diff options
Diffstat (limited to '')
| -rw-r--r-- | video/out/d3d11/context.c | 566 | ||||
| -rw-r--r-- | video/out/d3d11/context.h | 9 | ||||
| -rw-r--r-- | video/out/d3d11/hwdec_d3d11va.c | 258 | ||||
| -rw-r--r-- | video/out/d3d11/hwdec_dxva2dxgi.c | 478 | ||||
| -rw-r--r-- | video/out/d3d11/ra_d3d11.c | 2544 | ||||
| -rw-r--r-- | video/out/d3d11/ra_d3d11.h | 47 |
6 files changed, 3902 insertions, 0 deletions
diff --git a/video/out/d3d11/context.c b/video/out/d3d11/context.c new file mode 100644 index 0000000..05f04fd --- /dev/null +++ b/video/out/d3d11/context.c @@ -0,0 +1,566 @@ +/* + * This file is part of mpv. + * + * mpv is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * mpv is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with mpv. If not, see <http://www.gnu.org/licenses/>. + */ + +#include "common/msg.h" +#include "options/m_config.h" +#include "osdep/timer.h" +#include "osdep/windows_utils.h" + +#include "video/out/gpu/context.h" +#include "video/out/gpu/d3d11_helpers.h" +#include "video/out/gpu/spirv.h" +#include "video/out/w32_common.h" +#include "context.h" +#include "ra_d3d11.h" + +static int d3d11_validate_adapter(struct mp_log *log, + const struct m_option *opt, + struct bstr name, const char **value); + +struct d3d11_opts { + int feature_level; + int warp; + bool flip; + int sync_interval; + char *adapter_name; + int output_format; + int color_space; + bool exclusive_fs; +}; + +#define OPT_BASE_STRUCT struct d3d11_opts +const struct m_sub_options d3d11_conf = { + .opts = (const struct m_option[]) { + {"d3d11-warp", OPT_CHOICE(warp, + {"auto", -1}, + {"no", 0}, + {"yes", 1})}, + {"d3d11-feature-level", OPT_CHOICE(feature_level, + {"12_1", D3D_FEATURE_LEVEL_12_1}, + {"12_0", D3D_FEATURE_LEVEL_12_0}, + {"11_1", D3D_FEATURE_LEVEL_11_1}, + {"11_0", D3D_FEATURE_LEVEL_11_0}, + {"10_1", D3D_FEATURE_LEVEL_10_1}, + {"10_0", D3D_FEATURE_LEVEL_10_0}, + {"9_3", D3D_FEATURE_LEVEL_9_3}, + {"9_2", D3D_FEATURE_LEVEL_9_2}, + {"9_1", D3D_FEATURE_LEVEL_9_1})}, + {"d3d11-flip", OPT_BOOL(flip)}, + {"d3d11-sync-interval", OPT_INT(sync_interval), M_RANGE(0, 4)}, + {"d3d11-adapter", OPT_STRING_VALIDATE(adapter_name, + d3d11_validate_adapter)}, + {"d3d11-output-format", OPT_CHOICE(output_format, + {"auto", DXGI_FORMAT_UNKNOWN}, + {"rgba8", DXGI_FORMAT_R8G8B8A8_UNORM}, + {"bgra8", DXGI_FORMAT_B8G8R8A8_UNORM}, + {"rgb10_a2", DXGI_FORMAT_R10G10B10A2_UNORM}, + {"rgba16f", DXGI_FORMAT_R16G16B16A16_FLOAT})}, + {"d3d11-output-csp", OPT_CHOICE(color_space, + {"auto", -1}, + {"srgb", DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P709}, + {"linear", DXGI_COLOR_SPACE_RGB_FULL_G10_NONE_P709}, + {"pq", DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020}, + {"bt.2020", DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P2020})}, + {"d3d11-exclusive-fs", OPT_BOOL(exclusive_fs)}, + {0} + }, + .defaults = &(const struct d3d11_opts) { + .feature_level = D3D_FEATURE_LEVEL_12_1, + .warp = -1, + .flip = true, + .sync_interval = 1, + .adapter_name = NULL, + .output_format = DXGI_FORMAT_UNKNOWN, + .color_space = -1, + }, + .size = sizeof(struct d3d11_opts) +}; + +struct priv { + struct d3d11_opts *opts; + struct m_config_cache *opts_cache; + + struct mp_vo_opts *vo_opts; + struct m_config_cache *vo_opts_cache; + + struct ra_tex *backbuffer; + ID3D11Device *device; + IDXGISwapChain *swapchain; + struct mp_colorspace swapchain_csp; + + int64_t perf_freq; + unsigned sync_refresh_count; + int64_t sync_qpc_time; + int64_t vsync_duration_qpc; + int64_t last_submit_qpc; +}; + +static int d3d11_validate_adapter(struct mp_log *log, + const struct m_option *opt, + struct bstr name, const char **value) +{ + struct bstr param = bstr0(*value); + bool help = bstr_equals0(param, "help"); + bool adapter_matched = false; + struct bstr listing = { 0 }; + + if (bstr_equals0(param, "")) { + return 0; + } + + adapter_matched = mp_d3d11_list_or_verify_adapters(log, + help ? bstr0(NULL) : param, + help ? &listing : NULL); + + if (help) { + mp_info(log, "Available D3D11 adapters:\n%.*s", + BSTR_P(listing)); + talloc_free(listing.start); + return M_OPT_EXIT; + } + + if (!adapter_matched) { + mp_err(log, "No adapter matching '%.*s'!\n", BSTR_P(param)); + } + + return adapter_matched ? 0 : M_OPT_INVALID; +} + +static struct ra_tex *get_backbuffer(struct ra_ctx *ctx) +{ + struct priv *p = ctx->priv; + ID3D11Texture2D *backbuffer = NULL; + struct ra_tex *tex = NULL; + HRESULT hr; + + hr = IDXGISwapChain_GetBuffer(p->swapchain, 0, &IID_ID3D11Texture2D, + (void**)&backbuffer); + if (FAILED(hr)) { + MP_ERR(ctx, "Couldn't get swapchain image\n"); + goto done; + } + + tex = ra_d3d11_wrap_tex(ctx->ra, (ID3D11Resource *)backbuffer); +done: + SAFE_RELEASE(backbuffer); + return tex; +} + +static bool resize(struct ra_ctx *ctx) +{ + struct priv *p = ctx->priv; + HRESULT hr; + + if (p->backbuffer) { + MP_ERR(ctx, "Attempt at resizing while a frame was in progress!\n"); + return false; + } + + hr = IDXGISwapChain_ResizeBuffers(p->swapchain, 0, ctx->vo->dwidth, + ctx->vo->dheight, DXGI_FORMAT_UNKNOWN, 0); + if (FAILED(hr)) { + MP_FATAL(ctx, "Couldn't resize swapchain: %s\n", mp_HRESULT_to_str(hr)); + return false; + } + + return true; +} + +static bool d3d11_reconfig(struct ra_ctx *ctx) +{ + vo_w32_config(ctx->vo); + return resize(ctx); +} + +static int d3d11_color_depth(struct ra_swapchain *sw) +{ + struct priv *p = sw->priv; + DXGI_SWAP_CHAIN_DESC desc; + + HRESULT hr = IDXGISwapChain_GetDesc(p->swapchain, &desc); + if (FAILED(hr)) { + MP_ERR(sw->ctx, "Failed to query swap chain description: %s!\n", + mp_HRESULT_to_str(hr)); + return 0; + } + + const struct ra_format *ra_fmt = + ra_d3d11_get_ra_format(sw->ctx->ra, desc.BufferDesc.Format); + if (!ra_fmt) + return 0; + + return ra_fmt->component_depth[0]; +} + +static bool d3d11_start_frame(struct ra_swapchain *sw, struct ra_fbo *out_fbo) +{ + struct priv *p = sw->priv; + + if (!out_fbo) + return true; + + assert(!p->backbuffer); + + p->backbuffer = get_backbuffer(sw->ctx); + if (!p->backbuffer) + return false; + + *out_fbo = (struct ra_fbo) { + .tex = p->backbuffer, + .flip = false, + .color_space = p->swapchain_csp + }; + return true; +} + +static bool d3d11_submit_frame(struct ra_swapchain *sw, + const struct vo_frame *frame) +{ + struct priv *p = sw->priv; + + ra_d3d11_flush(sw->ctx->ra); + ra_tex_free(sw->ctx->ra, &p->backbuffer); + return true; +} + +static int64_t qpc_to_ns(struct ra_swapchain *sw, int64_t qpc) +{ + struct priv *p = sw->priv; + + // Convert QPC units (1/perf_freq seconds) to nanoseconds. This will work + // without overflow because the QPC value is guaranteed not to roll-over + // within 100 years, so perf_freq must be less than 2.9*10^9. + return qpc / p->perf_freq * INT64_C(1000000000) + + qpc % p->perf_freq * INT64_C(1000000000) / p->perf_freq; +} + +static int64_t qpc_ns_now(struct ra_swapchain *sw) +{ + LARGE_INTEGER perf_count; + QueryPerformanceCounter(&perf_count); + return qpc_to_ns(sw, perf_count.QuadPart); +} + +static void d3d11_swap_buffers(struct ra_swapchain *sw) +{ + struct priv *p = sw->priv; + + m_config_cache_update(p->opts_cache); + + LARGE_INTEGER perf_count; + QueryPerformanceCounter(&perf_count); + p->last_submit_qpc = perf_count.QuadPart; + + IDXGISwapChain_Present(p->swapchain, p->opts->sync_interval, 0); +} + +static void d3d11_get_vsync(struct ra_swapchain *sw, struct vo_vsync_info *info) +{ + struct priv *p = sw->priv; + HRESULT hr; + + m_config_cache_update(p->opts_cache); + + // The calculations below are only valid if mpv presents on every vsync + if (p->opts->sync_interval != 1) + return; + + // They're also only valid for flip model swapchains + DXGI_SWAP_CHAIN_DESC desc; + hr = IDXGISwapChain_GetDesc(p->swapchain, &desc); + if (FAILED(hr) || (desc.SwapEffect != DXGI_SWAP_EFFECT_FLIP_SEQUENTIAL && + desc.SwapEffect != DXGI_SWAP_EFFECT_FLIP_DISCARD)) + { + return; + } + + // GetLastPresentCount returns a sequential ID for the frame submitted by + // the last call to IDXGISwapChain::Present() + UINT submit_count; + hr = IDXGISwapChain_GetLastPresentCount(p->swapchain, &submit_count); + if (FAILED(hr)) + return; + + // GetFrameStatistics returns two pairs. The first is (PresentCount, + // PresentRefreshCount) which relates a present ID (on the same timeline as + // GetLastPresentCount) to the physical vsync it was displayed on. The + // second is (SyncRefreshCount, SyncQPCTime), which relates a physical vsync + // to a timestamp on the same clock as QueryPerformanceCounter. + DXGI_FRAME_STATISTICS stats; + hr = IDXGISwapChain_GetFrameStatistics(p->swapchain, &stats); + if (hr == DXGI_ERROR_FRAME_STATISTICS_DISJOINT) { + p->sync_refresh_count = 0; + p->sync_qpc_time = 0; + } + if (FAILED(hr)) + return; + + info->last_queue_display_time = 0; + info->vsync_duration = 0; + // Detecting skipped vsyncs is possible but not supported yet + info->skipped_vsyncs = -1; + + // Get the number of physical vsyncs that have passed since the start of the + // playback or disjoint event. + // Check for 0 here, since sometimes GetFrameStatistics returns S_OK but + // with 0s in some (all?) members of DXGI_FRAME_STATISTICS. + unsigned src_passed = 0; + if (stats.SyncRefreshCount && p->sync_refresh_count) + src_passed = stats.SyncRefreshCount - p->sync_refresh_count; + if (p->sync_refresh_count == 0) + p->sync_refresh_count = stats.SyncRefreshCount; + + // Get the elapsed time passed between the above vsyncs + unsigned sqt_passed = 0; + if (stats.SyncQPCTime.QuadPart && p->sync_qpc_time) + sqt_passed = stats.SyncQPCTime.QuadPart - p->sync_qpc_time; + if (p->sync_qpc_time == 0) + p->sync_qpc_time = stats.SyncQPCTime.QuadPart; + + // If any vsyncs have passed, estimate the physical frame rate + if (src_passed && sqt_passed) + p->vsync_duration_qpc = sqt_passed / src_passed; + if (p->vsync_duration_qpc) + info->vsync_duration = qpc_to_ns(sw, p->vsync_duration_qpc); + + // If the physical frame rate is known and the other members of + // DXGI_FRAME_STATISTICS are non-0, estimate the timing of the next frame + if (p->vsync_duration_qpc && stats.PresentCount && + stats.PresentRefreshCount && stats.SyncRefreshCount && + stats.SyncQPCTime.QuadPart) + { + // It's not clear if PresentRefreshCount and SyncRefreshCount can refer + // to different frames, but in case they can, assuming mpv presents on + // every frame, guess the present count that relates to SyncRefreshCount. + unsigned expected_sync_pc = stats.PresentCount + + (stats.SyncRefreshCount - stats.PresentRefreshCount); + + // Now guess the timestamp of the last submitted frame based on the + // timestamp of the frame at SyncRefreshCount and the frame rate + int queued_frames = submit_count - expected_sync_pc; + int64_t last_queue_display_time_qpc = stats.SyncQPCTime.QuadPart + + queued_frames * p->vsync_duration_qpc; + + // Only set the estimated display time if it's after the last submission + // time. It could be before if mpv skips a lot of frames. + if (last_queue_display_time_qpc >= p->last_submit_qpc) { + info->last_queue_display_time = mp_time_ns() + + (qpc_to_ns(sw, last_queue_display_time_qpc) - qpc_ns_now(sw)); + } + } +} + +static bool d3d11_set_fullscreen(struct ra_ctx *ctx) +{ + struct priv *p = ctx->priv; + HRESULT hr; + + m_config_cache_update(p->opts_cache); + + if (!p->swapchain) { + MP_ERR(ctx, "Full screen configuration was requested before D3D11 " + "swap chain was ready!"); + return false; + } + + // we only want exclusive FS if we are entering FS and + // exclusive FS is enabled. Otherwise disable exclusive FS. + bool enable_exclusive_fs = p->vo_opts->fullscreen && + p->opts->exclusive_fs; + + MP_VERBOSE(ctx, "%s full-screen exclusive mode while %s fullscreen\n", + enable_exclusive_fs ? "Enabling" : "Disabling", + ctx->vo->opts->fullscreen ? "entering" : "leaving"); + + hr = IDXGISwapChain_SetFullscreenState(p->swapchain, + enable_exclusive_fs, NULL); + if (FAILED(hr)) + return false; + + if (!resize(ctx)) + return false; + + return true; +} + +static int d3d11_control(struct ra_ctx *ctx, int *events, int request, void *arg) +{ + struct priv *p = ctx->priv; + int ret = -1; + bool fullscreen_switch_needed = false; + + switch (request) { + case VOCTRL_VO_OPTS_CHANGED: { + void *changed_option; + + while (m_config_cache_get_next_changed(p->vo_opts_cache, + &changed_option)) + { + struct mp_vo_opts *vo_opts = p->vo_opts_cache->opts; + + if (changed_option == &vo_opts->fullscreen) { + fullscreen_switch_needed = true; + } + } + + break; + } + default: + break; + } + + // if leaving full screen, handle d3d11 stuff first, then general + // windowing + if (fullscreen_switch_needed && !p->vo_opts->fullscreen) { + if (!d3d11_set_fullscreen(ctx)) + return VO_FALSE; + + fullscreen_switch_needed = false; + } + + ret = vo_w32_control(ctx->vo, events, request, arg); + + // if entering full screen, handle d3d11 after general windowing stuff + if (fullscreen_switch_needed && p->vo_opts->fullscreen) { + if (!d3d11_set_fullscreen(ctx)) + return VO_FALSE; + + fullscreen_switch_needed = false; + } + + if (*events & VO_EVENT_RESIZE) { + if (!resize(ctx)) + return VO_ERROR; + } + return ret; +} + +static void d3d11_uninit(struct ra_ctx *ctx) +{ + struct priv *p = ctx->priv; + + if (p->swapchain) + IDXGISwapChain_SetFullscreenState(p->swapchain, FALSE, NULL); + + if (ctx->ra) + ra_tex_free(ctx->ra, &p->backbuffer); + SAFE_RELEASE(p->swapchain); + vo_w32_uninit(ctx->vo); + SAFE_RELEASE(p->device); + + // Destroy the RA last to prevent objects we hold from showing up in D3D's + // leak checker + if (ctx->ra) + ctx->ra->fns->destroy(ctx->ra); +} + +static const struct ra_swapchain_fns d3d11_swapchain = { + .color_depth = d3d11_color_depth, + .start_frame = d3d11_start_frame, + .submit_frame = d3d11_submit_frame, + .swap_buffers = d3d11_swap_buffers, + .get_vsync = d3d11_get_vsync, +}; + +static bool d3d11_init(struct ra_ctx *ctx) +{ + struct priv *p = ctx->priv = talloc_zero(ctx, struct priv); + p->opts_cache = m_config_cache_alloc(ctx, ctx->global, &d3d11_conf); + p->opts = p->opts_cache->opts; + + p->vo_opts_cache = m_config_cache_alloc(ctx, ctx->vo->global, &vo_sub_opts); + p->vo_opts = p->vo_opts_cache->opts; + + LARGE_INTEGER