q0 to q1
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
movi.n a7, 0xf // 0xf alignment mask
.outer_loop_aligned_by_1byte:
// alignment check
and a15, a7, a3 // 0xf (alignment mask) AND dest_buff pointer
mov a12, a11 // a12 - local_dest_w_bytes = dest_w_bytes
beqz a15, _dest_buff_aligned_by_1byte // branch if a15 equals to zero
movi.n a14, 16 // a14 - 16
sub a15, a14, a15 // a15 = 16 - unalignment (lower 4 bits of dest_buff address)
sub a12, a12, a15 // local_dest_w_bytes = len - (16 - unalignment)
// keep setting until dest_buff is aligned
// Check modulo 8 of the unalignment, if - then set 8 bytes
bbci a15, 3, _aligning_mod_8_check_1byte// branch if 3-rd bit of unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 4 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_aligning_mod_8_check_1byte:
// Check modulo 4 of the unalignment, if - then set 4 bytes
bbci a15, 2, _aligning_mod_4_check_1byte // branch if 2-nd bit unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligning_mod_4_check_1byte:
// Check modulo 2 and 1 (the following 2 ifs do the same correction)
// modulo 2 and modulo 1 requires the same action, just once
bbci a15, 1, _aligning_mod_2_check_1byte
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
j _dest_buff_aligned_by_1byte
_aligning_mod_2_check_1byte:
bbci a15, 0, _dest_buff_aligned_by_1byte
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_dest_buff_aligned_by_1byte:
// Shift q reg, allowing to set 16-byte unaligned adata
wur.sar_byte a15 // apply unalignment to the SAR_BYTE
ee.src.q q2, q0, q1 // shift concat. of q0 and q1 to q2 by SAR_BYTE amount
// Calculate main loop_len
srli a9, a12, 4 // a9 - loop_len = local_dest_w_bytes / 16
// Main loop
loopnez a9, ._main_loop_unaligned_by_1byte // 16 bytes (4 argb8888) in one loop
ee.vst.128.ip q2, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_unaligned_by_1byte:
// Firstly check mod 1 and mod 2 - correcting the aligned memory access
// Go back in one Byte, allow to correct after ee.vst.128.ip aligned access
addi a3, a3, -4
// Check modulo 2 of the dest_w, if - then set 2 bytes
// set SSSS in 0xSSSS0000
bbci a12, 1, _aligned_mod_2_check_1byte // branch if 1-st bit of dest_w a12 is clear
srli a14, a10, 16 // shift a10 in 16, allowing s16i (saving of lower 16 bits)
s16i a14, a3, 2 // save 16 bits from a10 to dest_buff a3, offset 2 bytes
// Check modulo 1 of the dest_w, if - then set 1 byte
// additionally set SS in 0x0000SS00
bbci a12, 0, _aligned_end // branch if 0-th bit of dest_w a12 is clear
srli a14, a10, 8 // shift a10 in 8, allowing s8i
s8i a14, a3, 1 // save 8 bits from a10 to dest_buff a3, offset 1 byte
j _aligned_end
_aligned_mod_2_check_1byte:
// Check modulo 1 of the dest_w, if - then set 1 byte
// set SS in 0xSS000000
bbci a12, 0, _aligned_end // branch if 0-th bit of dest_w a12 is clear
srli a14, a10, 24 // shift a10 in 24, allowing s8i (saving of lower 8 bits)
s8i a14, a3, 3 // save 8 bits from a10 to dest_buff a3, offset 3 bytes
_aligned_end:
addi a3, a3, 4 // Increase the pointer back, correction for addi a3, a3, -4
// Check modulo 8 of the dest_w, if - then set 8 bytes
bbci a12, 3, _aligned_mod_8_check_1byte // branch if 3-rd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 4 bytes
//ee.vst.l.64.ip q2, a3, 8 // save lower 64 bits from q0 to dest_buff a3, increase dest_buff pointer by 8 bytes
_aligned_mod_8_check_1byte:
// Check modulo 4 of the dest_w, if - then set 4 bytes
bbci a12, 2, _aligned_mod_4_check_1byte // branch if 2-nd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligned_mod_4_check_1byte:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned_by_1byte
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
.lv_color_blend_to_argb8888_esp32_body:
srli a9, a4, 2 // a9 - loop_len = dest_w / 4
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
.outer_loop:
// Run main loop which sets 16 bytes in one loop run
loopnez a9, ._main_loop
