llvm-project

Commit Graph

Author	SHA1	Message	Date
Matt Arsenault	d88db6d7fc	AMDGPU: Always use s33 for global scratch wave offset Every called function could possibly need this to calculate the absolute address of stack objectst, and this avoids inserting a copy around every call site in the kernel. It's also somewhat cleaner to keep this in a callee saved SGPR. llvm-svn: 363990	2019-06-20 21:58:24 +00:00
Matt Arsenault	aa41e92e17	AMDGPU: Avoid most waitcnts before calls Currently you get extra waits, because waits are inserted for the register dependencies of the call, and the function prolog waits on everything. Currently waits are still inserted on returns. It may make sense to not do this, and wait in the caller instead. llvm-svn: 363465	2019-06-14 21:52:26 +00:00
Matt Arsenault	1509fde891	AMDGPU: Add baseline test for call waitcnt insertion llvm-svn: 363453	2019-06-14 21:01:23 +00:00

Author

SHA1

Message

Date

Matt Arsenault

d88db6d7fc

AMDGPU: Always use s33 for global scratch wave offset

Every called function could possibly need this to calculate the
absolute address of stack objectst, and this avoids inserting a copy
around every call site in the kernel. It's also somewhat cleaner to
keep this in a callee saved SGPR.

llvm-svn: 363990

2019-06-20 21:58:24 +00:00

Matt Arsenault

aa41e92e17

AMDGPU: Avoid most waitcnts before calls

Currently you get extra waits, because waits are inserted for the
register dependencies of the call, and the function prolog waits on
everything.

Currently waits are still inserted on returns. It may make sense to
not do this, and wait in the caller instead.

llvm-svn: 363465

2019-06-14 21:52:26 +00:00

Matt Arsenault

1509fde891

AMDGPU: Add baseline test for call waitcnt insertion

llvm-svn: 363453

2019-06-14 21:01:23 +00:00

3 Commits