perf_freq; + QueryPerformanceFrequency(&perf_freq); + p->perf_freq = perf_freq.QuadPart; + + struct ra_swapchain *sw = ctx->swapchain = talloc_zero(ctx, struct ra_swapchain); + sw->priv = p; + sw->ctx = ctx; + sw->fns = &d3d11_swapchain; + + struct d3d11_device_opts dopts = { + .debug = ctx->opts.debug, + .allow_warp = p->opts->warp != 0, + .force_warp = p->opts->warp == 1, + .max_feature_level = p->opts->feature_level, + .max_frame_latency = ctx->vo->opts->swapchain_depth, + .adapter_name = p->opts->adapter_name, + }; + if (!mp_d3d11_create_present_device(ctx->log, &dopts, &p->device)) + goto error; + + if (!spirv_compiler_init(ctx)) + goto error; + ctx->ra = ra_d3d11_create(p->device, ctx->log, ctx->spirv); + if (!ctx->ra) + goto error; + + if (!vo_w32_init(ctx->vo)) + goto error; + + UINT usage = DXGI_USAGE_RENDER_TARGET_OUTPUT | DXGI_USAGE_SHADER_INPUT; + if (ID3D11Device_GetFeatureLevel(p->device) >= D3D_FEATURE_LEVEL_11_0 && + p->opts->output_format != DXGI_FORMAT_B8G8R8A8_UNORM) + { + usage |= DXGI_USAGE_UNORDERED_ACCESS; + } + + struct d3d11_swapchain_opts scopts = { + .window = vo_w32_hwnd(ctx->vo), + .width = ctx->vo->dwidth, + .height = ctx->vo->dheight, + .format = p->opts->output_format, + .color_space = p->opts->color_space, + .configured_csp = &p->swapchain_csp, + .flip = p->opts->flip, + // Add one frame for the backbuffer and one frame of "slack" to reduce + // contention with the window manager when acquiring the backbuffer + .length = ctx->vo->opts->swapchain_depth + 2, + .usage = usage, + }; + if (!mp_d3d11_create_swapchain(p->device, ctx->log, &scopts, &p->swapchain)) + goto error; + + return true; + +error: + d3d11_uninit(ctx); + return false; +} + +IDXGISwapChain *ra_d3d11_ctx_get_swapchain(struct ra_ctx *ra) +{ + if (ra->swapchain->fns != &d3d11_swapchain) + return NULL; + + struct priv *p = ra->priv; + + IDXGISwapChain_AddRef(p->swapchain); + + return p->swapchain; +} + +const struct ra_ctx_fns ra_ctx_d3d11 = { + .type = "d3d11", + .name = "d3d11", + .reconfig = d3d11_reconfig, + .control = d3d11_control, + .init = d3d11_init, + .uninit = d3d11_uninit, +}; diff --git a/video/out/d3d11/context.h b/video/out/d3d11/context.h new file mode 100644 index 0000000..8a9ef4c --- /dev/null +++ b/video/out/d3d11/context.h @@ -0,0 +1,9 @@ +#pragma once + +#include <dxgi.h> + +#include "video/out/gpu/context.h" + +// Get the underlying D3D11 swap chain from an RA context. The returned swap chain is +// refcounted and must be released by the caller. +IDXGISwapChain *ra_d3d11_ctx_get_swapchain(struct ra_ctx *ra); diff --git a/video/out/d3d11/hwdec_d3d11va.c b/video/out/d3d11/hwdec_d3d11va.c new file mode 100644 index 0000000..6aaa12b --- /dev/null +++ b/video/out/d3d11/hwdec_d3d11va.c @@ -0,0 +1,258 @@ +/* + * This file is part of mpv. + * + * mpv is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * mpv is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with mpv. If not, see <http://www.gnu.org/licenses/>. + */ + +#include <windows.h> +#include <d3d11.h> +#include <d3d11_1.h> + +#include "common/common.h" +#include "options/m_config.h" +#include "osdep/windows_utils.h" +#include "video/hwdec.h" +#include "video/d3d.h" +#include "video/out/d3d11/ra_d3d11.h" +#include "video/out/gpu/hwdec.h" + +struct d3d11va_opts { + bool zero_copy; +}; + +#define OPT_BASE_STRUCT struct d3d11va_opts +const struct m_sub_options d3d11va_conf = { + .opts = (const struct m_option[]) { + {"d3d11va-zero-copy", OPT_BOOL(zero_copy)}, + {0} + }, + .defaults = &(const struct d3d11va_opts) {0}, + .size = sizeof(struct d3d11va_opts) +}; + +struct priv_owner { + struct d3d11va_opts *opts; + + struct mp_hwdec_ctx hwctx; + ID3D11Device *device; + ID3D11Device1 *device1; +}; + +struct priv { + // 1-copy path + ID3D11DeviceContext1 *ctx; + ID3D11Texture2D *copy_tex; + + // zero-copy path + int num_planes; + const struct ra_format *fmt[4]; +}; + +static void uninit(struct ra_hwdec *hw) +{ + struct priv_owner *p = hw->priv; + hwdec_devices_remove(hw->devs, &p->hwctx); + av_buffer_unref(&p->hwctx.av_device_ref); + SAFE_RELEASE(p->device); + SAFE_RELEASE(p->device1); +} + +static int init(struct ra_hwdec *hw) +{ + struct priv_owner *p = hw->priv; + HRESULT hr; + + if (!ra_is_d3d11(hw->ra_ctx->ra)) + return -1; + p->device = ra_d3d11_get_device(hw->ra_ctx->ra); + if (!p->device) + return -1; + + p->opts = mp_get_config_group(hw->priv, hw->global, &d3d11va_conf); + + // D3D11VA requires Direct3D 11.1, so this should always succeed + hr = ID3D11Device_QueryInterface(p->device, &IID_ID3D11Device1, + (void**)&p->device1); + if (FAILED(hr)) { + MP_ERR(hw, "Failed to get D3D11.1 interface: %s\n", + mp_HRESULT_to_str(hr)); + return -1; + } + + ID3D10Multithread *multithread; + hr = ID3D11Device_QueryInterface(p->device, &IID_ID3D10Multithread, + (void **)&multithread); + if (FAILED(hr)) { + MP_ERR(hw, "Failed to get Multithread interface: %s\n", + mp_HRESULT_to_str(hr)); + return -1; + } + ID3D10Multithread_SetMultithreadProtected(multithread, TRUE); + ID3D10Multithread_Release(multithread); + + static const int subfmts[] = {IMGFMT_NV12, IMGFMT_P010, 0}; + p->hwctx = (struct mp_hwdec_ctx){ + .driver_name = hw->driver->name, + .av_device_ref = d3d11_wrap_device_ref(p->device), + .supported_formats = subfmts, + .hw_imgfmt = IMGFMT_D3D11, + }; + + if (!p->hwctx.av_device_ref) { + MP_VERBOSE(hw, "Failed to create hwdevice_ctx\n"); + return -1; + } + + hwdec_devices_add(hw->devs, &p->hwctx); + return 0; +} + +static void mapper_uninit(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + for (int i = 0; i < 4; i++) + ra_tex_free(mapper->ra, &mapper->tex[i]); + SAFE_RELEASE(p->copy_tex); + SAFE_RELEASE(p->ctx); +} + +static int mapper_init(struct ra_hwdec_mapper *mapper) +{ + struct priv_owner *o = mapper->owner->priv; + struct priv *p = mapper->priv; + HRESULT hr; + + mapper->dst_params = mapper->src_params; + mapper->dst_params.imgfmt = mapper->src_params.hw_subfmt; + mapper->dst_params.hw_subfmt = 0; + + struct ra_imgfmt_desc desc = {0}; + + if (!ra_get_imgfmt_desc(mapper->ra, mapper->dst_params.imgfmt, &desc)) + return -1; + + if (o->opts->zero_copy) { + // In the zero-copy path, we create the ra_tex objects in the map + // operation, so we just need to store the format of each plane + p->num_planes = desc.num_planes; + for (int i = 0; i < desc.num_planes; i++) + p->fmt[i] = desc.planes[i]; + } else { + struct mp_image layout = {0}; + mp_image_set_params(&layout, &mapper->dst_params); + + DXGI_FORMAT copy_fmt; + switch (mapper->dst_params.imgfmt) { + case IMGFMT_NV12: copy_fmt = DXGI_FORMAT_NV12; break; + case IMGFMT_P010: copy_fmt = DXGI_FORMAT_P010; break; + default: return -1; + } + + D3D11_TEXTURE2D_DESC copy_desc = { + .Width = mapper->dst_params.w, + .Height = mapper->dst_params.h, + .MipLevels = 1, + .ArraySize = 1, + .SampleDesc.Count = 1, + .Format = copy_fmt, + .BindFlags = D3D11_BIND_SHADER_RESOURCE, + }; + hr = ID3D11Device_CreateTexture2D(o->device, ©_desc, NULL, + &p->copy_tex); + if (FAILED(hr)) { + MP_FATAL(mapper, "Could not create shader resource texture\n"); + return -1; + } + + for (int i = 0; i < desc.num_planes; i++) { + mapper->tex[i] = ra_d3d11_wrap_tex_video(mapper->ra, p->copy_tex, + mp_image_plane_w(&layout, i), mp_image_plane_h(&layout, i), 0, + desc.planes[i]); + if (!mapper->tex[i]) { + MP_FATAL(mapper, "Could not create RA texture view\n"); + return -1; + } + } + + // A ref to the immediate context is needed for CopySubresourceRegion + ID3D11Device1_GetImmediateContext1(o->device1, &p->ctx); + } + + return 0; +} + +static int mapper_map(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + ID3D11Texture2D *tex = (void *)mapper->src->planes[0]; + int subresource = (intptr_t)mapper->src->planes[1]; + + if (p->copy_tex) { + ID3D11DeviceContext1_CopySubresourceRegion1(p->ctx, + (ID3D11Resource *)p->copy_tex, 0, 0, 0, 0, + (ID3D11Resource *)tex, subresource, (&(D3D11_BOX) { + .left = 0, + .top = 0, + .front = 0, + .right = mapper->dst_params.w, + .bottom = mapper->dst_params.h, + .back = 1, + }), D3D11_COPY_DISCARD); + + // We no longer need the original texture after copying it. + mp_image_unrefp(&mapper->src); + } else { + D3D11_TEXTURE2D_DESC desc2d; + ID3D11Texture2D_GetDesc(tex, &desc2d); + + for (int i = 0; i < p->num_planes; i++) { + // The video decode texture may include padding, so the size of the + // ra_tex needs to be determined by the actual size of the Tex2D + bool chroma = i >= 1; + int w = desc2d.Width / (chroma ? 2 : 1); + int h = desc2d.Height / (chroma ? 2 : 1); + + mapper->tex[i] = ra_d3d11_wrap_tex_video(mapper->ra, tex, + w, h, subresource, p->fmt[i]); + if (!mapper->tex[i]) + return -1; + } + } + + return 0; +} + +static void mapper_unmap(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + if (p->copy_tex) + return; + for (int i = 0; i < 4; i++) + ra_tex_free(mapper->ra, &mapper->tex[i]); +} + +const struct ra_hwdec_driver ra_hwdec_d3d11va = { + .name = "d3d11va", + .priv_size = sizeof(struct priv_owner), + .imgfmts = {IMGFMT_D3D11, 0}, + .init = init, + .uninit = uninit, + .mapper = &(const struct ra_hwdec_mapper_driver){ + .priv_size = sizeof(struct priv), + .init = mapper_init, + .uninit = mapper_uninit, + .map = mapper_map, + .unmap = mapper_unmap, + }, +}; diff --git a/video/out/d3d11/hwdec_dxva2dxgi.c b/video/out/d3d11/hwdec_dxva2dxgi.c new file mode 100644 index 0000000..62158d4 --- /dev/null +++ b/video/out/d3d11/hwdec_dxva2dxgi.c @@ -0,0 +1,478 @@ +/* + * This file is part of mpv. + * + * mpv is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * mpv is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with mpv. If not, see <http://www.gnu.org/licenses/>. + */ + +#include <windows.h> +#include <d3d9.h> +#include <d3d11.h> +#include <dxva2api.h> + +#include "common/common.h" +#include "osdep/windows_utils.h" +#include "video/hwdec.h" +#include "video/d3d.h" +#include "video/out/d3d11/ra_d3d11.h" +#include "video/out/gpu/hwdec.h" + +struct priv_owner { + struct mp_hwdec_ctx hwctx; + ID3D11Device *dev11; + IDirect3DDevice9Ex *dev9; +}; + +struct queue_surf { + ID3D11Texture2D *tex11; + ID3D11Query *idle11; + ID3D11Texture2D *stage11; + IDirect3DTexture9 *tex9; + IDirect3DSurface9 *surf9; + IDirect3DSurface9 *stage9; + struct ra_tex *tex; + + bool busy11; // The surface is currently being used by D3D11 +}; + +struct priv { + ID3D11Device *dev11; + ID3D11DeviceContext *ctx11; + IDirect3DDevice9Ex *dev9; + + // Surface queue stuff. Following Microsoft recommendations, a queue of + // surfaces is used to share images between D3D9 and D3D11. This allows + // multiple D3D11 frames to be in-flight at once. + struct queue_surf **queue; + int queue_len; + int queue_pos; +}; + +static void uninit(struct ra_hwdec *hw) +{ + struct priv_owner *p = hw->priv; + hwdec_devices_remove(hw->devs, &p->hwctx); + av_buffer_unref(&p->hwctx.av_device_ref); + SAFE_RELEASE(p->dev11); + SAFE_RELEASE(p->dev9); +} + +static int init(struct ra_hwdec *hw) +{ + struct priv_owner *p = hw->priv; + IDirect3D9Ex *d3d9ex = NULL; + int ret = -1; + HRESULT hr; + + if (!ra_is_d3d11(hw->ra_ctx->ra)) + goto done; + p->dev11 = ra_d3d11_get_device(hw->ra_ctx->ra); + if (!p->dev11) + goto done; + + d3d_load_dlls(); + if (!d3d9_dll) { + MP_FATAL(hw, "Failed to load \"d3d9.dll\": %s\n", mp_LastError_to_str()); + goto done; + } + if (!dxva2_dll) { + MP_FATAL(hw, "Failed to load \"dxva2.dll\": %s\n", mp_LastError_to_str()); + goto done; + } + + HRESULT (WINAPI *Direct3DCreate9Ex)(UINT SDKVersion, IDirect3D9Ex **ppD3D); + Direct3DCreate9Ex = (void *)GetProcAddress(d3d9_dll, "Direct3DCreate9Ex"); + if (!Direct3DCreate9Ex) { + MP_FATAL(hw, "Direct3D 9Ex not supported\n"); + goto done; + } + + hr = Direct3DCreate9Ex(D3D_SDK_VERSION, &d3d9ex); + if (FAILED(hr)) { + MP_FATAL(hw, "Couldn't create Direct3D9Ex: %s\n", mp_HRESULT_to_str(hr)); + goto done; + } + + D3DPRESENT_PARAMETERS pparams = { + .BackBufferWidth = 16, + .BackBufferHeight = 16, + .BackBufferCount = 1, + .SwapEffect = D3DSWAPEFFECT_DISCARD, + .hDeviceWindow = GetDesktopWindow(), + .Windowed = TRUE, + .Flags = D3DPRESENTFLAG_VIDEO, + }; + hr = IDirect3D9Ex_CreateDeviceEx(d3d9ex, D3DADAPTER_DEFAULT, + D3DDEVTYPE_HAL, GetDesktopWindow(), D3DCREATE_NOWINDOWCHANGES | + D3DCREATE_FPU_PRESERVE | D3DCREATE_HARDWARE_VERTEXPROCESSING | + D3DCREATE_DISABLE_PSGP_THREADING | D3DCREATE_MULTITHREADED, &pparams, + NULL, &p->dev9); + if (FAILED(hr)) { + MP_FATAL(hw, "Failed to create Direct3D9Ex device: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + // Check if it's possible to StretchRect() from NV12 to XRGB surfaces + hr = IDirect3D9Ex_CheckDeviceFormatConversion(d3d9ex, D3DADAPTER_DEFAULT, + D3DDEVTYPE_HAL, MAKEFOURCC('N', 'V', '1', '2'), D3DFMT_X8R8G8B8); + if (hr != S_OK) { + MP_FATAL(hw, "Can't StretchRect from NV12 to XRGB surfaces\n"); + goto done; + } + + p->hwctx = (struct mp_hwdec_ctx){ + .driver_name = hw->driver->name, + .av_device_ref = d3d9_wrap_device_ref((IDirect3DDevice9 *)p->dev9), + .hw_imgfmt = IMGFMT_DXVA2, + }; + + if (!p->hwctx.av_device_ref) { + MP_VERBOSE(hw, "Failed to create hwdevice_ctx\n"); + goto done; + } + + hwdec_devices_add(hw->devs, &p->hwctx); + + ret = 0; +done: + SAFE_RELEASE(d3d9ex); + return ret; +} + +static int mapper_init(struct ra_hwdec_mapper *mapper) +{ + struct priv_owner *o = mapper->owner->priv; + struct priv *p = mapper->priv; + + ID3D11Device_AddRef(o->dev11); + p->dev11 = o->dev11; + IDirect3DDevice9Ex_AddRef(o->dev9); + p->dev9 = o->dev9; + ID3D11Device_GetImmediateContext(o->dev11, &p->ctx11); + + mapper->dst_params = mapper->src_params; + mapper->dst_params.imgfmt = IMGFMT_RGB0; + mapper->dst_params.hw_subfmt = 0; + return 0; +} + +static void surf_destroy(struct ra_hwdec_mapper *mapper, + struct queue_surf *surf) +{ + if (!surf) + return; + SAFE_RELEASE(surf->tex11); + SAFE_RELEASE(surf->idle11); + SAFE_RELEASE(surf->stage11); + SAFE_RELEASE(surf->tex9); + SAFE_RELEASE(surf->surf9); + SAFE_RELEASE(surf->stage9); + ra_tex_free(mapper->ra, &surf->tex); + talloc_free(surf); +} + +static struct queue_surf *surf_create(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + IDXGIResource *res11 = NULL; + bool success = false; + HRESULT hr; + + struct queue_surf *surf = talloc_ptrtype(p, surf); + + D3D11_TEXTURE2D_DESC desc11 = { + .Width = mapper->src->w, + .Height = mapper->src->h, + .MipLevels = 1, + .ArraySize = 1, + .Format = DXGI_FORMAT_B8G8R8X8_UNORM, + .SampleDesc.Count = 1, + .Usage = D3D11_USAGE_DEFAULT, + .BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET, + .MiscFlags = D3D11_RESOURCE_MISC_SHARED, + }; + hr = ID3D11Device_CreateTexture2D(p->dev11, &desc11, NULL, &surf->tex11); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to create D3D11 texture: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + // Try to use a 16x16 staging texture, unless the source surface is + // smaller. Ideally, a 1x1 texture would be sufficient, but Microsoft's + // D3D9ExDXGISharedSurf example uses 16x16 to avoid driver bugs. + D3D11_TEXTURE2D_DESC sdesc11 = { + .Width = MPMIN(16, desc11.Width), + .Height = MPMIN(16, desc11.Height), + .MipLevels = 1, + .ArraySize = 1, + .Format = DXGI_FORMAT_B8G8R8X8_UNORM, + .SampleDesc.Count = 1, + .Usage = D3D11_USAGE_STAGING, + .CPUAccessFlags = D3D11_CPU_ACCESS_READ, + }; + hr = ID3D11Device_CreateTexture2D(p->dev11, &sdesc11, NULL, &surf->stage11); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to create D3D11 staging texture: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + hr = ID3D11Texture2D_QueryInterface(surf->tex11, &IID_IDXGIResource, + (void**)&res11); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to get share handle: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + HANDLE share_handle; + hr = IDXGIResource_GetSharedHandle(res11, &share_handle); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to get share handle: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + hr = ID3D11Device_CreateQuery(p->dev11, + &(D3D11_QUERY_DESC) { D3D11_QUERY_EVENT }, &surf->idle11); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to create D3D11 query: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + // Share the D3D11 texture with D3D9Ex + hr = IDirect3DDevice9Ex_CreateTexture(p->dev9, desc11.Width, desc11.Height, + 1, D3DUSAGE_RENDERTARGET, D3DFMT_X8R8G8B8, D3DPOOL_DEFAULT, + &surf->tex9, &share_handle); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to create D3D9 texture: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + hr = IDirect3DTexture9_GetSurfaceLevel(surf->tex9, 0, &surf->surf9); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to get D3D9 surface: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + // As above, try to use a 16x16 staging texture to avoid driver bugs + hr = IDirect3DDevice9Ex_CreateRenderTarget(p->dev9, + MPMIN(16, desc11.Width), MPMIN(16, desc11.Height), D3DFMT_X8R8G8B8, + D3DMULTISAMPLE_NONE, 0, TRUE, &surf->stage9, NULL); + if (FAILED(hr)) { + MP_ERR(mapper, "Failed to create D3D9 staging surface: %s\n", + mp_HRESULT_to_str(hr)); + goto done; + } + + surf->tex = ra_d3d11_wrap_tex(mapper->ra, (ID3D11Resource *)surf->tex11); + if (!surf->tex) + goto done; + + success = true; +done: + if (!success) + surf_destroy(mapper, surf); + SAFE_RELEASE(res11); + return success ? surf : NULL; +} + +// true if the surface is currently in-use by the D3D11 graphics pipeline +static bool surf_is_idle11(struct ra_hwdec_mapper *mapper, + struct queue_surf *surf) +{ + struct priv *p = mapper->priv; + HRESULT hr; + BOOL idle; + + if (!surf->busy11) + return true; + + hr = ID3D11DeviceContext_GetData(p->ctx11, + (ID3D11Asynchronous *)surf->idle11, &idle, sizeof(idle), + D3D11_ASYNC_GETDATA_DONOTFLUSH); + if (FAILED(hr) || hr == S_FALSE || !idle) + return false; + + surf->busy11 = false; + return true; +} + +// If the surface is currently in-use by the D3D11 graphics pipeline, wait for +// it to become idle. Should only be called in the queue-underflow case. +static bool surf_wait_idle11(struct ra_hwdec_mapper *mapper, + struct queue_surf *surf) +{ + struct priv *p = mapper->priv; + HRESULT hr; + + ID3D11DeviceContext_CopySubresourceRegion(p->ctx11, + (ID3D11Resource *)surf->stage11, 0, 0, 0, 0, + (ID3D11Resource *)surf->tex11, 0, (&(D3D11_BOX){ + .right = MPMIN(16, mapper->src->w), + .bottom = MPMIN(16, mapper->src->h), + .back = 1, + })); + + // Block until the surface becomes idle (see surf_wait_idle9()) + D3D11_MAPPED_SUBRESOURCE map = {0}; + hr = ID3D11DeviceContext_Map(p->ctx11, (ID3D11Resource *)surf->stage11, 0, + D3D11_MAP_READ, 0, &map); + if (FAILED(hr)) { + MP_ERR(mapper, "Couldn't map D3D11 staging texture: %s\n", + mp_HRESULT_to_str(hr)); + return false; + } + + ID3D11DeviceContext_Unmap(p->ctx11, (ID3D11Resource *)surf->stage11, 0); + surf->busy11 = false; + return true; +} + +static bool surf_wait_idle9(struct ra_hwdec_mapper *mapper, + struct queue_surf *surf) +{ + struct priv *p = mapper->priv; + HRESULT hr; + + // Rather than polling for the surface to become idle, copy part of the + // surface to a staging texture and map it. This should block until the + // surface becomes idle. Microsoft's ISurfaceQueue does this as well. + RECT rc = {0, 0, MPMIN(16, mapper->src->w), MPMIN(16, mapper->src->h)}; + hr = IDirect3DDevice9Ex_StretchRect(p->dev9, surf->surf9, &rc, surf->stage9, + &rc, D3DTEXF_NONE); + if (FAILED(hr)) { + MP_ERR(mapper, "Couldn't copy to D3D9 staging texture: %s\n", + mp_HRESULT_to_str(hr)); + return false; + } + + D3DLOCKED_RECT lock; + hr = IDirect3DSurface9_LockRect(surf->stage9, &lock, NULL, D3DLOCK_READONLY); + if (FAILED(hr)) { + MP_ERR(mapper, "Couldn't map D3D9 staging texture: %s\n", + mp_HRESULT_to_str(hr)); + return false; + } + + IDirect3DSurface9_UnlockRect(surf->stage9); + p->queue[p->queue_pos]->busy11 = true; + return true; +} + +static struct queue_surf *surf_acquire(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + + if (!p->queue_len || !surf_is_idle11(mapper, p->queue[p->queue_pos])) { + if (p->queue_len < 16) { + struct queue_surf *surf = surf_create(mapper); + if (!surf) + return NULL; + + // The next surface is busy, so grow the queue + MP_TARRAY_INSERT_AT(p, p->queue, p->queue_len, p->queue_pos, surf); + MP_DBG(mapper, "Queue grew to %d surfaces\n", p->queue_len); + } else { + // For sanity, don't let the queue grow beyond 16 surfaces. It + // should never get this big. If it does, wait for the surface to + // become idle rather than polling it. + if (!surf_wait_idle11(mapper, p->queue[p->queue_pos])) + return NULL; + MP_WARN(mapper, "Queue underflow!\n"); + } + } + return p->queue[p->queue_pos]; +} + +static void surf_release(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + ID3D11DeviceContext_End(p->ctx11, + (ID3D11Asynchronous *)p->queue[p->queue_pos]->idle11); + + // The current surface is now in-flight, move to the next surface + p->queue_pos++; + if (p->queue_pos >= p->queue_len) + p->queue_pos = 0; +} + +static void mapper_uninit(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + + for (int i = 0; i < p->queue_len; i++) + surf_destroy(mapper, p->queue[i]); + + SAFE_RELEASE(p->ctx11); + SAFE_RELEASE(p->dev9); + SAFE_RELEASE(p->dev11); +} + +static int mapper_map(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + HRESULT hr; + + struct queue_surf *surf = surf_acquire(mapper); + if (!surf) + return -1; + + RECT rc = {0, 0, mapper->src->w, mapper->src->h}; + IDirect3DSurface9* hw_surface = (IDirect3DSurface9 *)mapper->src->planes[3]; + + hr = IDirect3DDevice9Ex_StretchRect(p->dev9, hw_surface, &rc, surf->surf9, + &rc, D3DTEXF_NONE); + if (FAILED(hr)) { + MP_ERR(mapper, "StretchRect() failed: %s\n", mp_HRESULT_to_str(hr)); + return -1; + } + + if (!surf_wait_idle9(mapper, surf)) + return -1; + + mapper->tex[0] = surf->tex; + return 0; +} + +static void mapper_unmap(struct ra_hwdec_mapper *mapper) +{ + struct priv *p = mapper->priv; + + if (p->queue_pos < p->queue_len && + p->queue[p->queue_pos]->tex == mapper->tex[0]) + { + surf_release(mapper); + mapper->tex[0] = NULL; + } +} + +const struct ra_hwdec_driver ra_hwdec_dxva2dxgi = { + .name = "dxva2-dxgi", + .priv_size = sizeof(struct priv_owner), + .imgfmts = {IMGFMT_DXVA2, 0}, + .init = init, + .uninit = uninit, + .mapper = &(const struct ra_hwdec_mapper_driver){ + .priv_size = sizeof(struct priv), + .init = mapper_init, + .uninit = mapper_uninit, + .map = mapper_map, + .unmap = mapper_unmap, + }, +}; diff --git a/video/out/d3d11/ra_d3d11.c b/video/out/d3d11/ra_d3d11.c new file mode 100644 index 0000000..84fd004 --- /dev/null +++ b/video/out/d3d11/ra_d3d11.c @@ -0,0 +1,2544 @@ +#include <windows.h> +#include <versionhelpers.h> +#include <d3d11_1.h> +#include <d3d11sdklayers.h> +#include <dxgi1_2.h> +#include <d3dcompiler.h> +#include <spirv_cross_c.h> + +#include "common/msg.h" +#include "osdep/io.h" +#include "osdep/subprocess.h" +#include "osdep/timer.h" +#include "osdep/windows_utils.h" +#include "video/out/gpu/spirv.h" +#include "video/out/gpu/utils.h" + +#include "ra_d3d11.h" + +#ifndef D3D11_1_UAV_SLOT_COUNT +#define D3D11_1_UAV_SLOT_COUNT (64) +#endif +#define D3D11_FORMAT_SUPPORT2_UAV_TYPED_STORE (0x80) + +// D3D11.3 message IDs, not present in mingw-w64 v9 +#define D3D11_MESSAGE_ID_CREATE_FENCE ((D3D11_MESSAGE_ID)0x300209) +#define D3D11_MESSAGE_ID_DESTROY_FENCE ((D3D11_MESSAGE_ID)0x30020b) + +struct dll_version { + uint16_t major; + uint16_t minor; + uint16_t build; + uint16_t revision; +}; + +struct ra_d3d11 { + struct spirv_compiler *spirv; + + ID3D11Device *dev; + ID3D11Device1 *dev1; + ID3D11DeviceContext *ctx; + ID3D11DeviceContext1 *ctx1; + pD3DCompile D3DCompile; + + struct dll_version d3d_compiler_ver; + + // Debug interfaces (--gpu-debug) + ID3D11Debug *debug; + ID3D11InfoQueue *iqueue; + + // Device capabilities + D3D_FEATURE_LEVEL fl; + bool has_clear_view; + bool has_timestamp_queries; + int max_uavs; + + // Streaming dynamic vertex buffer, which is used for all renderpasses + ID3D11Buffer *vbuf; + size_t vbuf_size; + size_t vbuf_used; + + // clear() renderpass resources (only used when has_clear_view is false) + ID3D11PixelShader *clear_ps; + ID3D11VertexShader *clear_vs; + ID3D11InputLayout *clear_layout; + ID3D11Buffer *clear_vbuf; + ID3D11Buffer *clear_cbuf; + + // blit() renderpass resources + ID3D11PixelShader *blit_float_ps; + ID3D11VertexShader *blit_vs; + ID3D11InputLayout *blit_layout; + ID3D11Buffer *blit_vbuf; + ID3D11SamplerState *blit_sampler; +}; + +struct d3d_tex { + // res mirrors one of tex1d, tex2d or tex3d for convenience. It does not + // hold an additional reference to the texture object. + ID3D11Resource *res; + + ID3D11Texture1D *tex1d; + ID3D11Texture2D *tex2d; + ID3D11Texture3D *tex3d; + int array_slice; + + // Staging texture for tex_download(), 2D only + ID3D11Texture2D *staging; + + ID3D11ShaderResourceView *srv; + ID3D11RenderTargetView *rtv; + ID3D11UnorderedAccessView *uav; + ID3D11SamplerState *sampler; +}; + +struct d3d_buf { + ID3D11Buffer *buf; + ID3D11UnorderedAccessView *uav; + void *data; // System-memory mirror of the data in buf + bool dirty; // Is buf out of date? +}; + +struct d3d_rpass { + ID3D11PixelShader *ps; + ID3D11VertexShader *vs; + ID3D11ComputeShader *cs; + ID3D11InputLayout *layout; + ID3D11BlendState *bstate; +}; + +struct d3d_timer { + ID3D11Query *ts_start; + ID3D11Query *ts_end; + ID3D11Query *disjoint; + uint64_t result; // Latches the result from the previous use of the timer +}; + +struct d3d_fmt { + const char *name; + int components; + int bytes; + int bits[4]; + DXGI_FORMAT fmt; + enum ra_ctype ctype; + bool unordered; +}; + +static const char clear_vs[] = "\ +float4 main(float2 pos : POSITION) : SV_Position\n\ +{\n\ + return float4(pos, 0.0, 1.0);\n\ +}\n\ +"; + +static const char clear_ps[] = "\ +cbuffer ps_cbuf : register(b0) {\n\ + float4 color : packoffset(c0);\n\ +}\n\ +\n\ +float4 main(float4 pos : SV_Position) : SV_Target\n\ +{\n\ + return color;\n\ +}\n\ +"; + +struct blit_vert { + float x, y, u, v; +}; + +static const char blit_vs[] = "\ +void main(float2 pos : POSITION, float2 coord : TEXCOORD0,\n\ + out float4 out_pos : SV_Position, out float2 out_coord : TEXCOORD0)\n\ +{\n\ + out_pos = float4(pos, 0.0, 1.0);\n\ + out_coord = coord;\n\ +}\n\ +"; + +static const char blit_float_ps[] = "\ +Texture2D<float4> tex : register(t0);\n\ +SamplerState samp : register(s0);\n\ +\n\ +float4 main(float4 pos : SV_Position, float2 coord : TEXCOORD0) : SV_Target\n\ +{\n\ + return tex.Sample(samp, coord);\n\ +}\n\ +"; + +#define DXFMT(f, t) .fmt = DXGI_FORMAT_##f##_##t, .ctype = RA_CTYPE_##t +static struct d3d_fmt formats[] = { + { "r8", 1, 1, { 8}, DXFMT(R8, UNORM) }, + { "rg8", 2, 2, { 8, 8}, DXFMT(R8G8, UNORM) }, + { "rgba8", 4, 4, { 8, 8, 8, 8}, DXFMT(R8G8B8A8, UNORM) }, + { "r16", 1, 2, {16}, DXFMT(R16, UNORM) }, + { "rg16", 2, 4, {16, 16}, DXFMT(R16G16, UNORM) }, + { "rgba16", 4, 8, {16, 16, 16, 16}, DXFMT(R16G16B16A16, UNORM) }, + + { "r32ui", 1, 4, {32}, DXFMT(R32, UINT) }, + { "rg32ui", 2, 8, {32, 32}, DXFMT(R32G32, UINT) }, + { "rgb32ui", 3, 12, {32, 32, 32}, DXFMT(R32G32B32, UINT) }, + { "rgba32ui", 4, 16, {32, 32, 32, 32}, DXFMT(R32G32B32A32, UINT) }, + + { "r16hf", 1, 2, {16}, DXFMT(R16, FLOAT) }, + { "rg16hf", 2, 4, {16, 16}, DXFMT(R16G16, FLOAT) }, + { "rgba16hf", 4, 8, {16, 16, 16, 16}, DXFMT(R16G16B16A16, FLOAT) }, + { "r32f", 1, 4, {32}, DXFMT(R32, FLOAT) }, + { "rg32f", 2, 8, {32, 32}, DXFMT(R32G32, FLOAT) }, + { "rgb32f", 3, 12, {32, 32, 32}, DXFMT(R32G32B32, FLOAT) }, + { "rgba32f", 4, 16, {32, 32, 32, 32}, DXFMT(R32G32B32A32, FLOAT) }, + + { "rgb10_a2", 4, 4, {10, 10, 10, 2}, DXFMT(R10G10B10A2, UNORM) }, + { "bgra8", 4, 4, { 8, 8, 8, 8}, DXFMT(B8G8R8A8, UNORM), .unordered = true }, + { "bgrx8", 3, 4, { 8, 8, 8}, DXFMT(B8G8R8X8, UNORM), .unordered = true }, +}; + +static bool dll_version_equal(struct dll_version a, struct dll_version b) +{ + return a.major == b.major && + a.minor == b.minor && + a.build == b.build && + a.revision == b.revision; +} + +DXGI_FORMAT ra_d3d11_get_format(const struct ra_format *fmt) +{ + struct d3d_fmt *d3d = fmt->priv; + return d3d->fmt; +} + +const struct ra_format *ra_d3d11_get_ra_format(struct ra *ra, DXGI_FORMAT fmt) +{ + for (int i = 0; i < ra->num_formats; i++) { + struct ra_format *ra_fmt = ra->formats[i]; + + if (ra_d3d11_get_format(ra_fmt) == fmt) + return ra_fmt; + } + + return NULL; +} + +static void setup_formats(struct ra *ra) +{ + // All formats must be usable as a 2D