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 8 // save 32 bits from a10 to dest_buff a3
s32i.n a10, a3, 12 // save 32 bits from a10 to dest_buff a3
addi.n a3, a3, 16 // increment dest_buff pointer by 16 bytes
._main_loop:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 8 bytes
bbci a11, 3, _mod_8_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_mod_8_check:
// Check modulo 4 of the dest_w_bytes, if - then set 4 bytes
bbci a11, 2, _mod_4_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_mod_4_check:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
Libraries/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_rgb565_esp32.S
+149
View File
@@ -0,0 +1,149 @@
/*
* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
// This is LVGL RGB565 simple fill for ESP32 processor
.section .text
.align 4
.global lv_color_blend_to_rgb565_esp
.type lv_color_blend_to_rgb565_esp,@function
// The function implements the following C code:
// void lv_color_blend_to_rgb565(_lv_draw_sw_blend_fill_dsc_t * dsc);
// Input params
//
// dsc - a2
// typedef struct {
// uint32_t opa; l32i 0
// void * dst_buf; l32i 4
// uint32_t dst_w; l32i 8
// uint32_t dst_h; l32i 12
// uint32_t dst_stride; l32i 16
// const void * src_buf; l32i 20
// uint32_t src_stride; l32i 24
// const lv_opa_t * mask_buf; l32i 28
// uint32_t mask_stride; l32i 32
// } asm_dsc_t;
lv_color_blend_to_rgb565_esp:
entry a1, 32
l32i.n a3, a2, 4 // a3 - dest_buff
l32i.n a4, a2, 8 // a4 - dest_w in uint16_t
l32i.n a5, a2, 12 // a5 - dest_h in uint16_t
l32i.n a6, a2, 16 // a6 - dest_stride in bytes
l32i.n a7, a2, 20 // a7 - src_buff (color)
l32i.n a8, a7, 0 // a8 - color as value
slli a11, a4, 1 // a11 - dest_w_bytes = sizeof(uint16_t) * dest_w
// Convert color to rgb656
l8ui a15, a7, 2 // red
movi.n a14, 0xf8
and a13, a15, a14
slli a10, a13, 8
l8ui a15, a7, 0 // blue
and a13, a15, a14
srli a12, a13, 3
add a10, a10, a12
l8ui a15, a7, 1 // green
movi.n a14, 0xfc
and a13, a15, a14
slli a12, a13, 3
add a12, a10, a12 // a12 = 16-bit color
slli a10, a12, 16
movi.n a13, 0xFFFF0000
and a10, a10, a13
or a10, a10, a12 // a10 = 32-bit color (16bit + (16bit << 16))
movi.n a8, 0x3 // a8 = 0x3, dest_buff align mask
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
// cache init
// Prepare main loop length and dest_w_bytes
srli a9, a4, 4 // a9 = loop_len = dest_w / 8, calculate main loop_len for original dest_w
slli a11, a4, 1 // a11 = dest_w_bytes = sizeof(uint16_t) * dest_w
addi a4, a4, -1 // a4-- (decrement a4)
s32i.n a9, a1, 0 // cache.orig.loop_len
s32i.n a11, a1, 4 // cache.orig.dest_w_bytes
// Prepare decreased main loop length and dest_w_bytes
srli a9, a4, 4 // a9 = loop_len = dest_w / 8, calculate main loop_len for dest_w - 1
slli a11, a4, 1 // a11 = dest_w_bytes = sizeof(uint16_t) * (dest_w - 1)
s32i.n a9, a1, 8 // cache.decr.loop_len
s32i.n a11, a1, 12 // cache.decr.dest_w_bytes
and a7, a8, a3 // a7 = dest_buff AND 0x3 (chck if the address is 4-byte aligned)
.outer_loop:
// Check if the des_buff is 2-byte aligned
beqz a7, _dest_buff_2_byte_aligned // branch if a7 is equal to zero
s16i a12, a3, 0 // save 16 bits from 16-bit color a12 to dest_buff a3, offset 0
l32i.n a9, a1, 8 // a9 = load cache.decr.loop_len
l32i.n a11, a1, 12 // a11 = load cache.decr.dest_w_bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2
j _dest_buff_unaligned
_dest_buff_2_byte_aligned:
l32i.n a9, a1, 0 // a11 = load cache.orig.loop_len
l32i.n a11, a1, 4 // a11 = load cache.orig.dest_w_bytes
_dest_buff_unaligned:
// Run main loop which sets 16 bytes in one loop run
loopnez a9, ._main_loop
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
s32i.n a10, a3, 8 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 8
s32i.n a10, a3, 12 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 12
s32i.n a10, a3, 16 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 16
s32i.n a10, a3, 20 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 20
