summaryrefslogtreecommitdiffstats
path: root/src/video_core/engines/maxwell_dma.cpp
blob: a12a95ce27bb89b3271e80794236766fae37bcfa (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later

#include "common/algorithm.h"
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/settings.h"
#include "core/core.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/engines/maxwell_dma.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_base.h"
#include "video_core/textures/decoders.h"

MICROPROFILE_DECLARE(GPU_DMAEngine);
MICROPROFILE_DEFINE(GPU_DMAEngine, "GPU", "DMA Engine", MP_RGB(224, 224, 128));

namespace Tegra::Engines {

using namespace Texture;

MaxwellDMA::MaxwellDMA(Core::System& system_, MemoryManager& memory_manager_)
    : system{system_}, memory_manager{memory_manager_} {}

MaxwellDMA::~MaxwellDMA() = default;

void MaxwellDMA::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
    rasterizer = rasterizer_;
}

void MaxwellDMA::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
    ASSERT_MSG(method < NUM_REGS, "Invalid MaxwellDMA register");

    regs.reg_array[method] = method_argument;

    if (method == offsetof(Regs, launch_dma) / sizeof(u32)) {
        Launch();
    }
}

void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                 u32 methods_pending) {
    for (size_t i = 0; i < amount; ++i) {
        CallMethod(method, base_start[i], methods_pending - static_cast<u32>(i) <= 1);
    }
}

void MaxwellDMA::Launch() {
    MICROPROFILE_SCOPE(GPU_DMAEngine);
    LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in),
              static_cast<GPUVAddr>(regs.offset_out));

    // TODO(Subv): Perform more research and implement all features of this engine.
    const LaunchDMA& launch = regs.launch_dma;
    ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE);
    ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED);

    const bool is_src_pitch = launch.src_memory_layout == LaunchDMA::MemoryLayout::PITCH;
    const bool is_dst_pitch = launch.dst_memory_layout == LaunchDMA::MemoryLayout::PITCH;

    if (!is_src_pitch && !is_dst_pitch) {
        // If both the source and the destination are in block layout, assert.
        UNIMPLEMENTED_MSG("Tiled->Tiled DMA transfers are not yet implemented");
        return;
    }

    if (is_src_pitch && is_dst_pitch) {
        CopyPitchToPitch();
    } else {
        ASSERT(launch.multi_line_enable == 1);

        if (!is_src_pitch && is_dst_pitch) {
            CopyBlockLinearToPitch();
        } else {
            CopyPitchToBlockLinear();
        }
    }
    ReleaseSemaphore();
}

void MaxwellDMA::CopyPitchToPitch() {
    // When `multi_line_enable` bit is enabled we copy a 2D image of dimensions
    // (line_length_in, line_count).
    // Otherwise the copy is performed as if we were copying a 1D buffer of length line_length_in.
    const bool remap_enabled = regs.launch_dma.remap_enable != 0;
    if (regs.launch_dma.multi_line_enable) {
        UNIMPLEMENTED_IF(remap_enabled);

        // Perform a line-by-line copy.
        // We're going to take a subrect of size (line_length_in, line_count) from the source
        // rectangle. There is no need to manually flush/invalidate the regions because CopyBlock
        // does that for us.
        for (u32 line = 0; line < regs.line_count; ++line) {
            const GPUVAddr source_line = regs.offset_in + static_cast<size_t>(line) * regs.pitch_in;
            const GPUVAddr dest_line = regs.offset_out + static_cast<size_t>(line) * regs.pitch_out;
            memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in);
        }
        return;
    }
    // TODO: allow multisized components.
    auto& accelerate = rasterizer->AccessAccelerateDMA();
    const bool is_const_a_dst = regs.remap_const.dst_x == RemapConst::Swizzle::CONST_A;
    const bool is_buffer_clear = remap_enabled && is_const_a_dst;
    if (is_buffer_clear) {
        ASSERT(regs.remap_const.component_size_minus_one == 3);
        accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value);
        std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value);
        memory_manager.WriteBlockUnsafe(regs.offset_out, reinterpret_cast<u8*>(tmp_buffer.data()),
                                        regs.line_length_in * sizeof(u32));
        return;
    }
    UNIMPLEMENTED_IF(remap_enabled);
    if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) {
        std::vector<u8> tmp_buffer(regs.line_length_in);
        memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(), regs.line_length_in);
        memory_manager.WriteBlock(regs.offset_out, tmp_buffer.data(), regs.line_length_in);
    }
}

void MaxwellDMA::CopyBlockLinearToPitch() {
    UNIMPLEMENTED_IF(regs.src_params.block_size.width != 0);
    UNIMPLEMENTED_IF(regs.src_params.layer != 0);

    const bool is_remapping = regs.launch_dma.remap_enable != 0;

    // Optimized path for micro copies.
    const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
    if (!is_remapping && dst_size < GOB_SIZE && regs.pitch_out <= GOB_SIZE_X &&
        regs.src_params.height > GOB_SIZE_Y) {
        FastCopyBlockLinearToPitch();
        return;
    }

