GNU Linux-libre 5.10.219-gnu1
[releases.git] / drivers / staging / media / hantro / hantro_h1_jpeg_enc.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Hantro VPU codec driver
4  *
5  * Copyright (C) 2018 Rockchip Electronics Co., Ltd.
6  */
7
8 #include <asm/unaligned.h>
9 #include <media/v4l2-mem2mem.h>
10 #include "hantro_jpeg.h"
11 #include "hantro.h"
12 #include "hantro_v4l2.h"
13 #include "hantro_hw.h"
14 #include "hantro_h1_regs.h"
15
16 #define H1_JPEG_QUANT_TABLE_COUNT 16
17
18 static void hantro_h1_set_src_img_ctrl(struct hantro_dev *vpu,
19                                        struct hantro_ctx *ctx)
20 {
21         struct v4l2_pix_format_mplane *pix_fmt = &ctx->src_fmt;
22         u32 reg;
23
24         reg = H1_REG_IN_IMG_CTRL_ROW_LEN(pix_fmt->width)
25                 | H1_REG_IN_IMG_CTRL_OVRFLR_D4(0)
26                 | H1_REG_IN_IMG_CTRL_OVRFLB(0)
27                 | H1_REG_IN_IMG_CTRL_FMT(ctx->vpu_src_fmt->enc_fmt);
28         vepu_write_relaxed(vpu, reg, H1_REG_IN_IMG_CTRL);
29 }
30
31 static void hantro_h1_jpeg_enc_set_buffers(struct hantro_dev *vpu,
32                                            struct hantro_ctx *ctx,
33                                            struct vb2_buffer *src_buf)
34 {
35         struct v4l2_pix_format_mplane *pix_fmt = &ctx->src_fmt;
36         dma_addr_t src[3];
37
38         WARN_ON(pix_fmt->num_planes > 3);
39
40         vepu_write_relaxed(vpu, ctx->jpeg_enc.bounce_buffer.dma,
41                            H1_REG_ADDR_OUTPUT_STREAM);
42         vepu_write_relaxed(vpu, ctx->jpeg_enc.bounce_buffer.size,
43                            H1_REG_STR_BUF_LIMIT);
44
45         if (pix_fmt->num_planes == 1) {
46                 src[0] = vb2_dma_contig_plane_dma_addr(src_buf, 0);
47                 /* single plane formats we supported are all interlaced */
48                 vepu_write_relaxed(vpu, src[0], H1_REG_ADDR_IN_PLANE_0);
49         } else if (pix_fmt->num_planes == 2) {
50                 src[0] = vb2_dma_contig_plane_dma_addr(src_buf, 0);
51                 src[1] = vb2_dma_contig_plane_dma_addr(src_buf, 1);
52                 vepu_write_relaxed(vpu, src[0], H1_REG_ADDR_IN_PLANE_0);
53                 vepu_write_relaxed(vpu, src[1], H1_REG_ADDR_IN_PLANE_1);
54         } else {
55                 src[0] = vb2_dma_contig_plane_dma_addr(src_buf, 0);
56                 src[1] = vb2_dma_contig_plane_dma_addr(src_buf, 1);
57                 src[2] = vb2_dma_contig_plane_dma_addr(src_buf, 2);
58                 vepu_write_relaxed(vpu, src[0], H1_REG_ADDR_IN_PLANE_0);
59                 vepu_write_relaxed(vpu, src[1], H1_REG_ADDR_IN_PLANE_1);
60                 vepu_write_relaxed(vpu, src[2], H1_REG_ADDR_IN_PLANE_2);
61         }
62 }
63
64 static void
65 hantro_h1_jpeg_enc_set_qtable(struct hantro_dev *vpu,
66                               unsigned char *luma_qtable,
67                               unsigned char *chroma_qtable)
68 {
69         u32 reg, i;
70         __be32 *luma_qtable_p;
71         __be32 *chroma_qtable_p;
72
73         luma_qtable_p = (__be32 *)luma_qtable;
74         chroma_qtable_p = (__be32 *)chroma_qtable;
75
76         /*
77          * Quantization table registers must be written in contiguous blocks.
78          * DO NOT collapse the below two "for" loops into one.