s32i.n a10, a3, 24 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 24
s32i.n a10, a3, 28 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 28
addi.n a3, a3, 32 // increment dest_buff pointer by 32
._main_loop:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 16 bytes
bbci a11, 4, _mod_16_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
s32i.n a10, a3, 8 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 8
s32i.n a10, a3, 12 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 12
addi.n a3, a3, 16 // increment dest_buff pointer by 16
_mod_16_check:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 8 bytes
bbci a11, 3, _mod_8_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_mod_8_check:
// Check modulo 4 of the dest_w_bytes, if - then set 4 bytes
bbci a11, 2, _mod_4_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
addi.n a3, a3, 4 // increment dest_buff pointer by 4
_mod_4_check:
// Check modulo 2 of the dest_w_bytes, if - then set 2 bytes
bbci a11, 1, _mod_2_check // branch if 1-st bit of dest_w_bytes is clear
s16i a12, a3, 0 // save 16 bits from 16-bit color a12 to dest_buff a3, offset 0
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_mod_2_check:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
and a7, a8, a3 // a7 = dest_buff AND 0x3 (chck if the address is 4-byte aligned)
bnez a5, .outer_loop
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
Libraries/esp_lvgl_port/src/lvgl9/simd/lv_color_blend_to_rgb565_esp32s3.S
+405
View File
@@ -0,0 +1,405 @@
/*
* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
// This is LVGL RGB565 simple fill for ESP32S3 processor
.section .text
.align 4
.global lv_color_blend_to_rgb565_esp
.type lv_color_blend_to_rgb565_esp,@function
// The function implements the following C code:
// void lv_color_blend_to_rgb565(_lv_draw_sw_blend_fill_dsc_t * dsc);
// Input params
//
// dsc - a2
// typedef struct {
// uint32_t opa; l32i 0
// void * dst_buf; l32i 4
// uint32_t dst_w; l32i 8
// uint32_t dst_h; l32i 12
// uint32_t dst_stride; l32i 16
// const void * src_buf; l32i 20
// uint32_t src_stride; l32i 24
// const lv_opa_t * mask_buf; l32i 28
// uint32_t mask_stride; l32i 32
// } asm_dsc_t;
lv_color_blend_to_rgb565_esp:
entry a1, 32
ee.zero.q q0 // dummy TIE instruction, to enable the TIE
l32i.n a3, a2, 4 // a3 - dest_buff
l32i.n a4, a2, 8 // a4 - dest_w in uint16_t
l32i.n a5, a2, 12 // a5 - dest_h in uint16_t
l32i.n a6, a2, 16 // a6 - dest_stride in bytes
l32i.n a7, a2, 20 // a7 - src_buff (color)
l32i.n a8, a7, 0 // a8 - color as value
slli a11, a4, 1 // a11 - dest_w_bytes = sizeof(uint16_t) * dest_w
// Convert color to rgb656
l8ui a15, a7, 2 // red
movi.n a14, 0xf8
and a13, a15, a14
slli a10, a13, 8
l8ui a15, a7, 0 // blue
and a13, a15, a14
srli a12, a13, 3
add a10, a10, a12
l8ui a15, a7, 1 // green
movi.n a14, 0xfc
and a13, a15, a14
slli a12, a13, 3
add a12, a10, a12 // a12 = 16-bit color
slli a10, a12, 16
movi.n a13, 0xFFFF0000
and a10, a10, a13
or a10, a10, a12 // a10 = 32-bit color (16bit + (16bit << 16))
// Check for short lengths
// dest_w should be at least 16, othewise it's not worth using esp32s3 TIE
bgei a4, 16, _esp32s3_implementation // Branch if dest_w is greater than or equal to 16
j .lv_color_blend_to_rgb565_esp32_body // Jump to esp32 implementation
_esp32s3_implementation:
ee.movi.32.q q0, a10, 0 // fill q0 register from a10 by 32 bits
ee.movi.32.q q0, a10, 1
ee.movi.32.q q0, a10, 2
ee.movi.32.q q0, a10, 3
// Check dest_buff alignment
movi.n a7, 0xf // 0xf alignment mask (16-byte alignment)
and a15, a7, a3 // 16-byte alignment mask AND dest_buff pointer
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
// Check dest_stride alignment
and a15, a7, a6 // 16-byte alignment mask AND dest_stride
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
// Check dest_w_bytes alignment
and a15, a7, a11 // 16-byte alignment mask AND dest_w_bytes
