[host] dxgi: use CopySubresourceRegion when possible

This commit adds damage tracking to the DXGI textures, and only copies the
damaged areas to the textures with ID3D11DeviceContext::CopySubresourceRegion.

The sleep logic in waitFrame makes it difficult for this to reduce the
latency, but removing it shows significant improvements (6-7 ms to ~3 ms)
when a tiny portion of the screen is damaged, while showing no difference on
full screen damage.
This commit is contained in:
Quantum 2021-08-11 02:43:42 -04:00 committed by Geoffrey McRae
parent 7e982a6658
commit 9bded74543

View File

@ -40,7 +40,7 @@
#include "dxgi_extra.h" #include "dxgi_extra.h"
#define LOCKED(x) INTERLOCKED_SECTION(this->deviceContextLock, x) #define LOCKED(...) INTERLOCKED_SECTION(this->deviceContextLock, __VA_ARGS__)
enum TextureState enum TextureState
{ {
@ -56,8 +56,10 @@ typedef struct Texture
ID3D11Texture2D * tex; ID3D11Texture2D * tex;
D3D11_MAPPED_SUBRESOURCE map; D3D11_MAPPED_SUBRESOURCE map;
uint64_t copyTime; uint64_t copyTime;
uint8_t damageRectsCount; uint32_t damageRectsCount;
FrameDamageRect damageRects[KVMFR_MAX_DAMAGE_RECTS]; FrameDamageRect damageRects[KVMFR_MAX_DAMAGE_RECTS];
int32_t texDamageCount;
FrameDamageRect texDamageRects[KVMFR_MAX_DAMAGE_RECTS];
} }
Texture; Texture;
@ -556,6 +558,7 @@ static bool dxgi_init(void)
for (int i = 0; i < this->maxTextures; ++i) for (int i = 0; i < this->maxTextures; ++i)
{ {
this->texture[i].texDamageCount = -1;
status = ID3D11Device_CreateTexture2D(this->device, &texDesc, NULL, &this->texture[i].tex); status = ID3D11Device_CreateTexture2D(this->device, &texDesc, NULL, &this->texture[i].tex);
if (FAILED(status)) if (FAILED(status))
{ {
@ -859,10 +862,36 @@ static CaptureResult dxgi_capture(void)
{ {
computeFrameDamage(tex); computeFrameDamage(tex);
if (tex->texDamageCount < 0 || tex->damageRectsCount == 0 ||
tex->texDamageCount + tex->damageRectsCount > KVMFR_MAX_DAMAGE_RECTS)
tex->texDamageCount = -1;
else
{
memcpy(tex->texDamageRects + tex->texDamageCount, tex->damageRects,
tex->damageRectsCount * sizeof(FrameDamageRect));
tex->texDamageCount += tex->damageRectsCount;
tex->texDamageCount = rectsMergeOverlapping(tex->texDamageRects, tex->texDamageCount);
}
// issue the copy from GPU to CPU RAM // issue the copy from GPU to CPU RAM
tex->copyTime = microtime(); tex->copyTime = microtime();
if (tex->texDamageCount < 0)
ID3D11DeviceContext_CopyResource(this->deviceContext, ID3D11DeviceContext_CopyResource(this->deviceContext,
(ID3D11Resource *)tex->tex, (ID3D11Resource *)src); (ID3D11Resource *)tex->tex, (ID3D11Resource *)src);
else
{
for (int i = 0; i < tex->texDamageCount; ++i)
{
FrameDamageRect * rect = tex->texDamageRects + i;
D3D11_BOX box = {
.left = rect->x, .top = rect->y, .front = 0, .back = 1,
.right = rect->x + rect->width, .bottom = rect->y + rect->height,
};
ID3D11DeviceContext_CopySubresourceRegion(this->deviceContext,
(ID3D11Resource *)tex->tex, 0, rect->x, rect->y, 0,
(ID3D11Resource *)src, 0, &box);
}
}
} }
if (copyPointer) if (copyPointer)
@ -879,6 +908,22 @@ static CaptureResult dxgi_capture(void)
{ {
ID3D11Texture2D_Release(src); ID3D11Texture2D_Release(src);
for (int i = 0; i < this->maxTextures; ++i)
{
Texture * t = this->texture + i;
if (i == this->texWIndex)
t->texDamageCount = 0;
else if (tex->damageRectsCount > 0 && t->texDamageCount >= 0 &&
t->texDamageCount + tex->damageRectsCount <= KVMFR_MAX_DAMAGE_RECTS)
{
memcpy(t->texDamageRects + t->texDamageCount, tex->damageRects,
tex->damageRectsCount * sizeof(FrameDamageRect));
t->texDamageCount += tex->damageRectsCount;
}
else
t->texDamageCount = -1;
}
// set the state, and signal // set the state, and signal
tex->state = TEXTURE_STATE_PENDING_MAP; tex->state = TEXTURE_STATE_PENDING_MAP;
tex->formatVer = this->formatVer; tex->formatVer = this->formatVer;