texture + static const UINT sup_basic = D3D11_FORMAT_SUPPORT_TEXTURE2D; + // SHADER_SAMPLE indicates support for linear sampling, point always works + static const UINT sup_filter = D3D11_FORMAT_SUPPORT_SHADER_SAMPLE; + // RA requires renderable surfaces to be blendable as well + static const UINT sup_render = D3D11_FORMAT_SUPPORT_RENDER_TARGET | + D3D11_FORMAT_SUPPORT_BLENDABLE; + // Typed UAVs are equivalent to images. RA only cares if they're storable. + static const UINT sup_store = D3D11_FORMAT_SUPPORT_TYPED_UNORDERED_ACCESS_VIEW; + static const UINT sup2_store = D3D11_FORMAT_SUPPORT2_UAV_TYPED_STORE; + + struct ra_d3d11 *p = ra->priv; + HRESULT hr; + + for (int i = 0; i < MP_ARRAY_SIZE(formats); i++) { + struct d3d_fmt *d3dfmt = &formats[i]; + UINT support = 0; + hr = ID3D11Device_CheckFormatSupport(p->dev, d3dfmt->fmt, &support); + if (FAILED(hr)) + continue; + if ((support & sup_basic) != sup_basic) + continue; + + D3D11_FEATURE_DATA_FORMAT_SUPPORT2 sup2 = { .InFormat = d3dfmt->fmt }; + ID3D11Device_CheckFeatureSupport(p->dev, D3D11_FEATURE_FORMAT_SUPPORT2, + ², sizeof(sup2)); + UINT support2 = sup2.OutFormatSupport2; + + struct ra_format *fmt = talloc_zero(ra, struct ra_format); + *fmt = (struct ra_format) { + .name = d3dfmt->name, + .priv = d3dfmt, + .ctype = d3dfmt->ctype, + .ordered = !d3dfmt->unordered, + .num_components = d3dfmt->components, + .pixel_size = d3dfmt->bytes, + .linear_filter = (support & sup_filter) == sup_filter, + .renderable = (support & sup_render) == sup_render, + .storable = p->fl >= D3D_FEATURE_LEVEL_11_0 && + (support & sup_store) == sup_store && + (support2 & sup2_store) == sup2_store, + }; + + if (support & D3D11_FORMAT_SUPPORT_TEXTURE1D) + ra->caps |= RA_CAP_TEX_1D; + + for (int j = 0; j < d3dfmt->components; j++) + fmt->component_size[j] = fmt->component_depth[j] = d3dfmt->bits[j]; + + fmt->glsl_format = ra_fmt_glsl_format(fmt); + + MP_TARRAY_APPEND(ra, ra->formats, ra->num_formats, fmt); + } +} + +static bool tex_init(struct ra *ra, struct ra_tex *tex) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_tex *tex_p = tex->priv; + struct ra_tex_params *params = &tex->params; + HRESULT hr; + + // A SRV is required for renderpasses and blitting, since blitting can use + // a renderpass internally + if (params->render_src || params->blit_src) { + // Always specify the SRV format for simplicity. This will match the + // texture format for textures created with tex_create, but it can be + // different for wrapped planar video textures. + D3D11_SHADER_RESOURCE_VIEW_DESC srvdesc = { + .Format = ra_d3d11_get_format(params->format), + }; + switch (params->dimensions) { + case 1: + if (tex_p->array_slice >= 0) { + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE1DARRAY; + srvdesc.Texture1DArray.MipLevels = 1; + srvdesc.Texture1DArray.FirstArraySlice = tex_p->array_slice; + srvdesc.Texture1DArray.ArraySize = 1; + } else { + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE1D; + srvdesc.Texture1D.MipLevels = 1; + } + break; + case 2: + if (tex_p->array_slice >= 0) { + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2DARRAY; + srvdesc.Texture2DArray.MipLevels = 1; + srvdesc.Texture2DArray.FirstArraySlice = tex_p->array_slice; + srvdesc.Texture2DArray.ArraySize = 1; + } else { + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D; + srvdesc.Texture2D.MipLevels = 1; + } + break; + case 3: + // D3D11 does not have Texture3D arrays + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE3D; + srvdesc.Texture3D.MipLevels = 1; + break; + } + hr = ID3D11Device_CreateShaderResourceView(p->dev, tex_p->res, &srvdesc, + &tex_p->srv); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create SRV: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + } + + // Samplers are required for renderpasses, but not blitting, since the blit + // code uses its own point sampler + if (params->render_src) { + D3D11_SAMPLER_DESC sdesc = { + .AddressU = D3D11_TEXTURE_ADDRESS_CLAMP, + .AddressV = D3D11_TEXTURE_ADDRESS_CLAMP, + .AddressW = D3D11_TEXTURE_ADDRESS_CLAMP, + .ComparisonFunc = D3D11_COMPARISON_NEVER, + .MinLOD = 0, + .MaxLOD = D3D11_FLOAT32_MAX, + .MaxAnisotropy = 1, + }; + if (params->src_linear) + sdesc.Filter = D3D11_FILTER_MIN_MAG_MIP_LINEAR; + if (params->src_repeat) { + sdesc.AddressU = sdesc.AddressV = sdesc.AddressW = + D3D11_TEXTURE_ADDRESS_WRAP; + } + // The runtime pools sampler state objects internally, so we don't have + // to worry about resource usage when creating one for every ra_tex + hr = ID3D11Device_CreateSamplerState(p->dev, &sdesc, &tex_p->sampler); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create sampler: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + } + + // Like SRVs, an RTV is required for renderpass output and blitting + if (params->render_dst || params->blit_dst) { + hr = ID3D11Device_CreateRenderTargetView(p->dev, tex_p->res, NULL, + &tex_p->rtv); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create RTV: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + } + + if (p->fl >= D3D_FEATURE_LEVEL_11_0 && params->storage_dst) { + hr = ID3D11Device_CreateUnorderedAccessView(p->dev, tex_p->res, NULL, + &tex_p->uav); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create UAV: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + } + + return true; +error: + return false; +} + +static void tex_destroy(struct ra *ra, struct ra_tex *tex) +{ + if (!tex) + return; + struct d3d_tex *tex_p = tex->priv; + + SAFE_RELEASE(tex_p->srv); + SAFE_RELEASE(tex_p->rtv); + SAFE_RELEASE(tex_p->uav); + SAFE_RELEASE(tex_p->sampler); + SAFE_RELEASE(tex_p->res); + SAFE_RELEASE(tex_p->staging); + talloc_free(tex); +} + +static struct ra_tex *tex_create(struct ra *ra, + const struct ra_tex_params *params) +{ + // Only 2D textures may be downloaded for now + if (params->downloadable && params->dimensions != 2) + return NULL; + + struct ra_d3d11 *p = ra->priv; + HRESULT hr; + + struct ra_tex *tex = talloc_zero(NULL, struct ra_tex); + tex->params = *params; + tex->params.initial_data = NULL; + + struct d3d_tex *tex_p = tex->priv = talloc_zero(tex, struct d3d_tex); + DXGI_FORMAT fmt = ra_d3d11_get_format(params->format); + + D3D11_SUBRESOURCE_DATA data; + D3D11_SUBRESOURCE_DATA *pdata = NULL; + if (params->initial_data) { + data = (D3D11_SUBRESOURCE_DATA) { + .pSysMem = params->initial_data, + .SysMemPitch = params->w * params->format->pixel_size, + }; + if (params->dimensions >= 3) + data.SysMemSlicePitch = data.SysMemPitch * params->h; + pdata = &data; + } + + D3D11_USAGE usage = D3D11_USAGE_DEFAULT; + D3D11_BIND_FLAG bind_flags = 0; + + if (params->render_src || params->blit_src) + bind_flags |= D3D11_BIND_SHADER_RESOURCE; + if (params->render_dst || params->blit_dst) + bind_flags |= D3D11_BIND_RENDER_TARGET; + if (p->fl >= D3D_FEATURE_LEVEL_11_0 && params->storage_dst) + bind_flags |= D3D11_BIND_UNORDERED_ACCESS; + + // Apparently IMMUTABLE textures are efficient, so try to infer whether we + // can use one + if (params->initial_data && !params->render_dst && !params->storage_dst && + !params->blit_dst && !params->host_mutable) + usage = D3D11_USAGE_IMMUTABLE; + + switch (params->dimensions) { + case 1:; + D3D11_TEXTURE1D_DESC desc1d = { + .Width = params->w, + .MipLevels = 1, + .ArraySize = 1, + .Format = fmt, + .Usage = usage, + .BindFlags = bind_flags, + }; + hr = ID3D11Device_CreateTexture1D(p->dev, &desc1d, pdata, &tex_p->tex1d); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create Texture1D: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + tex_p->res = (ID3D11Resource *)tex_p->tex1d; + break; + case 2:; + D3D11_TEXTURE2D_DESC desc2d = { + .Width = params->w, + .Height = params->h, + .MipLevels = 1, + .ArraySize = 1, + .SampleDesc.Count = 1, + .Format = fmt, + .Usage = usage, + .BindFlags = bind_flags, + }; + hr = ID3D11Device_CreateTexture2D(p->dev, &desc2d, pdata, &tex_p->tex2d); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create Texture2D: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + tex_p->res = (ID3D11Resource *)tex_p->tex2d; + + // Create a staging texture with CPU access for tex_download() + if (params->downloadable) { + desc2d.BindFlags = 0; + desc2d.CPUAccessFlags = D3D11_CPU_ACCESS_READ; + desc2d.Usage = D3D11_USAGE_STAGING; + + hr = ID3D11Device_CreateTexture2D(p->dev, &desc2d, NULL, + &tex_p->staging); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to staging texture: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + } + break; + case 3:; + D3D11_TEXTURE3D_DESC desc3d = { + .Width = params->w, + .Height = params->h, + .Depth = params->d, + .MipLevels = 1, + .Format = fmt, + .Usage = usage, + .BindFlags = bind_flags, + }; + hr = ID3D11Device_CreateTexture3D(p->dev, &desc3d, pdata, &tex_p->tex3d); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create Texture3D: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + tex_p->res = (ID3D11Resource *)tex_p->tex3d; + break; + default: + MP_ASSERT_UNREACHABLE(); + } + + tex_p->array_slice = -1; + + if (!tex_init(ra, tex)) + goto error; + + return tex; + +error: + tex_destroy(ra, tex); + return NULL; +} + +struct ra_tex *ra_d3d11_wrap_tex(struct ra *ra, ID3D11Resource *res) +{ + HRESULT hr; + + struct ra_tex *tex = talloc_zero(NULL, struct ra_tex); + struct ra_tex_params *params = &tex->params; + struct d3d_tex *tex_p = tex->priv = talloc_zero(tex, struct d3d_tex); + + DXGI_FORMAT fmt = DXGI_FORMAT_UNKNOWN; + D3D11_USAGE usage = D3D11_USAGE_DEFAULT; + D3D11_BIND_FLAG bind_flags = 0; + + D3D11_RESOURCE_DIMENSION type; + ID3D11Resource_GetType(res, &type); + switch (type) { + case D3D11_RESOURCE_DIMENSION_TEXTURE2D: + hr = ID3D11Resource_QueryInterface(res, &IID_ID3D11Texture2D, + (void**)&tex_p->tex2d); + if (FAILED(hr)) { + MP_ERR(ra, "Resource is not a ID3D11Texture2D\n"); + goto error; + } + tex_p->res = (ID3D11Resource *)tex_p->tex2d; + + D3D11_TEXTURE2D_DESC desc2d; + ID3D11Texture2D_GetDesc(tex_p->tex2d, &desc2d); + if (desc2d.MipLevels != 1) { + MP_ERR(ra, "Mipmapped textures not supported for wrapping\n"); + goto error; + } + if (desc2d.ArraySize != 1) { + MP_ERR(ra, "Texture arrays not supported for wrapping\n"); + goto error; + } + if (desc2d.SampleDesc.Count != 1) { + MP_ERR(ra, "Multisampled textures not supported for wrapping\n"); + goto error; + } + + params->dimensions = 2; + params->w = desc2d.Width; + params->h = desc2d.Height; + params->d = 1; + usage = desc2d.Usage; + bind_flags = desc2d.BindFlags; + fmt = desc2d.Format; + break; + default: + // We could wrap Texture1D/3D as well, but keep it simple, since this + // function is only used for swapchain backbuffers at the moment + MP_ERR(ra, "Resource is not suitable to wrap\n"); + goto error; + } + + for (int i = 0; i < ra->num_formats; i++) { + DXGI_FORMAT target_fmt = ra_d3d11_get_format(ra->formats[i]); + if (fmt == target_fmt) { + params->format = ra->formats[i]; + break; + } + } + if (!params->format) { + MP_ERR(ra, "Could not find a suitable RA format for wrapped resource\n"); + goto error; + } + + if (bind_flags & D3D11_BIND_SHADER_RESOURCE) { + params->render_src = params->blit_src = true; + params->src_linear = params->format->linear_filter; + } + if (bind_flags & D3D11_BIND_RENDER_TARGET) + params->render_dst = params->blit_dst = true; + if (bind_flags & D3D11_BIND_UNORDERED_ACCESS) + params->storage_dst = true; + + if (usage != D3D11_USAGE_DEFAULT) { + MP_ERR(ra, "Resource is not D3D11_USAGE_DEFAULT\n"); + goto error; + } + + tex_p->array_slice = -1; + + if (!tex_init(ra, tex)) + goto error; + + return tex; +error: + tex_destroy(ra, tex); + return NULL; +} + +struct ra_tex *ra_d3d11_wrap_tex_video(struct ra *ra, ID3D11Texture2D *res, + int w, int h, int array_slice, + const struct ra_format *fmt) +{ + struct ra_tex *tex = talloc_zero(NULL, struct ra_tex); + struct ra_tex_params *params = &tex->params; + struct d3d_tex *tex_p = tex->priv = talloc_zero(tex, struct d3d_tex); + + tex_p->tex2d = res; + tex_p->res = (ID3D11Resource *)tex_p->tex2d; + ID3D11Texture2D_AddRef(res); + + D3D11_TEXTURE2D_DESC desc2d; + ID3D11Texture2D_GetDesc(tex_p->tex2d, &desc2d); + if (!