bnez a15, _unaligned_by_4byte // branch if a15 not equals to zero
//**********************************************************************************************************************
// all aligned, the most ideal case
// dest_buff (a3) - 16-byte aligned
// dest_stride (a6) - 16-byte multiple
// dest_w (a4) - 16-byte multiple
srli a9, a4, 3 // a9 - loop_len = dest_w / 8
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
.outer_loop_aligned:
loopnez a9, ._main_loop_aligned // 16 bytes (8 rgb565) in one loop
ee.vst.128.ip q0, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_aligned:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
_unaligned_by_4byte:
// Check dest_buff alignment
movi.n a7, 0x3 // 0x3 alignment mask (4-byte alignment)
and a15, a7, a3 // 4-byte alignment mask AND dest_buff pointer
bnez a15, _unaligned_by_1byte // branch if a15 not equals to zero
// Check dest_stride alignment
and a15, a7, a6 // 4-byte alignment mask AND dest_stride pointer
bnez a15, _unaligned_by_1byte // branch if a15 not equals to zero
//**********************************************************************************************************************
// either dest_buff or dest_stride is not 16-byte aligned
// dest_w is always 4-byte multiple
// all of the following are 4-byte aligned
// dest_buff (a3) - 16-byte, or 4-byte aligned
// dest_stride (a6) - 16-byte, or 4-byte multiple
// dest_w (a4) - 4-byte multiple
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
movi.n a7, 0xf // 0xf alignment mask
.outer_loop_aligned_by_4byte:
// alignment check
and a15, a7, a3 // 0xf (alignment mask) AND dest_buff pointer
mov a12, a11 // a12 - local_dest_w_bytes = dest_w_bytes
beqz a15, _dest_buff_aligned_by_4byte // branch if a15 equals to zero
movi.n a14, 16 // a14 - 16
sub a15, a14, a15 // a15 = 16 - unalignment (lower 4 bits of dest_buff address)
sub a12, a12, a15 // local_dest_w_bytes = len - (16 - unalignment)
// keep setting until dest_buff is aligned
// Check modulo 8 of the unalignment, if - then set 8 bytes
bbci a15, 3, _aligning_mod_8_check_4byte // branch if 3-rd bit of unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 4 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_aligning_mod_8_check_4byte:
// Check modulo 4 of the unalignment, if - then set 4 bytes
bbci a15, 2, _aligning_mod_4_check_4byte // branch if 2-nd bit unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligning_mod_4_check_4byte:
_dest_buff_aligned_by_4byte:
// Calculate main loop_len
srli a9, a12, 4 // a9 - loop_len = local_dest_w_bytes / 16
// Main loop
loopnez a9, ._main_loop_unaligned_by_4byte // 16 bytes (8 rgb565) in one loop
ee.vst.128.ip q0, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_unaligned_by_4byte:
// Check modulo 8 of the dest_w, if - then set 8 bytes
bbci a12, 3, _aligned_mod_8_check_4byte // branch if 3-rd bit of local_dest_w_bytes a12 is clear
ee.vst.l.64.ip q0, a3, 8 // save lower 64 bits from q0 to dest_buff a3, increase dest_buff pointer by 8 bytes
_aligned_mod_8_check_4byte:
// Check modulo 4 of the dest_w, if - then set 4 bytes
bbci a12, 2, _aligned_mod_4_check_4byte // branch if 2-nd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligned_mod_4_check_4byte:
// Check modulo 2 of the dest_w, if - then set 2 bytes
bbci a12, 1, _aligned_mod_2_check_4byte // branch if 1-st bit of local_dest_w_bytes a12 is clear
s16i a10, a3, 0 // save 16 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_aligned_mod_2_check_4byte:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned_by_4byte
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
_unaligned_by_1byte:
//**********************************************************************************************************************
// either dest_buff or dest_stride is not 4-byte aligned
// dest_w is always 4-byte multiple
// dest_buff (a3) - 4-byte, or 1-byte aligned
// dest_stride (a6) - 4-byte, or 1-byte multiple
// dest_w (a4) - 4-byte multiple
