diff options
Diffstat (limited to 'video/out/d3d11/ra_d3d11.c')
| -rw-r--r-- | video/out/d3d11/ra_d3d11.c | 2235 |
1 files changed, 2235 insertions, 0 deletions
diff --git a/video/out/d3d11/ra_d3d11.c b/video/out/d3d11/ra_d3d11.c new file mode 100644 index 0000000000..372b65d49f --- /dev/null +++ b/video/out/d3d11/ra_d3d11.c @@ -0,0 +1,2235 @@ +#include <windows.h> +#include <versionhelpers.h> +#include <d3d11_1.h> +#include <d3d11sdklayers.h> +#include <dxgi1_2.h> +#include <d3dcompiler.h> +#include <crossc.h> + +#include "common/msg.h" +#include "osdep/io.h" +#include "osdep/subprocess.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 + +struct ra_d3d11 { + struct spirv_compiler *spirv; + + ID3D11Device *dev; + ID3D11Device1 *dev1; + ID3D11DeviceContext *ctx; + ID3D11DeviceContext1 *ctx1; + pD3DCompile D3DCompile; + + // Debug interfaces (--gpu-debug) + ID3D11Debug *debug; + ID3D11InfoQueue *iqueue; + + // Device capabilities + D3D_FEATURE_LEVEL fl; + bool has_clear_view; + 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; + + ID3D11ShaderResourceView *srv; + ID3D11RenderTargetView *rtv; + ID3D11UnorderedAccessView *uav; + ID3D11SamplerState *sampler; +}; + +struct d3d_buf { + ID3D11Buffer *buf; + ID3D11Buffer *staging; + ID3D11UnorderedAccessView *uav; + void *data; // Data for mapped staging texture +}; + +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 = false }, +}; + +static DXGI_FORMAT fmt_to_dxgi(const struct ra_format *fmt) +{ + struct d3d_fmt *d3d = fmt->priv; + return d3d->fmt; +} + +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; + + 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; + + 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, + }; + + 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 = fmt_to_dxgi(params->format), + }; + switch (params->dimensions) { + case 1: + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE1D; + srvdesc.Texture1D.MipLevels = 1; + break; + case 2: + srvdesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D; + srvdesc.Texture2D.MipLevels = 1; + break; + case 3: + 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); + talloc_free(tex); +} + +static struct ra_tex *tex_create(struct ra *ra, + const struct ra_tex_params *params) +{ + 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 = fmt_to_dxgi(params->format); + + D3D11_SUBRESOURCE_DATA *pdata = NULL; + if (params->initial_data) { + pdata = &(D3D11_SUBRESOURCE_DATA) { + .pSysMem = params->initial_data, + .SysMemPitch = params->w * params->format->pixel_size, + }; + if (params->dimensions >= 3) + pdata->SysMemSlicePitch = pdata->SysMemPitch * params->h; + } + + 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; + 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: + abort(); + } + + 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 || desc2d.ArraySize != 1) + goto error; + if (desc2d.SampleDesc.Count != 1) + 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 = fmt_to_dxgi(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; + 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; + } + + 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, + 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 (!tex_init(ra, tex)) + goto error; + + return tex; +error: + tex_destroy(ra, tex); + return NULL; +} + +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 = NULL; + + if (tex->params.dimensions == 2) { + stride = params->stride; + + // stride can be negative, but vo_gpu expects the RA backend to ignore + // the negative stride and upload the image "upside-down" for now + if (stride < 0) { + int h = params->rc ? mp_rect_h(*params->rc) : tex->params.h; + src += (h - 1) * stride; + stride = -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, + }; + invalidate = params->invalidate; + } + } + + if (p->ctx1) { + ID3D11DeviceContext1_UpdateSubresource1(p->ctx1, tex_p->res, 0, rc, + src, stride, pitch, invalidate ? D3D11_COPY_DISCARD : 0); + } else { + ID3D11DeviceContext_UpdateSubresource(p->ctx, tex_p->res, 0, rc, + src, stride, pitch); + } + + return true; +} + +static void buf_destroy(struct ra *ra, struct ra_buf *buf) +{ + if (!buf) + return; + struct ra_d3d11 *p = ra->priv; + struct d3d_buf *buf_p = buf->priv; + + if (buf_p->data) + ID3D11DeviceContext_Unmap(p->ctx, (ID3D11Resource *)buf_p->staging, 0); + SAFE_RELEASE(buf_p->buf); + SAFE_RELEASE(buf_p->staging); + 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 *pdata = NULL; + if (params->initial_data) + pdata = &(D3D11_SUBRESOURCE_DATA) { .pSysMem = params->initial_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; + } + + if (params->host_mutable) { + // 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, but the RA consumer is allowed to + // partially update an ra_buf. The best way to handle partial updates + // without causing a pipeline stall is probably to keep a copy of the + // data in a staging buffer. + + desc.Usage = D3D11_USAGE_STAGING; + desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE; + desc.BindFlags = 0; + hr = ID3D11Device_CreateBuffer(p->dev, &desc, NULL, &buf_p->staging); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to create staging buffer: %s\n", + mp_HRESULT_to_str(hr)); + goto error; + } + } + + 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; + + assert(buf->params.host_mutable); + if (!buf_p->data) + return; + + ID3D11DeviceContext_Unmap(p->ctx, (ID3D11Resource *)buf_p->staging, 0); + buf_p->data = NULL; + + // Synchronize the GPU buffer with the staging buffer + ID3D11DeviceContext_CopyResource(p->ctx, (ID3D11Resource *)buf_p->buf, + (ID3D11Resource *)buf_p->staging); +} + +static void buf_update(struct ra *ra, struct ra_buf *buf, ptrdiff_t offset, + const void *data, size_t size) +{ + struct ra_d3d11 *p = ra->priv; + struct d3d_buf *buf_p = buf->priv; + HRESULT hr; + + if (!buf_p->data) { + // If this is the first update after the buffer was created or after it + // has been used in a renderpass, it will be unmapped, so map it + D3D11_MAPPED_SUBRESOURCE map = {0}; + hr = ID3D11DeviceContext_Map(p->ctx, (ID3D11Resource *)buf_p->staging, + 0, D3D11_MAP_WRITE, 0, &map); + if (FAILED(hr)) { + MP_ERR(ra, "Failed to map resource\n"); + return; + } + buf_p->data = map.pData; + } + + char *cdata = buf_p->data; + memcpy(cdata + offset, data, size); +} + +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 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 = { |