(desc2d.BindFlags & D3D11_BIND_SHADER_RESOURCE)) { + MP_ERR(ra, "Video resource is not bindable\n"); + goto error; + } + + params->dimensions = 2; + params->w = w; + params->h = h; + params->d = 1; + params->render_src = true; + params->src_linear = true; + // fmt can be different to the texture format for planar video textures + params->format = fmt; + + if (desc2d.ArraySize > 1) { + tex_p->array_slice = array_slice; + } else { + tex_p->array_slice = -1; + } + + if (!tex_init(ra, tex)) + goto error; + + return tex; +error: + tex_destroy(ra, tex); + return NULL; +} + +ID3D11Resource *ra_d3d11_get_raw_tex(struct ra *ra, struct ra_tex *tex, + int *array_slice) +{ + struct d3d_tex *tex_p = tex->priv; + + ID3D11Resource_AddRef(tex_p->res); + if (array_slice) + *array_slice = tex_p->array_slice; + return tex_p->res; +} + +static bool tex_upload(struct ra *ra, const struct ra_tex_upload_params *params) +{ + struct ra_d3d11 *p = ra->priv; + struct ra_tex *tex = params->tex; + struct d3d_tex *tex_p = tex->priv; + + if (!params->src) { + MP_ERR(ra, "Pixel buffers are not supported\n"); + return false; + } + + const char *src = params->src; + ptrdiff_t stride = tex->params.dimensions >= 2 ? tex->params.w : 0; + ptrdiff_t pitch = tex->params.dimensions >= 3 ? stride * tex->params.h : 0; + bool invalidate = true; + D3D11_BOX rc; + D3D11_BOX *prc = NULL; + + if (tex->params.dimensions == 2) { + stride = params->stride; + + if (params->rc && (params->rc->x0 != 0 || params->rc->y0 != 0 || + params->rc->x1 != tex->params.w || params->rc->y1 != tex->params.h)) + { + rc = (D3D11_BOX) { + .left = params->rc->x0, + .top = params->rc->y0, + .front = 0, + .right = params->rc->x1, + .bottom = params->rc->y1, + .back = 1, + }; + prc = &rc; + invalidate = params->invalidate; + } + } + + int subresource = tex_p->array_slice >= 0 ? tex_p->array_slice : 0; + if (p->ctx1) { + ID3D11DeviceContext1_UpdateSubresource1(p->ctx1, tex_p->res, + subresource, prc, src, stride, pitch, + invalidate ? D3D11_COPY_DISCARD : 0); + } else { + ID3D11DeviceContext_UpdateSubresource(p->ctx, tex_p->res, subresource, + prc, src, stride, pitch); + } + + return true; +} + +static bool tex_download(struct ra *ra, struct ra_tex_download_params *params) +{ + struct ra_d3d11 *p = ra->priv; + struct ra_tex *tex = params->tex; + struct d3d_tex *tex_p = tex->priv; + HRESULT hr; + + if (!tex_p->staging) + return false; + + ID3D11DeviceContext_CopyResource(p->ctx, (ID3D11Resource*)tex_p->staging, + tex_p->res); + + D3D11_MAPPED_SUBRESOURCE lock; + hr = ID3D11DeviceContext_Map(p->ctx, (ID3D11Resource*)tex_p->staging, 0, + D3D11_MAP_READ, 0, &lock); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to map staging texture: %s\n", mp_HRESULT_to_str(hr)); + return false; + } + + char *cdst = params->dst; + char *csrc = lock.pData; + for (int y = 0; y < tex->params.h; y++) { + memcpy(cdst + y * params->stride, csrc + y * lock.RowPitch, + MPMIN(params->stride, lock.RowPitch)); + } + + ID3D11DeviceContext_Unmap(p->ctx, (ID3D11Resource*)tex_p->staging, 0); + + return true; +} + +static void buf_destroy(struct ra *ra, struct ra_buf *buf) +{ + if (!buf) + return; + struct d3d_buf *buf_p = buf->priv; + SAFE_RELEASE(buf_p->buf); + SAFE_RELEASE(buf_p->uav); + talloc_free(buf); +} + +static struct ra_buf *buf_create(struct ra *ra, + const struct ra_buf_params *params) +{ + // D3D11 does not support permanent mapping or pixel buffers + if (params->host_mapped || params->type == RA_BUF_TYPE_TEX_UPLOAD) + return NULL; + + struct ra_d3d11 *p = ra->priv; + HRESULT hr; + + struct ra_buf *buf = talloc_zero(NULL, struct ra_buf); + buf->params = *params; + buf->params.initial_data = NULL; + + struct d3d_buf *buf_p = buf->priv = talloc_zero(buf, struct d3d_buf); + + D3D11_SUBRESOURCE_DATA data; + D3D11_SUBRESOURCE_DATA *pdata = NULL; + if (params->initial_data) { + data = (D3D11_SUBRESOURCE_DATA) { .pSysMem = params->initial_data }; + pdata = &data; + } + + D3D11_BUFFER_DESC desc = { .ByteWidth = params->size }; + switch (params->type) { + case RA_BUF_TYPE_SHADER_STORAGE: + desc.BindFlags = D3D11_BIND_UNORDERED_ACCESS; + desc.ByteWidth = MP_ALIGN_UP(desc.ByteWidth, sizeof(float)); + desc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS; + break; + case RA_BUF_TYPE_UNIFORM: + desc.BindFlags = D3D11_BIND_CONSTANT_BUFFER; + desc.ByteWidth = MP_ALIGN_UP(desc.ByteWidth, sizeof(float[4])); + break; + } + + hr = ID3D11Device_CreateBuffer(p->dev, &desc, pdata, &buf_p->buf); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create buffer: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + + // D3D11 doesn't allow constant buffer updates that aren't aligned to a + // full constant boundary (vec4,) and some drivers don't allow partial + // constant buffer updates at all. To support partial buffer updates, keep + // a mirror of the buffer data in system memory and upload the whole thing + // before the buffer is used. + if (params->host_mutable) + buf_p->data = talloc_zero_size(buf, desc.ByteWidth); + + if (params->type == RA_BUF_TYPE_SHADER_STORAGE) { + D3D11_UNORDERED_ACCESS_VIEW_DESC udesc = { + .Format = DXGI_FORMAT_R32_TYPELESS, + .ViewDimension = D3D11_UAV_DIMENSION_BUFFER, + .Buffer = { + .NumElements = desc.ByteWidth / sizeof(float), + .Flags = D3D11_BUFFER_UAV_FLAG_RAW, + }, + }; + hr = ID3D11Device_CreateUnorderedAccessView(p->dev, + (ID3D11Resource *)buf_p->buf, &udesc, &buf_p->uav); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create UAV: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + } + + return buf; +error: + buf_destroy(ra, buf); + return NULL; +} + +static void buf_resolve(struct ra *ra, struct ra_buf *buf) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_buf *buf_p = buf->priv; + + if (!buf->params.host_mutable || !buf_p->dirty) + return; + + // Synchronize the GPU buffer with the system-memory copy + ID3D11DeviceContext_UpdateSubresource(p->ctx, (ID3D11Resource *)buf_p->buf, + 0, NULL, buf_p->data, 0, 0); + buf_p->dirty = false; +} + +static void buf_update(struct ra *ra, struct ra_buf *buf, ptrdiff_t offset, + const void *data, size_t size) +{ + struct d3d_buf *buf_p = buf->priv; + + char *cdata = buf_p->data; + memcpy(cdata + offset, data, size); + buf_p->dirty = true; +} + +static const char *get_shader_target(struct ra *ra, enum glsl_shader type) +{ + struct ra_d3d11 *p = ra->priv; + switch (p->fl) { + default: + switch (type) { + case GLSL_SHADER_VERTEX: return "vs_5_0"; + case GLSL_SHADER_FRAGMENT: return "ps_5_0"; + case GLSL_SHADER_COMPUTE: return "cs_5_0"; + } + break; + case D3D_FEATURE_LEVEL_10_1: + switch (type) { + case GLSL_SHADER_VERTEX: return "vs_4_1"; + case GLSL_SHADER_FRAGMENT: return "ps_4_1"; + case GLSL_SHADER_COMPUTE: return "cs_4_1"; + } + break; + case D3D_FEATURE_LEVEL_10_0: + switch (type) { + case GLSL_SHADER_VERTEX: return "vs_4_0"; + case GLSL_SHADER_FRAGMENT: return "ps_4_0"; + case GLSL_SHADER_COMPUTE: return "cs_4_0"; + } + break; + case D3D_FEATURE_LEVEL_9_3: + switch (type) { + case GLSL_SHADER_VERTEX: return "vs_4_0_level_9_3"; + case GLSL_SHADER_FRAGMENT: return "ps_4_0_level_9_3"; + } + break; + case D3D_FEATURE_LEVEL_9_2: + case D3D_FEATURE_LEVEL_9_1: + switch (type) { + case GLSL_SHADER_VERTEX: return "vs_4_0_level_9_1"; + case GLSL_SHADER_FRAGMENT: return "ps_4_0_level_9_1"; + } + break; + } + return NULL; +} + +static const char *shader_type_name(enum glsl_shader type) +{ + switch (type) { + case GLSL_SHADER_VERTEX: return "vertex"; + case GLSL_SHADER_FRAGMENT: return "fragment"; + case GLSL_SHADER_COMPUTE: return "compute"; + default: return "unknown"; + } +} + +static bool setup_clear_rpass(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + ID3DBlob *vs_blob = NULL; + ID3DBlob *ps_blob = NULL; + HRESULT hr; + + hr = p->D3DCompile(clear_vs, sizeof(clear_vs), NULL, NULL, NULL, "main", + get_shader_target(ra, GLSL_SHADER_VERTEX), + D3DCOMPILE_OPTIMIZATION_LEVEL3, 0, &vs_blob, NULL); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to compile clear() vertex shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = ID3D11Device_CreateVertexShader(p->dev, + ID3D10Blob_GetBufferPointer(vs_blob), ID3D10Blob_GetBufferSize(vs_blob), + NULL, &p->clear_vs); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create clear() vertex shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = p->D3DCompile(clear_ps, sizeof(clear_ps), NULL, NULL, NULL, "main", + get_shader_target(ra, GLSL_SHADER_FRAGMENT), + D3DCOMPILE_OPTIMIZATION_LEVEL3, 0, &ps_blob, NULL); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to compile clear() pixel shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = ID3D11Device_CreatePixelShader(p->dev, + ID3D10Blob_GetBufferPointer(ps_blob), ID3D10Blob_GetBufferSize(ps_blob), + NULL, &p->clear_ps); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create clear() pixel shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + D3D11_INPUT_ELEMENT_DESC in_descs[] = { + { "POSITION", 0, DXGI_FORMAT_R32G32_FLOAT, 0, 0 }, + }; + hr = ID3D11Device_CreateInputLayout(p->dev, in_descs, + MP_ARRAY_SIZE(in_descs), ID3D10Blob_GetBufferPointer(vs_blob), + ID3D10Blob_GetBufferSize(vs_blob), &p->clear_layout); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create clear() IA layout: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + // clear() always draws to a quad covering the whole viewport + static const float verts[] = { + -1, -1, + 1, -1, + 1, 1, + -1, 1, + -1, -1, + 1, 1, + }; + D3D11_BUFFER_DESC vdesc = { + .ByteWidth = sizeof(verts), + .Usage = D3D11_USAGE_IMMUTABLE, + .BindFlags = D3D11_BIND_VERTEX_BUFFER, + }; + D3D11_SUBRESOURCE_DATA vdata = { + .pSysMem = verts, + }; + hr = ID3D11Device_CreateBuffer(p->dev, &vdesc, &vdata, &p->clear_vbuf); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create clear() vertex buffer: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + D3D11_BUFFER_DESC cdesc = { + .ByteWidth = sizeof(float[4]), + .BindFlags = D3D11_BIND_CONSTANT_BUFFER, + }; + hr = ID3D11Device_CreateBuffer(p->dev, &cdesc, NULL, &p->clear_cbuf); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create clear() constant buffer: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(ps_blob); + return true; +error: + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(ps_blob); + return false; +} + +static void clear_rpass(struct ra *ra, struct ra_tex *tex, float color[4], + struct mp_rect *rc) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_tex *tex_p = tex->priv; + struct ra_tex_params *params = &tex->params; + + ID3D11DeviceContext_UpdateSubresource(p->ctx, + (ID3D11Resource *)p->clear_cbuf, 0, NULL, color, 0, 0); + + ID3D11DeviceContext_IASetInputLayout(p->ctx, p->clear_layout); + ID3D11DeviceContext_IASetVertexBuffers(p->ctx, 0, 1, &p->clear_vbuf, + &(UINT) { sizeof(float[2]) }, &(UINT) { 0 }); + ID3D11DeviceContext_IASetPrimitiveTopology(p->ctx, + D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST); + + ID3D11DeviceContext_VSSetShader(p->ctx, p->clear_vs, NULL, 0); + + ID3D11DeviceContext_RSSetViewports(p->ctx, 1, (&(D3D11_VIEWPORT) { + .Width = params->w, + .Height = params->h, + .MinDepth = 0, + .MaxDepth = 1, + })); + ID3D11DeviceContext_RSSetScissorRects(p->ctx, 1, (&(D3D11_RECT) { + .left = rc->x0, + .top = rc->y0, + .right = rc->x1, + .bottom = rc->y1, + })); + ID3D11DeviceContext_PSSetShader(p->ctx, p->clear_ps, NULL, 0); + ID3D11DeviceContext_PSSetConstantBuffers(p->ctx, 0, 1, &p->clear_cbuf); + + ID3D11DeviceContext_OMSetRenderTargets(p->ctx, 1, &tex_p->rtv, NULL); + ID3D11DeviceContext_OMSetBlendState(p->ctx, NULL, NULL, + D3D11_DEFAULT_SAMPLE_MASK); + + ID3D11DeviceContext_Draw(p->ctx, 6, 0); + + ID3D11DeviceContext_PSSetConstantBuffers(p->ctx, 0, 1, + &(ID3D11Buffer *){ NULL }); + ID3D11DeviceContext_OMSetRenderTargets(p->ctx, 0, NULL, NULL); +} + +static void clear(struct ra *ra, struct ra_tex *tex, float color[4], + struct mp_rect *rc) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_tex *tex_p = tex->priv; + struct ra_tex_params *params = &tex->params; + + if (!tex_p->rtv) + return; + + if (rc->x0 || rc->y0 || rc->x1 != params->w || rc->y1 != params->h) { + if (p->has_clear_view) { + ID3D11DeviceContext1_ClearView(p->ctx1, (ID3D11View *)tex_p->rtv, + color, (&(D3D11_RECT) { + .left = rc->x0, + .top = rc->y0, + .right = rc->x1, + .bottom = rc->y1, + }), 1); + } else { + clear_rpass(ra, tex, color, rc); + } + } else { + ID3D11DeviceContext_ClearRenderTargetView(p->ctx, tex_p->rtv, color); + } +} + +static bool setup_blit_rpass(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + ID3DBlob *vs_blob = NULL; + ID3DBlob *float_ps_blob = NULL; + HRESULT hr; + + hr = p->D3DCompile(blit_vs, sizeof(blit_vs), NULL, NULL, NULL, "main", + get_shader_target(ra, GLSL_SHADER_VERTEX), + D3DCOMPILE_OPTIMIZATION_LEVEL3, 0, &vs_blob, NULL); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to compile blit() vertex shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = ID3D11Device_CreateVertexShader(p->dev, + ID3D10Blob_GetBufferPointer(vs_blob), ID3D10Blob_GetBufferSize(vs_blob), + NULL, &p->blit_vs); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blit() vertex shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = p->D3DCompile(blit_float_ps, sizeof(blit_float_ps), NULL, NULL, NULL, + "main", get_shader_target(ra, GLSL_SHADER_FRAGMENT), + D3DCOMPILE_OPTIMIZATION_LEVEL3, 0, &float_ps_blob, NULL); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to compile blit() pixel shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + hr = ID3D11Device_CreatePixelShader(p->dev, + ID3D10Blob_GetBufferPointer(float_ps_blob), + ID3D10Blob_GetBufferSize(float_ps_blob), + NULL, &p->blit_float_ps); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blit() pixel shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + D3D11_INPUT_ELEMENT_DESC in_descs[] = { + { "POSITION", 0, DXGI_FORMAT_R32G32_FLOAT, 0, 0 }, + { "TEXCOORD", 0, DXGI_FORMAT_R32G32_FLOAT, 0, 8 }, + }; + hr = ID3D11Device_CreateInputLayout(p->dev, in_descs, + MP_ARRAY_SIZE(in_descs), ID3D10Blob_GetBufferPointer(vs_blob), + ID3D10Blob_GetBufferSize(vs_blob), &p->blit_layout); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blit() IA layout: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + D3D11_BUFFER_DESC vdesc = { + .ByteWidth = sizeof(struct blit_vert[6]), + .Usage = D3D11_USAGE_DEFAULT, + .BindFlags = D3D11_BIND_VERTEX_BUFFER, + }; + hr = ID3D11Device_CreateBuffer(p->dev, &vdesc, NULL, &p->blit_vbuf); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blit() vertex buffer: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + // Blit always uses point sampling, regardless of the source texture + D3D11_SAMPLER_DESC sdesc = { + .AddressU = D3D11_TEXTURE_ADDRESS_CLAMP, + .AddressV = D3D11_TEXTURE_ADDRESS_CLAMP, + .AddressW = D3D11_TEXTURE_ADDRESS_CLAMP, + .ComparisonFunc = D3D11_COMPARISON_NEVER, + .MinLOD = 0, + .MaxLOD = D3D11_FLOAT32_MAX, + .MaxAnisotropy = 1, + }; + hr = ID3D11Device_CreateSamplerState(p->dev, &sdesc, &p->blit_sampler); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blit() sampler: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(float_ps_blob); + return true; +error: + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(float_ps_blob); + return false; +} + +static void blit_rpass(struct ra *ra, struct ra_tex *dst, struct ra_tex *src, + struct mp_rect *dst_rc, struct mp_rect *src_rc) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_tex *dst_p = dst->priv; + struct d3d_tex *src_p = src->priv; + + float u_min = (double)src_rc->x0 / src->params.w; + float u_max = (double)src_rc->x1 / src->params.w; + float v_min = (double)src_rc->y0 / src->params.h; + float v_max = (double)src_rc->y1 / src->params.h; + + struct blit_vert verts[6] = { + { .x = -1, .y = -1, .u = u_min, .v = v_max }, + { .x = 1, .y = -1, .u = u_max, .v = v_max }, + { .x = 1, .y = 1, .u = u_max, .v = v_min }, + { .x = -1, .y = 1, .u = u_min, .v = v_min }, + }; + verts[4] = verts[0]; + verts[5] = verts[2]; + ID3D11DeviceContext_UpdateSubresource(p->ctx, + (ID3D11Resource *)p->blit_vbuf, 0, NULL, verts, 0, 0); + + ID3D11DeviceContext_IASetInputLayout(p->ctx, p->blit_layout); + ID3D11DeviceContext_IASetVertexBuffers(p->ctx, 0, 1, &p->blit_vbuf, + &(UINT) { sizeof(verts[0]) }, &(UINT) { 0 }); + ID3D11DeviceContext_IASetPrimitiveTopology(p->ctx, + D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST); + + ID3D11DeviceContext_VSSetShader(p->ctx, p->blit_vs, NULL, 0); + + ID3D11DeviceContext_RSSetViewports(p->ctx, 1, (&(D3D11_VIEWPORT) { + .TopLeftX = dst_rc->x0, + .TopLeftY = dst_rc->y0, + .Width = mp_rect_w(*dst_rc), + .Height = mp_rect_h(*dst_rc), + .MinDepth = 0, + .MaxDepth = 1, + })); + ID3D11DeviceContext_RSSetScissorRects(p->ctx, 1, (&(D3D11_RECT) { + .left = dst_rc->x0, + .top = dst_rc->y0, + .right = dst_rc->x1, + .bottom = dst_rc->y1, + })); + + ID3D11DeviceContext_PSSetShader(p->ctx, p->blit_float_ps, NULL, 0); + ID3D11DeviceContext_PSSetShaderResources(p->ctx, 0, 1, &src_p->srv); + ID3D11DeviceContext_PSSetSamplers(p->ctx, 0, 1, &p->blit_sampler); + + ID3D11DeviceContext_OMSetRenderTargets(p->ctx, 1, &dst_p->rtv, NULL); + ID3D11DeviceContext_OMSetBlendState(p->ctx, NULL, NULL, + D3D11_DEFAULT_SAMPLE_MASK); + + ID3D11DeviceContext_Draw(p->ctx, 6, 0); + + ID3D11DeviceContext_PSSetShaderResources(p->ctx, 0, 1, + &(ID3D11ShaderResourceView *) { NULL }); + ID3D11DeviceContext_PSSetSamplers(p->ctx, 0, 1, + &(ID3D11SamplerState *) { NULL }); + ID3D11DeviceContext_OMSetRenderTargets(p->ctx, 0, NULL, NULL); +} + +static void blit(struct ra *ra, struct ra_tex *dst, struct ra_tex *src, + struct mp_rect *dst_rc_ptr, struct mp_rect *src_rc_ptr) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_tex *dst_p = dst->priv; + struct d3d_tex *src_p = src->priv; + struct mp_rect dst_rc = *dst_rc_ptr; + struct mp_rect src_rc = *src_rc_ptr; + + assert(dst->params.dimensions == 2); + assert(src->params.dimensions == 2); + + // A zero-sized target rectangle is a no-op + if (!mp_rect_w(dst_rc) || !mp_rect_h(dst_rc)) + return; + + // ra.h seems to imply that both dst_rc and src_rc can be flipped, but it's + // easier for blit_rpass() if only src_rc can be flipped, so unflip dst_rc. + if (dst_rc.x0 > dst_rc.x1) { + MPSWAP(int, dst_rc.x0, dst_rc.x1); + MPSWAP(int, src_rc.x0, src_rc.x1); + } + if (dst_rc.y0 > dst_rc.y1) { + MPSWAP(int, dst_rc.y0, dst_rc.y1); + MPSWAP(int, src_rc.y0, src_rc.y1); + } + + // If format conversion, stretching or flipping is required, a renderpass + // must be used + if (dst->params.format != src->params.format || + mp_rect_w(dst_rc) != mp_rect_w(src_rc) || + mp_rect_h(dst_rc) != mp_rect_h(src_rc)) + { + blit_rpass(ra, dst, src, &dst_rc, &src_rc); + } else { + int dst_sr = dst_p->array_slice >= 0 ? dst_p->array_slice : 0; + int src_sr = src_p->array_slice >= 0 ? src_p->array_slice : 0; + ID3D11DeviceContext_CopySubresourceRegion(p->ctx, dst_p->res, dst_sr, + dst_rc.x0, dst_rc.y0, 0, src_p->res, src_sr, (&(D3D11_BOX) { + .left = src_rc.x0, + .top = src_rc.y0, + .front = 0, + .right = src_rc.x1, + .bottom = src_rc.y1, + .back = 1, + })); + } +} + +static int desc_namespace(struct ra *ra, enum ra_vartype type) +{ + // Images and SSBOs both use UAV bindings + if (type == RA_VARTYPE_IMG_W) + type = RA_VARTYPE_BUF_RW; + return type; +} + +static bool compile_glsl(struct ra *ra, enum glsl_shader type, + const char *glsl, ID3DBlob **out) +{ + struct ra_d3d11 *p = ra->priv; + struct spirv_compiler *spirv = p->spirv; + void *ta_ctx = talloc_new(NULL); + spvc_result sc_res = SPVC_SUCCESS; + spvc_context sc_ctx = NULL; + spvc_parsed_ir sc_ir = NULL; + spvc_compiler sc_compiler = NULL; + spvc_compiler_options sc_opts = NULL; + const char *hlsl = NULL; + ID3DBlob *errors = NULL; + bool success = false; + HRESULT hr; + + int sc_shader_model; + if (p->fl >= D3D_FEATURE_LEVEL_11_0) { + sc_shader_model = 50; + } else if (p->fl >= D3D_FEATURE_LEVEL_10_1) { + sc_shader_model = 41; + } else { + sc_shader_model = 40; + } + + int64_t start_ns = mp_time_ns(); + + bstr spv_module; + if (!spirv->fns->compile_glsl(spirv, ta_ctx, type, glsl, &spv_module)) + goto done; + + int64_t shaderc_ns = mp_time_ns(); + + sc_res = spvc_context_create(&sc_ctx); + if (sc_res != SPVC_SUCCESS) + goto done; + + sc_res = spvc_context_parse_spirv(sc_ctx, (SpvId *)spv_module.start, + spv_module.len / sizeof(SpvId), &sc_ir); + if (sc_res != SPVC_SUCCESS) + goto done; + + sc_res = spvc_context_create_compiler(sc_ctx, SPVC_BACKEND_HLSL, sc_ir, + SPVC_CAPTURE_MODE_TAKE_OWNERSHIP, + &sc_compiler); + if (sc_res != SPVC_SUCCESS) + goto done; + + sc_res = spvc_compiler_create_compiler_options(sc_compiler, &sc_opts); + if (sc_res != SPVC_SUCCESS) + goto done; + sc_res = spvc_compiler_options_set_uint(sc_opts, + SPVC_COMPILER_OPTION_HLSL_SHADER_MODEL, sc_shader_model); + if (sc_res != SPVC_SUCCESS) + goto done; + if (type == GLSL_SHADER_VERTEX) { + // FLIP_VERTEX_Y is only valid for vertex shaders + sc_res = spvc_compiler_options_set_bool(sc_opts, + SPVC_COMPILER_OPTION_FLIP_VERTEX_Y, SPVC_TRUE); + if (sc_res != SPVC_SUCCESS) + goto done; + } + sc_res = spvc_compiler_install_compiler_options(sc_compiler, sc_opts); + if (sc_res != SPVC_SUCCESS) + goto done; + + sc_res = spvc_compiler_compile(sc_compiler, &hlsl); + if (sc_res != SPVC_SUCCESS) + goto done; + + int64_t cross_ns = mp_time_ns(); + + hr = p->D3DCompile(hlsl, strlen(hlsl), NULL, NULL, NULL, "main", + get_shader_target(ra, type), D3DCOMPILE_OPTIMIZATION_LEVEL3, 0, out, + &errors); + if (FAILED(hr)) { + MP_ERR(ra, "D3DCompile failed: %s\n%.*s", mp_HRESULT_to_str(hr), + (int)ID3D10Blob_GetBufferSize(errors), + (char*)ID3D10Blob_GetBufferPointer(errors)); + goto done; + } + + int64_t d3dcompile_ns = mp_time_ns(); + + MP_VERBOSE(ra, "Compiled a %s shader in %lldns\n", shader_type_name(type), + d3dcompile_ns - start_ns); + MP_VERBOSE(ra, "shaderc: %lldns, SPIRV-Cross: %lldns, D3DCompile: %lldns\n", + shaderc_ns - start_ns, + cross_ns - shaderc_ns, + d3dcompile_ns - cross_ns); + + success = true; +done: + if (sc_res != SPVC_SUCCESS) { + MP_MSG(ra, MSGL_ERR, "SPIRV-Cross failed: %s\n", + spvc_context_get_last_error_string(sc_ctx)); + } + int level = success ? MSGL_DEBUG : MSGL_ERR; + MP_MSG(ra, level, "GLSL source:\n"); + mp_log_source(ra->log, level, glsl); + if (hlsl) { + MP_MSG(ra, level, "HLSL source:\n"); + mp_log_source(ra->log, level, hlsl); + } + SAFE_RELEASE(errors); + if (sc_ctx) + spvc_context_destroy(sc_ctx); + talloc_free(ta_ctx); + return success; +} + +static void renderpass_destroy(struct ra *ra, struct ra_renderpass *pass) +{ + if (!pass) + return; + struct d3d_rpass *pass_p = pass->priv; + + SAFE_RELEASE(pass_p->vs); + SAFE_RELEASE(pass_p->ps); + SAFE_RELEASE(pass_p->cs); + SAFE_RELEASE(pass_p->layout); + SAFE_RELEASE(pass_p->bstate); + talloc_free(pass); +} + +static D3D11_BLEND map_ra_blend(enum ra_blend blend) +{ + switch (blend) { + default: + case RA_BLEND_ZERO: return D3D11_BLEND_ZERO; + case RA_BLEND_ONE: return D3D11_BLEND_ONE; + case RA_BLEND_SRC_ALPHA: return D3D11_BLEND_SRC_ALPHA; + case RA_BLEND_ONE_MINUS_SRC_ALPHA: return D3D11_BLEND_INV_SRC_ALPHA; + }; +} + +static size_t vbuf_upload(struct ra *ra, void *data, size_t size) +{ + struct ra_d3d11 *p = ra->priv; + HRESULT hr; + + // Arbitrary size limit in case there is an insane number of vertices + if (size > 1e9) { + MP_ERR(ra, "Vertex buffer is too large\n"); + return -1; + } + + // If the vertex data doesn't fit, realloc the vertex buffer + if (size > p->vbuf_size) { + size_t new_size = p->vbuf_size; + // Arbitrary base size + if (!new_size) + new_size = 64 * 1024; + while (new_size < size) + new_size *= 2; + + ID3D11Buffer *new_buf; + D3D11_BUFFER_DESC vbuf_desc = { + .ByteWidth = new_size, + .Usage = D3D11_USAGE_DYNAMIC, + .BindFlags = D3D11_BIND_VERTEX_BUFFER, + .CPUAccessFlags = D3D11_CPU_ACCESS_WRITE, + }; + hr = ID3D11Device_CreateBuffer(p->dev, &vbuf_desc, NULL, &new_buf); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create vertex buffer: %s\n", + mp_HRESULT_to_str(hr)); + return -1; + } + + SAFE_RELEASE(p->vbuf); + p->vbuf = new_buf; + p->vbuf_size = new_size; + p->vbuf_used = 0; + } + + bool discard = false; + size_t offset = p->vbuf_used; + if (offset + size > p->vbuf_size) { + // We reached the end of the buffer, so discard and wrap around + discard = true; + offset = 0; + } + + D3D11_MAPPED_SUBRESOURCE map = { 0 }; + hr = ID3D11DeviceContext_Map(p->ctx, (ID3D11Resource *)p->vbuf, 0, + discard ? D3D11_MAP_WRITE_DISCARD : D3D11_MAP_WRITE_NO_OVERWRITE, + 0, &map); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to map vertex buffer: %s\n", mp_HRESULT_to_str(hr)); + return -1; + } + + char *cdata = map.pData; + memcpy(cdata + offset, data, size); + + ID3D11DeviceContext_Unmap(p->ctx, (ID3D11Resource *)p->vbuf, 0); + + p->vbuf_used = offset + size; + return offset; +} + +static const char cache_magic[4] = "RD11"; +static const int cache_version = 3; + +struct cache_header { + char magic[sizeof(cache_magic)]; + int cache_version; + char compiler[SPIRV_NAME_MAX_LEN]; + int spv_compiler_version; + unsigned spvc_compiler_major; + unsigned spvc_compiler_minor; + unsigned spvc_compiler_patch; + struct dll_version d3d_compiler_version; + int feature_level; + size_t vert_bytecode_len; + size_t frag_bytecode_len; + size_t comp_bytecode_len; +}; + +static void load_cached_program(struct ra *ra, + const struct ra_renderpass_params *params, + bstr *vert_bc, + bstr *frag_bc, + bstr *comp_bc) +{ + struct ra_d3d11 *p = ra->priv; + struct spirv_compiler *spirv = p->spirv; + bstr cache = params->cached_program; + + if (cache.len < sizeof(struct cache_header)) + return; + + struct cache_header *header = (struct cache_header *)cache.start; + cache = bstr_cut(cache, sizeof(*header)); + + unsigned spvc_major, spvc_minor, spvc_patch; + spvc_get_version(&spvc_major, &spvc_minor, &spvc_patch); + + if (strncmp(header->magic, cache_magic, sizeof(cache_magic)) != 0) + return; + if (header->cache_version != cache_version) + return; + if (strncmp(header->compiler, spirv->name, sizeof(header->compiler)) != 0) + return; + if (header->spv_compiler_version != spirv->compiler_version) + return; + if (header->spvc_compiler_major != spvc_major) + return; + if (header->spvc_compiler_minor != spvc_minor) + return; + if (header->spvc_compiler_patch != spvc_patch) + return; + if (!dll_version_equal(header->d3d_compiler_version, p->d3d_compiler_ver)) + return; + if (header->feature_level != p->fl) + return; + + if (header->vert_bytecode_len && vert_bc) { + *vert_bc = bstr_splice(cache, 0, header->vert_bytecode_len); + MP_VERBOSE(ra, "Using cached vertex shader\n"); + } + cache = bstr_cut(cache, header->vert_bytecode_len); + + if (header->frag_bytecode_len && frag_bc) { + *frag_bc = bstr_splice(cache, 0, header->frag_bytecode_len); + MP_VERBOSE(ra, "Using cached fragment shader\n"); + } + cache = bstr_cut(cache, header->frag_bytecode_len); + + if (header->comp_bytecode_len && comp_bc) { + *comp_bc = bstr_splice(cache, 0, header->comp_bytecode_len); + MP_VERBOSE(ra, "Using cached compute shader\n"); + } + cache = bstr_cut(cache, header->comp_bytecode_len); +} + +static void save_cached_program(struct ra *ra, struct ra_renderpass *pass, + bstr vert_bc, + bstr frag_bc, + bstr comp_bc) +{ + struct ra_d3d11 *p = ra->priv; + struct spirv_compiler *spirv = p->spirv; + + unsigned spvc_major, spvc_minor, spvc_patch; + spvc_get_version(&spvc_major, &spvc_minor, &spvc_patch); + + struct cache_header header = { + .cache_version = cache_version, + .spv_compiler_version = p->spirv->compiler_version, + .spvc_compiler_major = spvc_major, + .spvc_compiler_minor = spvc_minor, + .spvc_compiler_patch = spvc_patch, + .d3d_compiler_version = p->d3d_compiler_ver, + .feature_level = p->fl, + .vert_bytecode_len = vert_bc.len, + .frag_bytecode_len = frag_bc.len, + .comp_bytecode_len = comp_bc.len, + }; + memcpy(header.magic, cache_magic, sizeof(header.magic)); + strncpy(header.compiler, spirv->name, sizeof(header.compiler)); + + struct bstr *prog = &pass->params.cached_program; + bstr_xappend(pass, prog, (bstr){ (char *) &header, sizeof(header) }); + bstr_xappend(pass, prog, vert_bc); + bstr_xappend(pass, prog, frag_bc); + bstr_xappend(pass, prog, comp_bc); +} + +static struct ra_renderpass *renderpass_create_raster(struct ra *ra, + struct ra_renderpass *pass, const struct ra_renderpass_params *params) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_rpass *pass_p = pass->priv; + ID3DBlob *vs_blob = NULL; + ID3DBlob *ps_blob = NULL; + HRESULT hr; + + // load_cached_program will load compiled shader bytecode into vert_bc and + // frag_bc if the cache is valid. If not, vert_bc/frag_bc will remain NULL. + bstr vert_bc = {0}; + bstr frag_bc = {0}; + load_cached_program(ra, params, &vert_bc, &frag_bc, NULL); + + if (!vert_bc.start) { + if (!compile_glsl(ra, GLSL_SHADER_VERTEX, params->vertex_shader, + &vs_blob)) + goto error; + vert_bc = (bstr){ + ID3D10Blob_GetBufferPointer(vs_blob), + ID3D10Blob_GetBufferSize(vs_blob), + }; + } + + hr = ID3D11Device_CreateVertexShader(p->dev, vert_bc.start, vert_bc.len, + NULL, &pass_p->vs); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create vertex shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + if (!frag_bc.start) { + if (!compile_glsl(ra, GLSL_SHADER_FRAGMENT, params->frag_shader, + &ps_blob)) + goto error; + frag_bc = (bstr){ + ID3D10Blob_GetBufferPointer(ps_blob), + ID3D10Blob_GetBufferSize(ps_blob), + }; + } + + hr = ID3D11Device_CreatePixelShader(p->dev, frag_bc.start, frag_bc.len, + NULL, &pass_p->ps); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create pixel shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + D3D11_INPUT_ELEMENT_DESC *in_descs = talloc_array(pass, + D3D11_INPUT_ELEMENT_DESC, params->num_vertex_attribs); + for (int i = 0; i < params->num_vertex_attribs; i++) { + struct ra_renderpass_input *inp = ¶ms->vertex_attribs[i]; + + DXGI_FORMAT fmt = DXGI_FORMAT_UNKNOWN; + switch (inp->type) { + case RA_VARTYPE_FLOAT: + switch (inp->dim_v) { + case 1: fmt = DXGI_FORMAT_R32_FLOAT; break; + case 2: fmt = DXGI_FORMAT_R32G32_FLOAT; break; + case 3: fmt = DXGI_FORMAT_R32G32B32_FLOAT; break; + case 4: fmt = DXGI_FORMAT_R32G32B32A32_FLOAT; break; + } + break; + case RA_VARTYPE_BYTE_UNORM: + switch (inp->dim_v) { + case 1: fmt = DXGI_FORMAT_R8_UNORM; break; + case 2: fmt = DXGI_FORMAT_R8G8_UNORM; break; + // There is no 3-component 8-bit DXGI format + case 4: fmt = DXGI_FORMAT_R8G8B8A8_UNORM; break; + } + break; + } + if (fmt == DXGI_FORMAT_UNKNOWN) { + MP_ERR(ra, "Could not find suitable vertex input format\n"); + goto error; + } + + in_descs[i] = (D3D11_INPUT_ELEMENT_DESC) { + // The semantic name doesn't mean much and is just used to verify + // the input description matches the shader. SPIRV-Cross always + // uses TEXCOORD, so we should too. + .SemanticName = "TEXCOORD", + .SemanticIndex = i, + .AlignedByteOffset = inp->offset, + .Format = fmt, + }; + } + + hr = ID3D11Device_CreateInputLayout(p->dev, in_descs, + params->num_vertex_attribs, vert_bc.start, vert_bc.len, + &pass_p->layout); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create IA layout: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + talloc_free(in_descs); + in_descs = NULL; + + D3D11_BLEND_DESC bdesc = { + .RenderTarget[0] = { + .BlendEnable = params->enable_blend, + .SrcBlend = map_ra_blend(params->blend_src_rgb), + .DestBlend = map_ra_blend(params->blend_dst_rgb), + .BlendOp = D3D11_BLEND_OP_ADD, + .SrcBlendAlpha = map_ra_blend(params->blend_src_alpha), + .DestBlendAlpha = map_ra_blend(params->blend_dst_alpha), + .BlendOpAlpha = D3D11_BLEND_OP_ADD, + .RenderTargetWriteMask = D3D11_COLOR_WRITE_ENABLE_ALL, + }, + }; + hr = ID3D11Device_CreateBlendState(p->dev, &bdesc, &pass_p->bstate); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create blend state: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + + save_cached_program(ra, pass, vert_bc, frag_bc, (bstr){0}); + + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(ps_blob); + return pass; + +error: + renderpass_destroy(ra, pass); + SAFE_RELEASE(vs_blob); + SAFE_RELEASE(ps_blob); + return NULL; +} + +static struct ra_renderpass *renderpass_create_compute(struct ra *ra, + struct ra_renderpass *pass, const struct ra_renderpass_params *params) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_rpass *pass_p = pass->priv; + ID3DBlob *cs_blob = NULL; + HRESULT hr; + + bstr comp_bc = {0}; + load_cached_program(ra, params, NULL, NULL, &comp_bc); + + if (!comp_bc.start) { + if (!compile_glsl(ra, GLSL_SHADER_COMPUTE, params->compute_shader, + &cs_blob)) + goto error; + comp_bc = (bstr){ + ID3D10Blob_GetBufferPointer(cs_blob), + ID3D10Blob_GetBufferSize(cs_blob), + }; + } + hr = ID3D11Device_CreateComputeShader(p->dev, comp_bc.start, comp_bc.len, + NULL, &pass_p->cs); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create compute shader: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + + save_cached_program(ra, pass, (bstr){0}, (bstr){0}, comp_bc); + + SAFE_RELEASE(cs_blob); + return pass; +error: + renderpass_destroy(ra, pass); + SAFE_RELEASE(cs_blob); + return NULL; +} + +static struct ra_renderpass *renderpass_create(struct ra *ra, + const struct ra_renderpass_params *params) +{ + struct ra_renderpass *pass = talloc_zero(NULL, struct ra_renderpass); + pass->params = *ra_renderpass_params_copy(pass, params); + pass->params.cached_program = (bstr){0}; + pass->priv = talloc_zero(pass, struct d3d_rpass); + + if (params->type == RA_RENDERPASS_TYPE_COMPUTE) { + return renderpass_create_compute(ra, pass, params); + } else { + return renderpass_create_raster(ra, pass, params); + } +} + +static void renderpass_run_raster(struct ra *ra, + const struct ra_renderpass_run_params *params, + ID3D11Buffer *ubos[], int ubos_len, + ID3D11SamplerState *samplers[], + ID3D11ShaderResourceView *srvs[], + int samplers_len, + ID3D11UnorderedAccessView *uavs[], + int uavs_len) +{ + struct ra_d3d11 *p = ra->priv; + struct ra_renderpass *pass = params->pass; + struct d3d_rpass *pass_p = pass->priv; + + UINT vbuf_offset = vbuf_upload(ra, params->vertex_data, + pass->params.vertex_stride * params->vertex_count); + if (vbuf_offset == (UINT)-1) + return; + + ID3D11DeviceContext_IASetInputLayout(p->ctx, pass_p->layout); + ID3D11DeviceContext_IASetVertexBuffers(p->ctx, 0, 1, &p->vbuf, + &pass->params.vertex_stride, &vbuf_offset); + ID3D11DeviceContext_IASetPrimitiveTopology(p->ctx, + D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST); + + ID3D11DeviceContext_VSSetShader(p->ctx, pass_p->vs, NULL, 0); + + ID3D11DeviceContext_RSSetViewports(p->ctx, 1, (&(D3D11_VIEWPORT) { + .TopLeftX = params->viewport.x0, + .TopLeftY = params->viewport.y0, + .Width = mp_rect_w(params->viewport), + .Height = mp_rect_h(params->viewport), + .MinDepth = 0, + .MaxDepth = 1, + })); + ID3D11DeviceContext_RSSetScissorRects(p->ctx, 1, (&(D3D11_RECT) { + .left = params->scissors.x0, + .top = params->scissors.y0, + .right = params->scissors.x1, + .bottom = params->scissors.y1, + })); + ID3D11DeviceContext_PSSetShader(p->ctx, pass_p->ps, NULL, 0); + ID3D11DeviceContext_PSSetConstantBuffers(p->ctx, 0, ubos_len, ubos); + ID3D11DeviceContext_PSSetShaderResources(p->ctx, 0, samplers_len, srvs); + ID3D11DeviceContext_PSSetSamplers(p->ctx, 0, samplers_len, samplers); + + struct ra_tex *target = params->target; + struct d3d_tex *target_p = target->priv; + ID3D11DeviceContext_OMSetRenderTargetsAndUnorderedAccessViews(p->ctx, 1, + &target_p->rtv, NULL, 1, uavs_len, uavs, NULL); + ID3D11DeviceContext_OMSetBlendState(p->ctx, pass_p->bstate, NULL, + D3D11_DEFAULT_SAMPLE_MASK); + + ID3D11DeviceContext_Draw(p->ctx, params->vertex_count, 0); + + // Unbind everything. It's easier to do this than to actually track state, + // and if we leave the RTV bound, it could trip up D3D's conflict checker. + for (int i = 0; i < ubos_len; i++) + ubos[i] = NULL; + for (int i = 0; i < samplers_len; i++) { + samplers[i] = NULL; + srvs[i] = NULL; + } + for (int i = 0; i < uavs_len; i++) + uavs[i] = NULL; + ID3D11DeviceContext_PSSetConstantBuffers(p->ctx, 0, ubos_len, ubos); + ID3D11DeviceContext_PSSetShaderResources(p->ctx, 0, samplers_len, srvs); + ID3D11DeviceContext_PSSetSamplers(p->ctx, 0, samplers_len, samplers); + ID3D11DeviceContext_OMSetRenderTargetsAndUnorderedAccessViews(p->ctx, 0, + NULL, NULL, 1, uavs_len, uavs, NULL); +} + +static void renderpass_run_compute(struct ra *ra, + const struct ra_renderpass_run_params *params, + ID3D11Buffer *ubos[], int ubos_len, + ID3D11SamplerState *samplers[], + ID3D11ShaderResourceView *srvs[], + int samplers_len, + ID3D11UnorderedAccessView *uavs[], + int uavs_len) +{ + struct ra_d3d11 *p = ra->priv; + struct ra_renderpass *pass = params->pass; + struct d3d_rpass *pass_p = pass->priv; + + ID3D11DeviceContext_CSSetShader(p->ctx, pass_p->cs, NULL, 0); + ID3D11DeviceContext_CSSetConstantBuffers(p->ctx, 0, ubos_len, ubos); + ID3D11DeviceContext_CSSetShaderResources(p->ctx, 0, samplers_len, srvs); + ID3D11DeviceContext_CSSetSamplers(p->ctx, 0, samplers_len, samplers); + ID3D11DeviceContext_CSSetUnorderedAccessViews(p->ctx, 0, uavs_len, uavs, + NULL); + + ID3D11DeviceContext_Dispatch(p->ctx, params->compute_groups[0], + params->compute_groups[1], + params->compute_groups[2]); + + for (int i = 0; i < ubos_len; i++) + ubos[i] = NULL; + for (int i = 0; i < samplers_len; i++) { + samplers[i] = NULL; + srvs[i] = NULL; + } + for (int i = 0; i < uavs_len; i++) + uavs[i] = NULL; + ID3D11DeviceContext_CSSetConstantBuffers(p->ctx, 0, ubos_len, ubos); + ID3D11DeviceContext_CSSetShaderResources(p->ctx, 0, samplers_len, srvs); + ID3D11DeviceContext_CSSetSamplers(p->ctx, 0, samplers_len, samplers); + ID3D11DeviceContext_CSSetUnorderedAccessViews(p->ctx, 0, uavs_len, uavs, + NULL); +} + +static void renderpass_run(struct ra *ra, + const struct ra_renderpass_run_params *params) +{ + struct ra_d3d11 *p = ra->priv; + struct ra_renderpass *pass = params->pass; + enum ra_renderpass_type type = pass->params.type; + + ID3D11Buffer *ubos[D3D11_COMMONSHADER_CONSTANT_BUFFER_API_SLOT_COUNT] = {0}; + int ubos_len = 0; + + ID3D11SamplerState *samplers[D3D11_COMMONSHADER_SAMPLER_SLOT_COUNT] = {0}; + ID3D11ShaderResourceView *srvs[D3D11_COMMONSHADER_SAMPLER_SLOT_COUNT] = {0}; + int samplers_len = 0; + + ID3D11UnorderedAccessView *uavs[D3D11_1_UAV_SLOT_COUNT] = {0}; + int uavs_len = 0; + + // In a raster pass, one of the UAV slots is used by the runtime for the RTV + int uavs_max = type == RA_RENDERPASS_TYPE_COMPUTE ? p->max_uavs + : p->max_uavs - 1; + + // Gather the input variables used in this pass. These will be mapped to + // HLSL registers. + for (int i = 0; i < params->num_values; i++) { + struct ra_renderpass_input_val *val = ¶ms->values[i]; + int binding = pass->params.inputs[val->index].binding; + switch (pass->params.inputs[val->index].type) { + case RA_VARTYPE_BUF_RO: + if (binding >= MP_ARRAY_SIZE(ubos)) { + MP_ERR(ra, "Too many constant buffers in pass\n"); + return; + } + struct ra_buf *buf_ro = *(struct ra_buf **)val->data; + buf_resolve(ra, buf_ro); + struct d3d_buf *buf_ro_p = buf_ro->priv; + ubos[binding] = buf_ro_p->buf; + ubos_len = MPMAX(ubos_len, binding + 1); + break; + case RA_VARTYPE_BUF_RW: + if (binding > uavs_max) { + MP_ERR(ra, "Too many UAVs in pass\n"); + return; + } + struct ra_buf *buf_rw = *(struct ra_buf **)val->data; + buf_resolve(ra, buf_rw); + struct d3d_buf *buf_rw_p = buf_rw->priv; + uavs[binding] = buf_rw_p->uav; + uavs_len = MPMAX(uavs_len, binding + 1); + break; + case RA_VARTYPE_TEX: + if (binding >= MP_ARRAY_SIZE(samplers)) { + MP_ERR(ra, "Too many textures in pass\n"); + return; + } + struct ra_tex *tex = *(struct ra_tex **)val->data; + struct d3d_tex *tex_p = tex->priv; + samplers[binding] = tex_p->sampler; + srvs[binding] = tex_p->srv; + samplers_len = MPMAX(samplers_len, binding + 1); + break; + case RA_VARTYPE_IMG_W: + if (binding > uavs_max) { + MP_ERR(ra, "Too many UAVs in pass\n"); + return; + } + struct ra_tex *img = *(struct ra_tex **)val->data; + struct d3d_tex *img_p = img->priv; + uavs[binding] = img_p->uav; + uavs_len = MPMAX(uavs_len, binding + 1); + break; + } + } + + if (type == RA_RENDERPASS_TYPE_COMPUTE) { + renderpass_run_compute(ra, params, ubos, ubos_len, samplers, srvs, + samplers_len, uavs, uavs_len); + } else { + renderpass_run_raster(ra, params, ubos, ubos_len, samplers, srvs, + samplers_len, uavs, uavs_len); + } +} + +static void timer_destroy(struct ra *ra, ra_timer *ratimer) +{ + if (!ratimer) + return; + struct d3d_timer *timer = ratimer; + + SAFE_RELEASE(timer->ts_start); + SAFE_RELEASE(timer->ts_end); + SAFE_RELEASE(timer->disjoint); + talloc_free(timer); +} + +static ra_timer *timer_create(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + if (!p->has_timestamp_queries) + return NULL; + + struct d3d_timer *timer = talloc_zero(NULL, struct d3d_timer); + HRESULT hr; + + hr = ID3D11Device_CreateQuery(p->dev, + &(D3D11_QUERY_DESC) { D3D11_QUERY_TIMESTAMP }, &timer->ts_start); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create start query: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + + hr = ID3D11Device_CreateQuery(p->dev, + &(D3D11_QUERY_DESC) { D3D11_QUERY_TIMESTAMP }, &timer->ts_end); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create end query: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + + // Measuring duration in D3D11 requires three queries: start and end + // timestamps, and a disjoint query containing a flag which says whether + // the timestamps are usable or if a discontinuity occurred between them, + // like a change in power state or clock speed. The disjoint query also + // contains the timer frequency, so the timestamps are useless without it. + hr = ID3D11Device_CreateQuery(p->dev, + &(D3D11_QUERY_DESC) { D3D11_QUERY_TIMESTAMP_DISJOINT }, &timer->disjoint); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create timer query: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + + return timer; +error: + timer_destroy(ra, timer); + return NULL; +} + +static uint64_t timestamp_to_ns(uint64_t timestamp, uint64_t freq) +{ + static const uint64_t ns_per_s = 1000000000llu; + return timestamp / freq * ns_per_s + timestamp % freq * ns_per_s / freq; +} + +static uint64_t timer_get_result(struct ra *ra, ra_timer *ratimer) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_timer *timer = ratimer; + HRESULT hr; + + UINT64 start, end; + D3D11_QUERY_DATA_TIMESTAMP_DISJOINT dj; + + hr = ID3D11DeviceContext_GetData(p->ctx, + (ID3D11Asynchronous *)timer->ts_end, &end, sizeof(end), + D3D11_ASYNC_GETDATA_DONOTFLUSH); + if (FAILED(hr) || hr == S_FALSE) + return 0; + hr = ID3D11DeviceContext_GetData(p->ctx, + (ID3D11Asynchronous *)timer->ts_start, &start, sizeof(start), + D3D11_ASYNC_GETDATA_DONOTFLUSH); + if (FAILED(hr) || hr == S_FALSE) + return 0; + hr = ID3D11DeviceContext_GetData(p->ctx, + (ID3D11Asynchronous *)timer->disjoint, &dj, sizeof(dj), + D3D11_ASYNC_GETDATA_DONOTFLUSH); + if (FAILED(hr) || hr == S_FALSE || dj.Disjoint || !dj.Frequency) + return 0; + + return timestamp_to_ns(end - start, dj.Frequency); +} + +static void timer_start(struct ra *ra, ra_timer *ratimer) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_timer *timer = ratimer; + + // Latch the last result of this ra_timer (returned by timer_stop) + timer->result = timer_get_result(ra, ratimer); + + ID3D11DeviceContext_Begin(p->ctx, (ID3D11Asynchronous *)timer->disjoint); + ID3D11DeviceContext_End(p->ctx, (ID3D11Asynchronous *)timer->ts_start); +} + +static uint64_t timer_stop(struct ra *ra, ra_timer *ratimer) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_timer *timer = ratimer; + + ID3D11DeviceContext_End(p->ctx, (ID3D11Asynchronous *)timer->ts_end); + ID3D11DeviceContext_End(p->ctx, (ID3D11Asynchronous *)timer->disjoint); + + return timer->result; +} + +static int map_msg_severity(D3D11_MESSAGE_SEVERITY sev) +{ + switch (sev) { + case D3D11_MESSAGE_SEVERITY_CORRUPTION: + return MSGL_FATAL; + case D3D11_MESSAGE_SEVERITY_ERROR: + return MSGL_ERR; + case D3D11_MESSAGE_SEVERITY_WARNING: + return MSGL_WARN; + default: + case D3D11_MESSAGE_SEVERITY_INFO: + case D3D11_MESSAGE_SEVERITY_MESSAGE: + return MSGL_DEBUG; + } +} + +static int map_msg_severity_by_id(D3D11_MESSAGE_ID id, + D3D11_MESSAGE_SEVERITY sev) +{ + switch (id) { + // These are normal. The RA timer queue habitually reuses timer objects + // without retrieving the results. + case D3D11_MESSAGE_ID_QUERY_BEGIN_ABANDONING_PREVIOUS_RESULTS: + case D3D11_MESSAGE_ID_QUERY_END_ABANDONING_PREVIOUS_RESULTS: + return MSGL_TRACE; + + // D3D11 writes log messages every time an object is created or + // destroyed. That results in a lot of log spam, so force