    // Deswizzle the input and copy it over.
    UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0);
    const Parameters& src_params = regs.src_params;

    const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1;
    const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1;

    const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size;

    u32 width = src_params.width;
    u32 x_elements = regs.line_length_in;
    u32 x_offset = src_params.origin.x;
    u32 bpp_shift = 0U;
    if (!is_remapping) {
        bpp_shift = Common::FoldRight(
            4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); },
            width, x_elements, x_offset, static_cast<u32>(regs.offset_in));
        width >>= bpp_shift;
        x_elements >>= bpp_shift;
        x_offset >>= bpp_shift;
    }

    const u32 bytes_per_pixel = base_bpp << bpp_shift;
    const u32 height = src_params.height;
    const u32 depth = src_params.depth;
    const u32 block_height = src_params.block_size.height;
    const u32 block_depth = src_params.block_size.depth;
    const size_t src_size =
        CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);

    if (read_buffer.size() < src_size) {
        read_buffer.resize(src_size);
    }
    if (write_buffer.size() < dst_size) {
        write_buffer.resize(dst_size);
    }

    memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
    memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);

    UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset,
                     src_params.origin.y, x_elements, regs.line_count, block_height, block_depth,
                     regs.pitch_out);

    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
}

void MaxwellDMA::CopyPitchToBlockLinear() {
    UNIMPLEMENTED_IF_MSG(regs.dst_params.block_size.width != 0, "Block width is not one");
    UNIMPLEMENTED_IF(regs.dst_params.layer != 0);
    UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0);

    const bool is_remapping = regs.launch_dma.remap_enable != 0;
    const u32 num_remap_components = regs.remap_const.num_dst_components_minus_one + 1;
    const u32 remap_components_size = regs.remap_const.component_size_minus_one + 1;

    const auto& dst_params = regs.dst_params;

    const u32 base_bpp = !is_remapping ? 1U : num_remap_components * remap_components_size;

    u32 width = dst_params.width;
    u32 x_elements = regs.line_length_in;
    u32 x_offset = dst_params.origin.x;
    u32 bpp_shift = 0U;
    if (!is_remapping) {
        bpp_shift = Common::FoldRight(
            4U, [](u32 x, u32 y) { return std::min(x, static_cast<u32>(std::countr_zero(y))); },
            width, x_elements, x_offset, static_cast<u32>(regs.offset_out));
        width >>= bpp_shift;
        x_elements >>= bpp_shift;
        x_offset >>= bpp_shift;
    }

    const u32 bytes_per_pixel = base_bpp << bpp_shift;
    const u32 height = dst_params.height;
    const u32 depth = dst_params.depth;
    const u32 block_height = dst_params.block_size.height;
    const u32 block_depth = dst_params.block_size.depth;
    const size_t dst_size =
        CalculateSize(true, bytes_per_pixel, width, height, depth, block_height, block_depth);
    const size_t src_size = static_cast<size_t>(regs.pitch_in) * regs.line_count;

    if (read_buffer.size() < src_size) {
        read_buffer.resize(src_size);
    }
    if (write_buffer.size() < dst_size) {
        write_buffer.resize(dst_size);
    }

    memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
    if (Settings::IsGPULevelExtreme()) {
        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
    } else {
        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
    }

    // If the input is linear and the output is tiled, swizzle the input and copy it over.
    SwizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, width, height, depth, x_offset,
                   dst_params.origin.y, x_elements, regs.line_count, block_height, block_depth,
                   regs.pitch_in);

    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
}

void MaxwellDMA::FastCopyBlockLinearToPitch() {
    const u32 bytes_per_pixel = 1U;
    const size_t src_size = GOB_SIZE;
    const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
    u32 pos_x = regs.src_params.origin.x;
    u32 pos_y = regs.src_params.origin.y;
    const u64 offset = GetGOBOffset(regs.src_params.width, regs.src_params.height, pos_x, pos_y,
                                    regs.src_params.block_size.height, bytes_per_pixel);
    const u32 x_in_gob = 64 / bytes_per_pixel;
    pos_x = pos_x % x_in_gob;
    pos_y = pos_y % 8;

    if (read_buffer.size() < src_size) {
        read_buffer.resize(src_size);
    }
    if (write_buffer.size() < dst_size) {
        write_buffer.resize(dst_size);
    }

    if (Settings::IsGPULevelExtreme()) {
        memory_manager.ReadBlock(regs.offset_in + offset, read_buffer.data(), src_size);
        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
    } else {
        memory_manager.ReadBlockUnsafe(regs.offset_in + offset, read_buffer.data(), src_size);
        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
    }

    UnswizzleSubrect(write_buffer, read_buffer, bytes_per_pixel, regs.src_params.width,
                     regs.src_params.height, 1, pos_x, pos_y, regs.line_length_in, regs.line_count,
                     regs.src_params.block_size.height, regs.src_params.block_size.depth,
                     regs.pitch_out);

    memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
}

void MaxwellDMA::ReleaseSemaphore() {
    const auto type = regs.launch_dma.semaphore_type;
    const GPUVAddr address = regs.semaphore.address;
    switch (type) {
    case LaunchDMA::SemaphoreType::NONE:
        break;
    case LaunchDMA::SemaphoreType::RELEASE_ONE_WORD_SEMAPHORE:
        memory_manager.Write<u32>(address, regs.semaphore.payload);
        break;
    case LaunchDMA::SemaphoreType::RELEASE_FOUR_WORD_SEMAPHORE:
        memory_manager.Write<u64>(address, static_cast<u64>(regs.semaphore.payload));
        memory_manager.Write<u64>(address + 8, system.GPU().GetTicks());
        break;
    default:
        ASSERT_MSG(false, "Unknown semaphore type: {}", static_cast<u32>(type.Value()));
    }
}

} // namespace Tegra::Engines