joejulian/gio
1
0
Fork 1
mirror of https://git.sr.ht/~eliasnaur/gio synced 2026-07-06 09:55:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

gpu/internal/d3d11,internal/d3d11: add DirectCompute support

Browse Source 
This change implements support for compute programs in the Direct3D 11
driver. The compute renderer doesn't work for me yet; my NVIDIA GTX 970
and Intel GPUs both display corrupted output and hangs.

Signed-off-by: Elias Naur <mail@eliasnaur.com>
This commit is contained in:
Elias Naur
2021-08-28 10:18:03 +02:00
parent e841288b88
commit 1362acb85a
4 changed files with 327 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files
internal/d3d11/d3d11_windows.go
+151 -8
View File
@@ -74,16 +74,52 @@ type SHADER_RESOURCE_VIEW_DESC_TEX2D struct {
Texture2D TEX2D_SRV
}
type SHADER_RESOURCE_VIEW_DESC_BUFFEREX struct {
SHADER_RESOURCE_VIEW_DESC
Buffer BUFFEREX_SRV
}
type UNORDERED_ACCESS_VIEW_DESC_TEX2D struct {
UNORDERED_ACCESS_VIEW_DESC
Texture2D TEX2D_UAV
}
type UNORDERED_ACCESS_VIEW_DESC_BUFFER struct {
UNORDERED_ACCESS_VIEW_DESC
Buffer BUFFER_UAV
}
type SHADER_RESOURCE_VIEW_DESC struct {
Format uint32
ViewDimension uint32
}
type UNORDERED_ACCESS_VIEW_DESC struct {
Format uint32
ViewDimension uint32
}
type TEX2D_SRV struct {
MostDetailedMip uint32
MipLevels uint32
}
type BUFFEREX_SRV struct {
FirstElement uint32
NumElements uint32
Flags uint32
}
type TEX2D_UAV struct {
MipSlice uint32
}
type BUFFER_UAV struct {
FirstElement uint32
NumElements uint32
Flags uint32
}
type INPUT_ELEMENT_DESC struct {
SemanticName *byte
SemanticIndex uint32
@@ -336,6 +372,12 @@ type ShaderResourceView struct {
}
}
type UnorderedAccessView struct {
Vtbl *struct {
_IUnknownVTbl
}
}
type DepthStencilView struct {
Vtbl *struct {
_IUnknownVTbl
@@ -360,6 +402,12 @@ type VertexShader struct {
}
}
type ComputeShader struct {
Vtbl *struct {
_IUnknownVTbl
}
}
type RasterizerState struct {
Vtbl *struct {
_IUnknownVTbl
@@ -590,9 +638,11 @@ const (
DXGI_FORMAT_UNKNOWN = 0
DXGI_FORMAT_R16_FLOAT = 54
DXGI_FORMAT_R32_FLOAT = 41
DXGI_FORMAT_R32_TYPELESS = 39
DXGI_FORMAT_R32G32_FLOAT = 16
DXGI_FORMAT_R32G32B32_FLOAT = 6
DXGI_FORMAT_R32G32B32A32_FLOAT = 2
DXGI_FORMAT_R8G8B8A8_UNORM = 28
DXGI_FORMAT_R8G8B8A8_UNORM_SRGB = 29
DXGI_FORMAT_R16_SINT = 59
DXGI_FORMAT_R16G16_SINT = 38
@@ -623,12 +673,13 @@ const (
USAGE_IMMUTABLE = 1
USAGE_STAGING = 3
BIND_VERTEX_BUFFER = 0x1
BIND_INDEX_BUFFER = 0x2
BIND_CONSTANT_BUFFER = 0x4
BIND_SHADER_RESOURCE = 0x8
BIND_RENDER_TARGET = 0x20
BIND_DEPTH_STENCIL = 0x40
BIND_VERTEX_BUFFER = 0x1
BIND_INDEX_BUFFER = 0x2
BIND_CONSTANT_BUFFER = 0x4
BIND_SHADER_RESOURCE = 0x8
BIND_RENDER_TARGET = 0x20
BIND_DEPTH_STENCIL = 0x40
BIND_UNORDERED_ACCESS = 0x80
PRIMITIVE_TOPOLOGY_TRIANGLELIST = 4
PRIMITIVE_TOPOLOGY_TRIANGLESTRIP = 5
@@ -640,7 +691,16 @@ const (
TEXTURE_ADDRESS_CLAMP = 3
TEXTURE_ADDRESS_WRAP = 1
SRV_DIMENSION_BUFFER = 1
UAV_DIMENSION_BUFFER = 1
SRV_DIMENSION_BUFFEREX = 11
SRV_DIMENSION_TEXTURE2D = 4
UAV_DIMENSION_TEXTURE2D = 4
BUFFER_UAV_FLAG_RAW = 0x1
BUFFEREX_SRV_FLAG_RAW = 0x1
RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS = 0x20
