* Clarified aspects of Zce, MISA.C, and the D extension
* Clarified that invoking *ret instructions in a lesser privilege mode results in an illegal-instruction exception
* Reworded sentence about load/store-multiple instructions
* Removed excess D reference, since D implies F
* Reworded to use __x__ forms
* hypervisor: translate hstatus to wavedrom
The bytefield is hardly maintainable.
Signed-off-by: Radim Krčmář <rkrcmar@ventanamicro.com>
* hypervisor: add HUPMM to hstatus
HUPMM was forgotten when the pointer masking extension was added.
Signed-off-by: Radim Krčmář <rkrcmar@ventanamicro.com>
---------
Signed-off-by: Radim Krčmář <rkrcmar@ventanamicro.com>
The text refers to spimm as "the number of additional 16-byte address increments allocated for the stack frame."
and "The total stack adjustment represents the total size of the stack frame, which is stack_adj_base added
to spimm scaled by 16"
The "Example RV32I PUSH/POP sequences" show spimm with values of 0-3.
Based on those sections, I think spimm is a 2-bit value rather than a 6-bit value.
This PR fixes an editing error in which "C" should have been changed to
"Zca" when the various Zc extensions were incorporated. It also clarifies
that any compressed instruction, not just those in Zca, that map to one of the
permitted "I" instructions can be used in a constrained LR/SC loop. This
was always the case, but it was stated obliquely.
Resolves#1938
This is pursuant to making the manual contain only ratified content.
Note, this action does not connote any deprecation of the forthcoming Zp
extensions for packed SIMD. They, like all in-progress extensions, will
be incorporated into the manual upon ratification.
Setting UNRELIABLE_BUT_FASTER_INCREMENTAL_BUILDS=1 on the make command line
will avoid wiping the entire build directory before builds. Since each
output file uses a different build directory there should be no
need to wipe the whole build directory. The Makefile already triggers a
rebuild if any file under the source directory is modified so this should
be safe.
This significantly speeds up the build for me since we no longer regenerate
all the diagrams for each build. For the unprivileged manual the build time
goes from about 4:30 to 3:40.
However, it has been shown that non-clean builds don't always produce the
same output PDF, so this is gated by an opt-in flag.
Fixes: 1929d45a059979c55b070fba414d97d530a44c99
Author: mmoizhussain
On branch mmhus/fix-smcntrpmf-heading
Changes to be committed:
modified: src/smcntrpmf.adoc
This could be fixed by adding `2>&1` to the `docker info` command, but
it seems cleaner to correctly infer the value of SKIP_DOCKER and avoid
entering the whole `ifneq ($(SKIP_DOCKER),true)` block.
. Conversions narrowing to BF16 end up at BF16, not the target
precision (which is the origin of the conversion).
. Capitalize an 'i' at sentence start.
. Remove repetition of C bit in *stateen0 not being custom state.
. Remove a misplaced 'of'.
