11e6c1f478
Add dynamic sampler writes into frame capture record
2016-10-21 21:44:24 +02:00
6060bb9fcb
Fix indirect enum order
2016-10-21 21:44:24 +02:00
c23633b85d
Handle ExecuteIndirect by cracking lists apart and patching during read
2016-10-21 21:44:24 +02:00
df903cd191
Allocate indirect buffers for storing immutable patched copies of args
2016-10-21 21:44:24 +02:00
5978305794
Optionally skip adding event usage (if it will be done post-patching)
2016-10-21 21:44:24 +02:00
9f319e035e
Store list ID and relative EID for each drawcall use, for lookup later
2016-10-21 21:44:24 +02:00
9a6a05bf6d
Sure type and nodemask when resetting a list to allow cracking
2016-10-21 21:44:24 +02:00
5921a50bc4
Close readback list after creation, so we can just always reset it
2016-10-21 21:44:24 +02:00
12d9bd511a
Remove unused chunk type parameter from AddEvent() calls
2016-10-21 21:44:24 +02:00
bd73731ba8
Execute command lists one-by-one during reading
2016-10-21 21:44:24 +02:00
8dffc24383
Record command lists twice, so they can be cracked apart around executes
2016-10-21 21:44:24 +02:00
aeb5eb76f8
Track complete state vector during reading pass
...
* This will be needed for cracking apart command lists with indirect
executes.
2016-10-21 21:44:24 +02:00
19685c80a4
Handle replaying of ExecuteIndirect non-indirectly and add fake events
2016-10-21 21:44:24 +02:00
52232ad5b8
For now, skip replaying query functions
2016-10-21 21:44:24 +02:00
2ae4241174
Frame reference resources in query functions
2016-10-21 21:44:24 +02:00
2893d67b93
Handle OPCODE_CUSTOMDATA when counting declarations
2016-10-21 21:44:24 +02:00
d4d9a855c4
Fix replaying through indirect draws on vulkan
2016-10-21 21:44:24 +02:00
106308a3fa
Note if a command signature is graphics or compute
2016-10-21 21:44:24 +02:00
ac53611f07
If any list executed contains an ExecuteIndirect, ref all buffers
...
* Since buffers can be referenced indirectly on the GPU by their GPU
address, there's no feasible way to know if the buffer is actually
used or not. If an ExecuteIndirect is seen at all we just have to
pessimistically include all buffers.
* Generally textures take up the bulk of VRAM usage, so this likely
won't be too bad.
2016-10-21 21:44:24 +02:00
febe2b1ad9
Patch ExecuteIndirect call data to point to replayed GPU addresses
2016-10-21 21:44:24 +02:00
1c3b4dfc94
Add support for SM5.1 bytecode changes - primarily resource arrays
2016-10-21 21:44:24 +02:00
7434077718
Extract register space from DXBC reflection info
2016-10-21 21:44:24 +02:00
c2553e0d65
Don't fill in register spaces if there's no root signature set
2016-10-21 21:44:24 +02:00
c997cad967
Set plane slice as necessary on stencil SRVs
2016-10-21 21:44:24 +02:00
f756cf363e
Remove legacy DXBC padding of cbuffers array. Also stub reg spaces
...
* Historically there was no bindpoint mapping so it was convenient to
just pad the cbuffers array so that the elements in it were indexed by
their bind point. It doesn't make any sense anymore especially with
D3D12's bind model, so just remove it and roll bind points into a
struct member.
2016-10-21 21:44:24 +02:00
cc35215a12
Always call SetVariables from UI thread, don't use InvokeRequired
2016-10-21 21:44:24 +02:00
8856d3a6f2
Force-stop Android package before starting it for capture.
2016-10-19 20:32:00 +02:00
652c4dfe22
Update expected size of FetchFrameRecord to account for increased stats
2016-10-19 19:41:58 +02:00
9fb7cb81db
Don't pass potentially remote path to Path.GetFileName. Refs #400
2016-10-19 19:14:58 +02:00
459619ff0c
Don't try to access file on local disk in statistics. Refs #400
2016-10-19 19:12:41 +02:00
d1e5149527
Fix missing <math.h> header in qrenderdoc for Arch/clang++
...
Add the <math.h> header to some of the source files for the target
qrenderdoc. The build would not have succeded without these headers
declaration under Arch Linux/clang++.
2016-10-19 18:55:50 +02:00
ae06f86a1c
32-bit compile fixes (casts from uint64_t to size_t)
2016-10-18 21:02:58 +02:00
b8520a274d
Fix compile errors on CI targets
2016-10-18 19:35:01 +02:00
2fe22e96b5
Add support for tracking and saving mid-frame and persistent maps
2016-10-18 19:10:06 +02:00
681eb21b0a
Fetch debug messages before clearing the list
2016-10-18 19:08:47 +02:00
e959d4b826
Implement query functions (without applying predication on replay)
2016-10-18 16:27:30 +02:00
812e4affd3
Tidy up remaining functions marking not-implemented stuff
2016-10-18 15:54:23 +02:00
4dbfe17747
Use event usage to guess at buffer creation flags
2016-10-18 15:31:44 +02:00
cf5bf6f166
Track buffer usage as Indirect parameter buffer
2016-10-18 15:31:43 +02:00
17493e9158
Track resources created with a cubemap SRV, to mark them as cubemaps
2016-10-18 15:31:42 +02:00
b220a61d79
Not sure complex resource releasing logic is needed
...
* I think we're mostly covered by the requirement that resources stay
alive until GPU execution is finished.
2016-10-18 15:31:42 +02:00
4253094557
Track resource event usage
2016-10-18 15:31:41 +02:00
2d19d953c5
Handle debug messages on list execute
...
* It's not clear if the debug layer can usefully provide messages at a
specific event or even a specific submission.
2016-10-18 13:06:57 +02:00
5329bd4695
Serialise Wait() and SOSetTargets()
2016-10-18 12:30:55 +02:00
99b45aefb5
Minimal possible residency handling code
...
* We unwrap objects and track residency state, and when preparing init
states if necessary then we make those resources resident and sync the
list execute so we can evict them again immediately afterwards.
* On replay, all resources are created and left as resident, which could
go over the budget available. If a frame uses close to or more than
its total budget over the course of a frame this could fail. One fix
to that would be to track referenced resources in command lists at
replay time, and only make resident what we need for each list and
execute them in isolation, then evict right after (so all frame
resources are 'default evicted').
2016-10-18 12:12:54 +02:00
ca3e394c19
Add proper locking around check of CAPFRAME state
2016-10-18 12:04:07 +02:00
e4cc738a13
Ignore D3D12 debug message about NULL map ranges
2016-10-18 12:03:16 +02:00
529f230e4e
Don't directly frame-reference lists being executed
...
* It's no longer needed as they don't have direct parents anymore, and
it can break in cases where a list is being re-recorded while a
frame capture is on-going (but hasn't been baked).
2016-10-18 12:02:21 +02:00
63f505c95e
Delete frame capture record on shutdown
2016-10-18 10:38:53 +02:00
3cf71f158a
Perform copies on wrapped descriptors so internal tracking is updated
2016-10-18 10:38:53 +02:00
baldurk