CREATE_DEVICE_DEBUG = 0x2
@@ -855,14 +915,32 @@ func (d *Device) CreateVertexShader(bytecode []byte) (*VertexShader, error) {
return shader, nil
}
func (d *Device) CreateShaderResourceViewTEX2D(res *Resource, desc *SHADER_RESOURCE_VIEW_DESC_TEX2D) (*ShaderResourceView, error) {
func (d *Device) CreateComputeShader(bytecode []byte) (*ComputeShader, error) {
var shader *ComputeShader
r, _, _ := syscall.Syscall6(
d.Vtbl.CreateComputeShader,
5,
uintptr(unsafe.Pointer(d)),
uintptr(unsafe.Pointer(&bytecode[0])),
uintptr(len(bytecode)),
0, // pClassLinkage
uintptr(unsafe.Pointer(&shader)),
0,
)
if r != 0 {
return nil, ErrorCode{Name: "DeviceCreateComputeShader", Code: uint32(r)}
}
return shader, nil
}
func (d *Device) CreateShaderResourceView(res *Resource, desc unsafe.Pointer) (*ShaderResourceView, error) {
var resView *ShaderResourceView
r, _, _ := syscall.Syscall6(
d.Vtbl.CreateShaderResourceView,
4,
uintptr(unsafe.Pointer(d)),
uintptr(unsafe.Pointer(res)),
uintptr(unsafe.Pointer(desc)),
uintptr(desc),
uintptr(unsafe.Pointer(&resView)),
0, 0,
)
@@ -872,6 +950,23 @@ func (d *Device) CreateShaderResourceViewTEX2D(res *Resource, desc *SHADER_RESOU
return resView, nil
}
func (d *Device) CreateUnorderedAccessView(res *Resource, desc unsafe.Pointer) (*UnorderedAccessView, error) {
var uaView *UnorderedAccessView
r, _, _ := syscall.Syscall6(
d.Vtbl.CreateUnorderedAccessView,
4,
uintptr(unsafe.Pointer(d)),
uintptr(unsafe.Pointer(res)),
uintptr(desc),
uintptr(unsafe.Pointer(&uaView)),
0, 0,
)
if r != 0 {
return nil, ErrorCode{Name: "DeviceCreateUnorderedAccessView", Code: uint32(r)}
}
return uaView, nil
}
func (d *Device) CreateRasterizerState(desc *RASTERIZER_DESC) (*RasterizerState, error) {
var state *RasterizerState
r, _, _ := syscall.Syscall(
@@ -1162,6 +1257,42 @@ func (c *DeviceContext) ClearRenderTargetView(target *RenderTargetView, color *[
)
}
func (c *DeviceContext) CSSetShaderResources(startSlot uint32, s *ShaderResourceView) {
syscall.Syscall6(
c.Vtbl.CSSetShaderResources,
4,
uintptr(unsafe.Pointer(c)),
uintptr(startSlot),
1, // NumViews
uintptr(unsafe.Pointer(&s)),
0, 0,
)
}
func (c *DeviceContext) CSSetUnorderedAccessViews(startSlot uint32, v *UnorderedAccessView) {
syscall.Syscall6(
c.Vtbl.CSSetUnorderedAccessViews,
4,
uintptr(unsafe.Pointer(c)),
uintptr(startSlot),
1, // NumViews
uintptr(unsafe.Pointer(&v)),
0, 0,
)
}
func (c *DeviceContext) CSSetShader(s *ComputeShader) {
syscall.Syscall6(
c.Vtbl.CSSetShader,
4,
uintptr(unsafe.Pointer(c)),
uintptr(unsafe.Pointer(s)),
0, // ppClassInstances
0, // NumClassInstances
0, 0,
)
}
func (c *DeviceContext) RSSetViewports(viewport *VIEWPORT) {
syscall.Syscall(
c.Vtbl.RSSetViewports,
@@ -1365,6 +1496,18 @@ func (c *DeviceContext) DrawIndexed(count, start uint32, base int32) {
)
}
func (c *DeviceContext) Dispatch(x, y, z uint32) {
syscall.Syscall6(
c.Vtbl.Dispatch,
4,
uintptr(unsafe.Pointer(c)),
uintptr(x),
uintptr(y),
uintptr(z),
0, 0,
)
}
func (c *DeviceContext) OMSetBlendState(state *BlendState, factor *f32color.RGBA, sampleMask uint32) {
syscall.Syscall6(
c.Vtbl.OMSetBlendState,