79          */
80         for (i = 0; i < H1_JPEG_QUANT_TABLE_COUNT; i++) {
81                 reg = get_unaligned_be32(&luma_qtable_p[i]);
82                 vepu_write_relaxed(vpu, reg, H1_REG_JPEG_LUMA_QUAT(i));
83         }
84
85         for (i = 0; i < H1_JPEG_QUANT_TABLE_COUNT; i++) {
86                 reg = get_unaligned_be32(&chroma_qtable_p[i]);
87                 vepu_write_relaxed(vpu, reg, H1_REG_JPEG_CHROMA_QUAT(i));
88         }
89 }
90
91 void hantro_h1_jpeg_enc_run(struct hantro_ctx *ctx)
92 {
93         struct hantro_dev *vpu = ctx->dev;
94         struct vb2_v4l2_buffer *src_buf, *dst_buf;
95         struct hantro_jpeg_ctx jpeg_ctx;
96         u32 reg;
97
98         src_buf = hantro_get_src_buf(ctx);
99         dst_buf = hantro_get_dst_buf(ctx);
100
101         hantro_start_prepare_run(ctx);
102
103         memset(&jpeg_ctx, 0, sizeof(jpeg_ctx));
104         jpeg_ctx.buffer = vb2_plane_vaddr(&dst_buf->vb2_buf, 0);
105         jpeg_ctx.width = ctx->dst_fmt.width;
106         jpeg_ctx.height = ctx->dst_fmt.height;
107         jpeg_ctx.quality = ctx->jpeg_quality;
108         hantro_jpeg_header_assemble(&jpeg_ctx);
109
110         /* Switch to JPEG encoder mode before writing registers */
111         vepu_write_relaxed(vpu, H1_REG_ENC_CTRL_ENC_MODE_JPEG,
112                            H1_REG_ENC_CTRL);
113
114         hantro_h1_set_src_img_ctrl(vpu, ctx);
115         hantro_h1_jpeg_enc_set_buffers(vpu, ctx, &src_buf->vb2_buf);
116         hantro_h1_jpeg_enc_set_qtable(vpu,
117                                       hantro_jpeg_get_qtable(0),
118                                       hantro_jpeg_get_qtable(1));
119
120         reg = H1_REG_AXI_CTRL_OUTPUT_SWAP16
121                 | H1_REG_AXI_CTRL_INPUT_SWAP16
122                 | H1_REG_AXI_CTRL_BURST_LEN(16)
123                 | H1_REG_AXI_CTRL_OUTPUT_SWAP32
124                 | H1_REG_AXI_CTRL_INPUT_SWAP32
125                 | H1_REG_AXI_CTRL_OUTPUT_SWAP8
126                 | H1_REG_AXI_CTRL_INPUT_SWAP8;
127         /* Make sure that all registers are written at this point. */
128         vepu_write(vpu, reg, H1_REG_AXI_CTRL);
129
130         reg = H1_REG_ENC_CTRL_WIDTH(MB_WIDTH(ctx->src_fmt.width))
131                 | H1_REG_ENC_CTRL_HEIGHT(MB_HEIGHT(ctx->src_fmt.height))
132                 | H1_REG_ENC_CTRL_ENC_MODE_JPEG
133                 | H1_REG_ENC_PIC_INTRA
134                 | H1_REG_ENC_CTRL_EN_BIT;
135
136         hantro_end_prepare_run(ctx);
137
138         vepu_write(vpu, reg, H1_REG_ENC_CTRL);
139 }
140
141 void hantro_jpeg_enc_done(struct hantro_ctx *ctx)
142 {
143         struct hantro_dev *vpu = ctx->dev;
144         u32 bytesused = vepu_read(vpu, H1_REG_STR_BUF_LIMIT) / 8;
145         struct vb2_v4l2_buffer *dst_buf = hantro_get_dst_buf(ctx);
146
147         /*
148          * TODO: Rework the JPEG encoder to eliminate the need
149          * for a bounce buffer.
150          */
151         memcpy(vb2_plane_vaddr(&dst_buf->vb2_buf, 0) +
152                ctx->vpu_dst_fmt->header_size,
153                ctx->jpeg_enc.bounce_buffer.cpu, bytesused);
154         vb2_set_plane_payload(&dst_buf->vb2_buf, 0,
155                               ctx->vpu_dst_fmt->header_size + bytesused);
156 }