software, integration/export: (re-)expose CSR subregister accessors #366
Conversation
|
FYI - @xobs |
litex/soc/integration/export.py
Outdated
| r += "extern uint32_t csr_rd_uint32(unsigned long a);\n" | ||
| r += "extern uint64_t csr_rd_uint64(unsigned long a);\n" | ||
| r += "extern void csr_write_sub(unsigned long v, unsigned long a);\n" | ||
| r += "extern unsigned long csr_read_sub(unsigned long a);\n" |
There was a problem hiding this comment.
These being marked as extern will mean the compiler can't optimize them without lto turned on (see #253 for that).
There was a problem hiding this comment.
this is the case where they're externally provided by downstream, and downstream defines CSR_ACCESSORS_DEFINED to let us know we shouldn't be using the built-in ones.
That's the way it was before I ever got involved :) The litex-provided accessors (in common.h) are definitely NOT marked as external
| #define CSR_DW_BYTES (CONFIG_CSR_DATA_WIDTH/8) | ||
|
|
||
| /* CSR subregisters are embedded inside native CPU word aligned locations: */ | ||
| #define MMPTR(a) (*((volatile unsigned long *)(a))) | ||
|
|
||
| static inline void csr_write_sub(unsigned long v, unsigned long a) |
There was a problem hiding this comment.
Maybe csr_write_part or csr_write_section are better names than csr_write_sub?
There was a problem hiding this comment.
'sub' is for "subregister" (how about 'subreg' then)? I could also go with "slice". Both "part" and "section" are IMHO terms that are too overloaded to ensure we don't confuse future readers of the source code...
There was a problem hiding this comment.
Or should we call it csr_write_simple to mach the simple_csrs used in https://github.com/enjoy-digital/litex/blob/master/litex/soc/interconnect/csr.py? (if simple is confusing we could change it to something better in both software/gateware later)
There was a problem hiding this comment.
Or should we call it
csr_write_simpleor to mach thesimple_csrs...
That works for me, thanks! Force-pushed an amended commit reflecting that.
Expose a pair of `csr_[read|write]_simple()` subregister accessors, and restore the way dedicated accessors are generated in "generated/csr.h" to use hard-coded combinations of shifts and subregister accessor calls. This restores downstream ability to override CSR handling at the subregister accessor level. Signed-off-by: Gabriel Somlo <gsomlo@gmail.com>
|
This works for me. It should get inlined easily, and should still work on both 32 and 64 bit architectures. And the new name should prevent confusion, and is guaranteed to work on any new architectures we come across. Maybe we'll find someday that we need to have it accept a Looks good! |
Expose a pair of
csr_[read|write]_sub()subregister accessors, andrestore the way dedicated accessors are generated in "generated/csr.h"
to use hard-coded combinations of shifts and subregister accessor call.
This restores downstream ability to override CSR handling at the
subregister accessor level.
Signed-off-by: Gabriel Somlo gsomlo@gmail.com
This is a followup to PR #339, and should hopefully address issue #365.