On 30.06.2026 00:49, Randy Schifflin wrote:
SH-4 manual for reference:
https://0x04.net/~mwk/doc/sh/e602156_sh4.pdf
There are two issues with the current TCG code generation
for the `fipr` and `ftrv` instructions. Firstly, they incorrectly
check that the PR flag of the FPSCR register is high. The manual
states these instructions are only defined when that flag is 0.
Secondly, they take as operands a vector index 0-3, but treat
them as offsets into the floating point register list. Vectors
span four floating point registers, and thus each index needs to be
multiplied by 4 to translate to the correct register offset.
This PR adds simple TCG tests for the `fipr` and `ftrv` instructions,
and fixes the issues above by ensuring the PR flag of FPSCR is 0 and
properly translating the vector index operand to its register offset.
Signed-off-by: Randy Schifflin <[email protected]>
---
Randy Schifflin (2):
target/sh4: add tests for sh4 fipr and ftrv instructions
target/sh4: fixup tcg for sh4 fipr/ftrv instructions
target/sh4/op_helper.c | 2 +-
target/sh4/translate.c | 15 +++----
tests/tcg/sh4/Makefile.target | 7 +++
tests/tcg/sh4/test-fpu-fipr.c | 64 ++++++++++++++++++++++++++++
tests/tcg/sh4/test-fpu-ftrv.c | 99 +++++++++++++++++++++++++++++++++++++++++++
5 files changed, 176 insertions(+), 11 deletions(-)
This looks like qemu-stable material.
I'm picking this up for currently active qemu-stable series.
Please let me know if I should not.
It'd be nice to have the testcase applied to master too (I'll
pick it up for stable series as well, if I wont forget :)
Thanks,
/mjt