MSGL_TRACE. +#define OBJ_LIFETIME_MESSAGES(obj) \ + case D3D11_MESSAGE_ID_CREATE_ ## obj: \ + case D3D11_MESSAGE_ID_DESTROY_ ## obj + + OBJ_LIFETIME_MESSAGES(CONTEXT): + OBJ_LIFETIME_MESSAGES(BUFFER): + OBJ_LIFETIME_MESSAGES(TEXTURE1D): + OBJ_LIFETIME_MESSAGES(TEXTURE2D): + OBJ_LIFETIME_MESSAGES(TEXTURE3D): + OBJ_LIFETIME_MESSAGES(SHADERRESOURCEVIEW): + OBJ_LIFETIME_MESSAGES(RENDERTARGETVIEW): + OBJ_LIFETIME_MESSAGES(DEPTHSTENCILVIEW): + OBJ_LIFETIME_MESSAGES(VERTEXSHADER): + OBJ_LIFETIME_MESSAGES(HULLSHADER): + OBJ_LIFETIME_MESSAGES(DOMAINSHADER): + OBJ_LIFETIME_MESSAGES(GEOMETRYSHADER): + OBJ_LIFETIME_MESSAGES(PIXELSHADER): + OBJ_LIFETIME_MESSAGES(INPUTLAYOUT): + OBJ_LIFETIME_MESSAGES(SAMPLER): + OBJ_LIFETIME_MESSAGES(BLENDSTATE): + OBJ_LIFETIME_MESSAGES(DEPTHSTENCILSTATE): + OBJ_LIFETIME_MESSAGES(RASTERIZERSTATE): + OBJ_LIFETIME_MESSAGES(QUERY): + OBJ_LIFETIME_MESSAGES(PREDICATE): + OBJ_LIFETIME_MESSAGES(COUNTER): + OBJ_LIFETIME_MESSAGES(COMMANDLIST): + OBJ_LIFETIME_MESSAGES(CLASSINSTANCE): + OBJ_LIFETIME_MESSAGES(CLASSLINKAGE): + OBJ_LIFETIME_MESSAGES(COMPUTESHADER): + OBJ_LIFETIME_MESSAGES(UNORDEREDACCESSVIEW): + OBJ_LIFETIME_MESSAGES(VIDEODECODER): + OBJ_LIFETIME_MESSAGES(VIDEOPROCESSORENUM): + OBJ_LIFETIME_MESSAGES(VIDEOPROCESSOR): + OBJ_LIFETIME_MESSAGES(DECODEROUTPUTVIEW): + OBJ_LIFETIME_MESSAGES(PROCESSORINPUTVIEW): + OBJ_LIFETIME_MESSAGES(PROCESSOROUTPUTVIEW): + OBJ_LIFETIME_MESSAGES(DEVICECONTEXTSTATE): + OBJ_LIFETIME_MESSAGES(FENCE): + return MSGL_TRACE; + +#undef OBJ_LIFETIME_MESSAGES + + default: + return map_msg_severity(sev); + } +} + +static void debug_marker(struct ra *ra, const char *msg) +{ + struct ra_d3d11 *p = ra->priv; + void *talloc_ctx = talloc_new(NULL); + HRESULT hr; + + if (!p->iqueue) + goto done; + + // Copy debug-layer messages to mpv's log output + bool printed_header = false; + uint64_t messages = ID3D11InfoQueue_GetNumStoredMessages(p->iqueue); + for (uint64_t i = 0; i < messages; i++) { + SIZE_T len; + hr = ID3D11InfoQueue_GetMessage(p->iqueue, i, NULL, &len); + if (FAILED(hr) || !len) + goto done; + + D3D11_MESSAGE *d3dmsg = talloc_size(talloc_ctx, len); + hr = ID3D11InfoQueue_GetMessage(p->iqueue, i, d3dmsg, &len); + if (FAILED(hr)) + goto done; + + int msgl = map_msg_severity_by_id(d3dmsg->ID, d3dmsg->Severity); + if (mp_msg_test(ra->log, msgl)) { + if (!printed_header) + MP_INFO(ra, "%s:\n", msg); + printed_header = true; + + MP_MSG(ra, msgl, "%d: %.*s\n", (int)d3dmsg->ID, + (int)d3dmsg->DescriptionByteLength, d3dmsg->pDescription); + talloc_free(d3dmsg); + } + } + + ID3D11InfoQueue_ClearStoredMessages(p->iqueue); +done: + talloc_free(talloc_ctx); +} + +static void destroy(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + + // Release everything except the interfaces needed to perform leak checking + SAFE_RELEASE(p->clear_ps); + SAFE_RELEASE(p->clear_vs); + SAFE_RELEASE(p->clear_layout); + SAFE_RELEASE(p->clear_vbuf); + SAFE_RELEASE(p->clear_cbuf); + SAFE_RELEASE(p->blit_float_ps); + SAFE_RELEASE(p->blit_vs); + SAFE_RELEASE(p->blit_layout); + SAFE_RELEASE(p->blit_vbuf); + SAFE_RELEASE(p->blit_sampler); + SAFE_RELEASE(p->vbuf); + SAFE_RELEASE(p->ctx1); + SAFE_RELEASE(p->dev1); + SAFE_RELEASE(p->dev); + + if (p->ctx) { + // Destroy the device context synchronously so referenced objects don't + // show up in the leak check + ID3D11DeviceContext_ClearState(p->ctx); + ID3D11DeviceContext_Flush(p->ctx); + } + SAFE_RELEASE(p->ctx); + + if (p->debug) { + // Report any leaked objects + debug_marker(ra, "after destroy"); + ID3D11Debug_ReportLiveDeviceObjects(p->debug, D3D11_RLDO_DETAIL); + debug_marker(ra, "after leak check"); + ID3D11Debug_ReportLiveDeviceObjects(p->debug, D3D11_RLDO_SUMMARY); + debug_marker(ra, "after leak summary"); + } + SAFE_RELEASE(p->debug); + SAFE_RELEASE(p->iqueue); + + talloc_free(ra); +} + +static struct ra_fns ra_fns_d3d11 = { + .destroy = destroy, + .tex_create = tex_create, + .tex_destroy = tex_destroy, + .tex_upload = tex_upload, + .tex_download = tex_download, + .buf_create = buf_create, + .buf_destroy = buf_destroy, + .buf_update = buf_update, + .clear = clear, + .blit = blit, + .uniform_layout = std140_layout, + .desc_namespace = desc_namespace, + .renderpass_create = renderpass_create, + .renderpass_destroy = renderpass_destroy, + .renderpass_run = renderpass_run, + .timer_create = timer_create, + .timer_destroy = timer_destroy, + .timer_start = timer_start, + .timer_stop = timer_stop, + .debug_marker = debug_marker, +}; + +void ra_d3d11_flush(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + ID3D11DeviceContext_Flush(p->ctx); +} + +static void init_debug_layer(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + HRESULT hr; + + hr = ID3D11Device_QueryInterface(p->dev, &IID_ID3D11Debug, + (void**)&p->debug); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to get debug device: %s\n", mp_HRESULT_to_str(hr)); + return; + } + + hr = ID3D11Device_QueryInterface(p->dev, &IID_ID3D11InfoQueue, + (void**)&p->iqueue); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to get info queue: %s\n", mp_HRESULT_to_str(hr)); + return; + } + + // Store an unlimited amount of messages in the buffer. This is fine + // because we flush stored messages regularly (in debug_marker.) + ID3D11InfoQueue_SetMessageCountLimit(p->iqueue, -1); + + // Push empty filter to get everything + D3D11_INFO_QUEUE_FILTER filter = {0}; + ID3D11InfoQueue_PushStorageFilter(p->iqueue, &filter); +} + +static struct dll_version get_dll_version(HMODULE dll) +{ + void *ctx = talloc_new(NULL); + struct dll_version ret = { 0 }; + + HRSRC rsrc = FindResourceW(dll, MAKEINTRESOURCEW(VS_VERSION_INFO), + VS_FILE_INFO); + if (!rsrc) + goto done; + DWORD size = SizeofResource(dll, rsrc); + HGLOBAL res = LoadResource(dll, rsrc); + if (!res) + goto done; + void *ptr = LockResource(res); + if (!ptr) + goto done; + void *copy = talloc_memdup(ctx, ptr, size); + + VS_FIXEDFILEINFO *ffi; + UINT ffi_len; + if (!VerQueryValueW(copy, L"\\", (void**)&ffi, &ffi_len)) + goto done; + if (ffi_len < sizeof(*ffi)) + goto done; + + ret.major = HIWORD(ffi->dwFileVersionMS); + ret.minor = LOWORD(ffi->dwFileVersionMS); + ret.build = HIWORD(ffi->dwFileVersionLS); + ret.revision = LOWORD(ffi->dwFileVersionLS); + +done: + talloc_free(ctx); + return ret; +} + +static bool load_d3d_compiler(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + HMODULE d3dcompiler = NULL; + + // Try the inbox D3DCompiler first (Windows 8.1 and up) + if (IsWindows8Point1OrGreater()) { + d3dcompiler = LoadLibraryExW(L"d3dcompiler_47.dll", NULL, + LOAD_LIBRARY_SEARCH_SYSTEM32); + } + // Check for a packaged version of d3dcompiler_47.dll + if (!d3dcompiler) + d3dcompiler = LoadLibraryW(L"d3dcompiler_47.dll"); + // Try d3dcompiler_46.dll from the Windows 8 SDK + if (!d3dcompiler) + d3dcompiler = LoadLibraryW(L"d3dcompiler_46.dll"); + // Try d3dcompiler_43.dll from the June 2010 DirectX SDK + if (!d3dcompiler) + d3dcompiler = LoadLibraryW(L"d3dcompiler_43.dll"); + // Can't find any compiler DLL, so give up + if (!d3dcompiler) + return false; + + p->d3d_compiler_ver = get_dll_version(d3dcompiler); + + p->D3DCompile = (pD3DCompile)GetProcAddress(d3dcompiler, "D3DCompile"); + if (!p->D3DCompile) + return false; + return true; +} + +static void find_max_texture_dimension(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + + D3D11_TEXTURE2D_DESC desc = { + .Width = ra->max_texture_wh, + .Height = ra->max_texture_wh, + .MipLevels = 1, + .ArraySize = 1, + .SampleDesc.Count = 1, + .Format = DXGI_FORMAT_R8_UNORM, + .BindFlags = D3D11_BIND_SHADER_RESOURCE, + }; + while (true) { + desc.Height = desc.Width *= 2; + if (desc.Width >= 0x8000000u) + return; + if (FAILED(ID3D11Device_CreateTexture2D(p->dev, &desc, NULL, NULL))) + return; + ra->max_texture_wh = desc.Width; + } +} + +struct ra *ra_d3d11_create(ID3D11Device *dev, struct mp_log *log, + struct spirv_compiler *spirv) +{ + HRESULT hr; + + struct ra *ra = talloc_zero(NULL, struct ra); + ra->log = log; + ra->fns = &ra_fns_d3d11; + + // Even Direct3D 10level9 supports 3D textures + ra->caps = RA_CAP_TEX_3D | RA_CAP_DIRECT_UPLOAD | RA_CAP_BUF_RO | + RA_CAP_BLIT | spirv->ra_caps; + + ra->glsl_version = spirv->glsl_version; + ra->glsl_vulkan = true; + + struct ra_d3d11 *p = ra->priv = talloc_zero(ra, struct ra_d3d11); + p->spirv = spirv; + + int minor = 0; + ID3D11Device_AddRef(dev); + p->dev = dev; + ID3D11Device_GetImmediateContext(p->dev, &p->ctx); + hr = ID3D11Device_QueryInterface(p->dev, &IID_ID3D11Device1, + (void**)&p->dev1); + if (SUCCEEDED(hr)) { + minor = 1; + ID3D11Device1_GetImmediateContext1(p->dev1, &p->ctx1); + + D3D11_FEATURE_DATA_D3D11_OPTIONS fopts = { 0 }; + hr = ID3D11Device_CheckFeatureSupport(p->dev, + D3D11_FEATURE_D3D11_OPTIONS, &fopts, sizeof(fopts)); + if (SUCCEEDED(hr)) { + p->has_clear_view = fopts.ClearView; + } + } + + MP_VERBOSE(ra, "Using Direct3D 11.%d runtime\n", minor); + + p->fl = ID3D11Device_GetFeatureLevel(p->dev); + if (p->fl >= D3D_FEATURE_LEVEL_11_0) { + ra->max_texture_wh = D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION; + } else if (p->fl >= D3D_FEATURE_LEVEL_10_0) { + ra->max_texture_wh = D3D10_REQ_TEXTURE2D_U_OR_V_DIMENSION; + } else if (p->fl >= D3D_FEATURE_LEVEL_9_3) { + ra->max_texture_wh = D3D_FL9_3_REQ_TEXTURE2D_U_OR_V_DIMENSION; + } else { + ra->max_texture_wh = D3D_FL9_1_REQ_TEXTURE2D_U_OR_V_DIMENSION; + } + + if (p->fl >= D3D_FEATURE_LEVEL_11_0) + ra->caps |= RA_CAP_GATHER; + if (p->fl >= D3D_FEATURE_LEVEL_10_0) + ra->caps |= RA_CAP_FRAGCOORD; + + // Some 10_0 hardware has compute shaders, but only 11_0 has image load/store + if (p->fl >= D3D_FEATURE_LEVEL_11_0) { + ra->caps |= RA_CAP_COMPUTE | RA_CAP_BUF_RW; + ra->max_shmem = 32 * 1024; + ra->max_compute_group_threads = + D3D11_CS_THREAD_GROUP_MAX_THREADS_PER_GROUP; + } + + if (p->fl >= D3D_FEATURE_LEVEL_11_1) { + p->max_uavs = D3D11_1_UAV_SLOT_COUNT; + } else { + p->max_uavs = D3D11_PS_CS_UAV_REGISTER_COUNT; + } + + if (ID3D11Device_GetCreationFlags(p->dev) & D3D11_CREATE_DEVICE_DEBUG) + init_debug_layer(ra); + + // Some level 9_x devices don't have timestamp queries + hr = ID3D11Device_CreateQuery(p->dev, + &(D3D11_QUERY_DESC) { D3D11_QUERY_TIMESTAMP }, NULL); + p->has_timestamp_queries = SUCCEEDED(hr); + + debug_marker(ra, "before maximum Texture2D size lookup"); + + // According to MSDN, the above texture sizes are just minimums and drivers + // may support larger textures. See: + // https://msdn.microsoft.com/en-us/library/windows/desktop/ff476874.aspx + find_max_texture_dimension(ra); + + // Ignore any messages during find_max_texture_dimension + if (p->iqueue) + ID3D11InfoQueue_ClearStoredMessages(p->iqueue); + + MP_VERBOSE(ra, "Maximum Texture2D size: %dx%d\n", ra->max_texture_wh, + ra->max_texture_wh); + + if (!load_d3d_compiler(ra)) { + MP_FATAL(ra, "Could not find D3DCompiler DLL\n"); + goto error; + } + + MP_VERBOSE(ra, "D3DCompiler version: %u.%u.%u.%u\n", + p->d3d_compiler_ver.major, p->d3d_compiler_ver.minor, + p->d3d_compiler_ver.build, p->d3d_compiler_ver.revision); + + setup_formats(ra); + + // The rasterizer state never changes, so set it up here + ID3D11RasterizerState *rstate; + D3D11_RASTERIZER_DESC rdesc = { + .FillMode = D3D11_FILL_SOLID, + .CullMode = D3D11_CULL_NONE, + .FrontCounterClockwise = FALSE, + .DepthClipEnable = TRUE, // Required for 10level9 + .ScissorEnable = TRUE, + }; + hr = ID3D11Device_CreateRasterizerState(p->dev, &rdesc, &rstate); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create rasterizer state: %s\n", mp_HRESULT_to_str(hr)); + goto error; + } + ID3D11DeviceContext_RSSetState(p->ctx, rstate); + SAFE_RELEASE(rstate); + + // If the device doesn't support ClearView, we have to set up a + // shader-based clear() implementation + if (!p->has_clear_view && !setup_clear_rpass(ra)) + goto error; + + if (!setup_blit_rpass(ra)) + goto error; + + return ra; + +error: + destroy(ra); + return NULL; +} + +ID3D11Device *ra_d3d11_get_device(struct ra *ra) +{ + struct ra_d3d11 *p = ra->priv; + ID3D11Device_AddRef(p->dev); + return p->dev; +} + +bool ra_is_d3d11(struct ra *ra) +{ + return ra->fns == &ra_fns_d3d11; +} diff --git a/video/out/d3d11/ra_d3d11.h b/video/out/d3d11/ra_d3d11.h new file mode 100644 index 0000000..6f62a7f --- /dev/null +++ b/video/out/d3d11/ra_d3d11.h @@ -0,0 +1,47 @@ +#pragma once + +#include <stdbool.h> +#include <windows.h> +#include <d3d11.h> +#include <dxgi1_2.h> + +#include "video/out/gpu/ra.h" +#include "video/out/gpu/spirv.h" + +// Get the underlying DXGI format from an RA format +DXGI_FORMAT ra_d3d11_get_format(const struct ra_format *fmt); + +// Gets the matching ra_format for a given DXGI format. +// Returns a nullptr in case of no known match. +const struct ra_format *ra_d3d11_get_ra_format(struct ra *ra, DXGI_FORMAT fmt); + +// Create an RA instance from a D3D11 device. This takes a reference to the +// device, which is released when the RA instance is destroyed. +struct ra *ra_d3d11_create(ID3D11Device *device, struct mp_log *log, + struct spirv_compiler *spirv); + +// Flush the immediate context of the wrapped D3D11 device +void ra_d3d11_flush(struct ra *ra); + +// Create an RA texture from a D3D11 resource. This takes a reference to the +// texture, which is released when the RA texture is destroyed. +struct ra_tex *ra_d3d11_wrap_tex(struct ra *ra, ID3D11Resource *res); + +// As above, but for a D3D11VA video resource. The fmt parameter selects which +// plane of a planar format will be mapped when the RA texture is used. +// array_slice should be set for texture arrays and is ignored for non-arrays. +struct ra_tex *ra_d3d11_wrap_tex_video(struct ra *ra, ID3D11Texture2D *res, + int w, int h, int array_slice, + const struct ra_format *fmt); + +// Get the underlying D3D11 resource from an RA texture. The returned resource +// is refcounted and must be released by the caller. +ID3D11Resource *ra_d3d11_get_raw_tex(struct ra *ra, struct ra_tex *tex, + int *array_slice); + +// Get the underlying D3D11 device from an RA instance. The returned device is +// refcounted and must be released by the caller. +ID3D11Device *ra_d3d11_get_device(struct ra *ra); + +// True if the RA instance was created with ra_d3d11_create() +bool ra_is_d3d11(struct ra *ra); |