ee.zero.q q1 // clear q1
ee.orq q1, q1, q0 // copy q0 to q1
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
movi.n a7, 0xf // 0xf alignment mask
.outer_loop_aligned_by_1byte:
// alignment check
and a15, a7, a3 // 0xf (alignment mask) AND dest_buff pointer
mov a12, a11 // a12 - local_dest_w_bytes = dest_w_bytes
beqz a15, _dest_buff_aligned_by_1byte // branch if a15 equals to zero
movi.n a14, 16 // a14 - 16
sub a15, a14, a15 // a15 = 16 - unalignment (lower 4 bits of dest_buff address)
sub a12, a12, a15 // local_dest_w_bytes = len - (16 - unalignment)
// keep setting until dest_buff is aligned
// Check modulo 8 of the unalignment, if - then set 8 bytes
bbci a15, 3, _aligning_mod_8_check_1byte// branch if 3-rd bit of unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 4 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_aligning_mod_8_check_1byte:
// Check modulo 4 of the unalignment, if - then set 4 bytes
bbci a15, 2, _aligning_mod_4_check_1byte // branch if 2-nd bit unalignment a15 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligning_mod_4_check_1byte:
// Check modulo 2 and 1
// modulo 2 and modulo 1 requires the same action
bbci a15, 1, _aligning_mod_2_check_1byte // branch if 1-st bit unalignment a15 is clear
s16i a10, a3, 0 // save 16 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_aligning_mod_2_check_1byte:
bbci a15, 0, _dest_buff_aligned_by_1byte // branch if 0-st bit unalignment a15 is clear
s16i a10, a3, 0 // save 16 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_dest_buff_aligned_by_1byte:
// Shift q reg, allowing to set 16-byte unaligned adata
wur.sar_byte a15 // apply unalignment to the SAR_BYTE
ee.src.q q2, q0, q1 // shift concat. of q0 and q1 to q2 by SAR_BYTE amount
// Calculate main loop_len
srli a9, a12, 4 // a9 - loop_len = local_dest_w_bytes / 16
// Main loop
loopnez a9, ._main_loop_unaligned_by_1byte // 16 bytes (8 rgb565) in one loop
ee.vst.128.ip q2, a3, 16 // store 16 bytes from q0 to dest_buff a3
._main_loop_unaligned_by_1byte:
// Firstly check mod 1 and mod 2 - correcting the aligned memory access
// Go back in one Byte, allow to correct after ee.vst.128.ip aligned access
addi a3, a3, -4
// Check modulo 2 of the dest_w, if - then set 2 bytes
// set SSSS in 0xSSSS0000
bbci a12, 1, _aligned_mod_2_check_1byte_corr // branch if 1-st bit of dest_w a12 is clear
srli a14, a10, 16 // shift a10 in 16, allowing s16i (saving of lower 16 bits)
s16i a14, a3, 2 // save 16 bits from a10 to dest_buff a3, offset 2 bytes
// Check modulo 1 of the dest_w, if - then set 1 byte
// additionally set SS in 0x0000SS00
bbci a12, 0, _aligned_end // branch if 0-th bit of dest_w a12 is clear
srli a14, a10, 8 // shift a10 in 8, allowing s8i
s8i a14, a3, 1 // save 8 bits from a10 to dest_buff a3, offset 1 byte
j _aligned_end
_aligned_mod_2_check_1byte_corr:
// Check modulo 1 of the dest_w, if - then set 1 byte
// set SS in 0xSS000000
bbci a12, 0, _aligned_end // branch if 0-th bit of dest_w a12 is clear
srli a14, a10, 24 // shift a10 in 24, allowing s8i (saving of lower 8 bits)
s8i a14, a3, 3 // save 8 bits from a10 to dest_buff a3, offset 3 bytes
_aligned_end:
addi a3, a3, 4 // Increase the pointer back, correction for addi a3, a3, -4
// Check modulo 8 of the dest_w, if - then set 8 bytes
bbci a12, 3, _aligned_mod_8_check_1byte // branch if 3-rd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
s32i.n a10, a3, 4 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 8 // increment dest_buff pointer by 4 bytes
_aligned_mod_8_check_1byte:
// Check modulo 4 of the dest_w, if - then set 4 bytes
bbci a12, 2, _aligned_mod_4_check_1byte // branch if 2-nd bit of local_dest_w_bytes a12 is clear
s32i.n a10, a3, 0 // save 32 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 4 // increment dest_buff pointer by 4 bytes
_aligned_mod_4_check_1byte:
// Check modulo 2 of the dest_w, if - then set 2 bytes
