This is a major reworking of the way the debugger is implemented.
Instead of generating shader code that is continually driving `vk::dbg` state, the debugger now:
* Maintains a full shadow copy of all Intermediate values (SSA)
* Only calls out to C++ whenever a trap is set
* Only constructs and updates the `vk::dbg` state when a trap is hit
The main goal of this reworking is to properly support OpenCL.Debug.100's DebugValue instructions that may use deref expressions
(See https://www.khronos.org/registry/spir-v/specs/unified1/OpenCL.DebugInfo.100.html#_debug_operations_a_id_operation_a)
These are remarkably hard to implement without having everything backed with real memory.
The resulting reimplementation is:
* Much, much faster (>1000x faster for sample apps I've tried). Only ~2x slower than non-debug enabled SwiftShader now.
* Uses much less runtime memory.
* Much cleaner, removing a load of template magic for setting variable state that was extremely hard to follow
* More thoroughly commented, explaining each component and method.
Bug: b/145351270
Change-Id: I1b48e9defae095109ebfaf90dfb7c7e22d335816
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/48729Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
