On 10/7/2026 5:37 PM, Brian Cain wrote: > MMVector is for HVX register state. Now with HVX floating point/QFloat, > MMVector includes internal extended-precision bits in addition to the > register content. Use VECTOR_SIZE_BYTE for GVec op lengths, vec mem alignment > and xfer sizes, scatter/gather processing, and histogram bounds. > > The standalone scatter/gather helper does not include the semantics macros > that define fVECSIZE(). Use sizeof_field(MMVector, ub) there to express the > same payload length without depending on that macro. > > Signed-off-by: Brian Cain <[email protected]> > --- > target/hexagon/gen_tcg_hvx.h | 172 ++++++++++++------------ > target/hexagon/mmvec/macros.h | 8 +- > target/hexagon/genptr.c | 18 +-- > target/hexagon/mmvec/system_ext_mmvec.c | 14 +- > target/hexagon/op_helper.c | 34 ++--- > 5 files changed, 123 insertions(+), 123 deletions(-) >
Reviewed-by: Pierrick Bouvier <[email protected]>