bbci a12, 1, _aligned_mod_2_check_1byte // branch if 1-st bit of local_dest_w_bytes a12 is clear
s16i a10, a3, 0 // save 16 bits from a10 to dest_buff a3, offset 0 bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_aligned_mod_2_check_1byte:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
bnez a5, .outer_loop_aligned_by_1byte
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return
.lv_color_blend_to_rgb565_esp32_body:
movi.n a8, 0x3 // a8 = 0x3, dest_buff align mask
sub a6, a6, a11 // dest_stride = dest_stride - dest_w_bytes
// cache init
// Prepare main loop length and dest_w_bytes
srli a9, a4, 4 // a9 = loop_len = dest_w / 8, calculate main loop_len for original dest_w
slli a11, a4, 1 // a11 = dest_w_bytes = sizeof(uint16_t) * dest_w
addi a4, a4, -1 // a4-- (decrement a4)
s32i.n a9, a1, 0 // cache.orig.loop_len
s32i.n a11, a1, 4 // cache.orig.dest_w_bytes
// Prepare decreased main loop length and dest_w_bytes
srli a9, a4, 4 // a9 = loop_len = dest_w / 8, calculate main loop_len for dest_w - 1
slli a11, a4, 1 // a11 = dest_w_bytes = sizeof(uint16_t) * (dest_w - 1)
s32i.n a9, a1, 8 // cache.decr.loop_len
s32i.n a11, a1, 12 // cache.decr.dest_w_bytes
and a7, a8, a3 // a7 = dest_buff AND 0x3 (chck if the address is 4-byte aligned)
.outer_loop:
// Check if the des_buff is 2-byte aligned
beqz a7, _dest_buff_2_byte_aligned // branch if a7 is equal to zero
s16i a12, a3, 0 // save 16 bits from 16-bit color a12 to dest_buff a3, offset 0
l32i.n a9, a1, 8 // a9 = load cache.decr.loop_len
l32i.n a11, a1, 12 // a11 = load cache.decr.dest_w_bytes
addi.n a3, a3, 2 // increment dest_buff pointer by 2
j _dest_buff_unaligned
_dest_buff_2_byte_aligned:
l32i.n a9, a1, 0 // a11 = load cache.orig.loop_len
l32i.n a11, a1, 4 // a11 = load cache.orig.dest_w_bytes
_dest_buff_unaligned:
// Run main loop which sets 16 bytes in one loop run
loopnez a9, ._main_loop
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
s32i.n a10, a3, 8 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 8
s32i.n a10, a3, 12 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 12
s32i.n a10, a3, 16 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 16
s32i.n a10, a3, 20 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 20
s32i.n a10, a3, 24 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 24
s32i.n a10, a3, 28 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 28
addi.n a3, a3, 32 // increment dest_buff pointer by 32
._main_loop:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 16 bytes
bbci a11, 4, _mod_16_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
s32i.n a10, a3, 8 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 8
s32i.n a10, a3, 12 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 12
addi.n a3, a3, 16 // increment dest_buff pointer by 16
_mod_16_check:
// Finish the remaining bytes out of the loop
// Check modulo 8 of the dest_w_bytes, if - then set 8 bytes
bbci a11, 3, _mod_8_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
s32i.n a10, a3, 4 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 4
addi.n a3, a3, 8 // increment dest_buff pointer by 8 bytes
_mod_8_check:
// Check modulo 4 of the dest_w_bytes, if - then set 4 bytes
bbci a11, 2, _mod_4_check // branch if 2-nd bit of dest_w_bytes is clear
s32i.n a10, a3, 0 // save 32 bits from 32-bit color a10 to dest_buff a3, offset 0
addi.n a3, a3, 4 // increment dest_buff pointer by 4
_mod_4_check:
// Check modulo 2 of the dest_w_bytes, if - then set 2 bytes
bbci a11, 1, _mod_2_check // branch if 1-st bit of dest_w_bytes is clear
s16i a12, a3, 0 // save 16 bits from 16-bit color a12 to dest_buff a3, offset 0
addi.n a3, a3, 2 // increment dest_buff pointer by 2 bytes
_mod_2_check:
add a3, a3, a6 // dest_buff + dest_stride
addi.n a5, a5, -1 // decrease the outer loop
and a7, a8, a3 // a7 = dest_buff AND 0x3 (chck if the address is 4-byte aligned)
bnez a5, .outer_loop
movi.n a2, 1 // return LV_RESULT_OK = 1
retw.n // return