On 2026-08-15 12:45, Richard Henderson wrote: > This merges vext_continuous_ldst_host into the ldst_host function. > We can then handle the byte little-endian optimization at compile-time. > > Signed-off-by: Richard Henderson <[email protected]> > --- > target/riscv/tcg/vector_helper.c | 83 +++++++++++++++----------------- > 1 file changed, 40 insertions(+), 43 deletions(-) > > diff --git a/target/riscv/tcg/vector_helper.c > b/target/riscv/tcg/vector_helper.c > index 661f1ab2f6..3786a17231 100644 > --- a/target/riscv/tcg/vector_helper.c > +++ b/target/riscv/tcg/vector_helper.c > @@ -214,7 +214,8 @@ static inline MemOpIdx vext_make_memop_idx(CPURISCVState > *env, size_t size) > /* elements operations for load and store */ > typedef void vext_ldst_elem_fn_tlb(CPURISCVState *env, abi_ptr addr, > uint32_t idx, void *vd, uintptr_t > retaddr); > -typedef void vext_ldst_elem_fn_host(void *vd, uint32_t idx, void *host); > +typedef void vext_ldst_elem_fn_host(void *vd, void *host, > + uint32_t idx, uint32_t evl); > > #define GEN_VEXT_TLB_LD_ELEM(NAME, ETYPE, H, LDSUF) \ > static inline QEMU_ALWAYS_INLINE \ > @@ -226,12 +227,15 @@ void NAME##_tlb(CPURISCVState *env, abi_ptr addr, > \ > *cur = cpu_##LDSUF##_mmu(env, addr, oi, retaddr); \ > } \ > > -#define GEN_VEXT_HOST_LD_ELEM(NAME, ETYPE, H, LDSUF) \ > -static inline QEMU_ALWAYS_INLINE \ > -void NAME##_host(void *vd, uint32_t idx, void *host) \ > -{ \ > - ETYPE *cur = ((ETYPE *)vd + H(idx)); \ > - *cur = (ETYPE)LDSUF##_p(host); \ > +#define GEN_VEXT_HOST_LD_ELEM(NAME, ETYPE, H, LDSUF) \ > +static inline QEMU_ALWAYS_INLINE \ > +void NAME##_host(void *vd, void *host, uint32_t idx, uint32_t evl) \ > +{ \ > + do { \ > + ETYPE *cur = (ETYPE *)vd + H(idx); \ > + *cur = LDSUF##_p(host); \ > + host += sizeof(ETYPE); \ > + } while (++idx < evl); \ > } > > GEN_VEXT_TLB_LD_ELEM(lde_b, uint8_t, H1, ldb) > @@ -239,7 +243,16 @@ GEN_VEXT_TLB_LD_ELEM(lde_h, uint16_t, H2, ldw) > GEN_VEXT_TLB_LD_ELEM(lde_w, uint32_t, H4, ldl) > GEN_VEXT_TLB_LD_ELEM(lde_d, uint64_t, H8, ldq) > > +#if HOST_BIG_ENDIAN > GEN_VEXT_HOST_LD_ELEM(lde_b, uint8_t, H1, ldub) > +#else > +static inline QEMU_ALWAYS_INLINE > +void lde_b_host(void *vd, void *host, uint32_t idx, uint32_t evl) > +{ > + memcpy(vd + idx, host, evl - idx); > +} > +#endif > + > GEN_VEXT_HOST_LD_ELEM(lde_h, uint16_t, H2, lduw_le) > GEN_VEXT_HOST_LD_ELEM(lde_w, uint32_t, H4, ldl_le) > GEN_VEXT_HOST_LD_ELEM(lde_d, uint64_t, H8, ldq_le) > @@ -254,12 +267,15 @@ void NAME##_tlb(CPURISCVState *env, abi_ptr addr, > \ > cpu_##STSUF##_mmu(env, addr, data, oi, retaddr); \ > } \ > > -#define GEN_VEXT_HOST_ST_ELEM(NAME, ETYPE, H, STSUF) \ > -static inline QEMU_ALWAYS_INLINE \ > -void NAME##_host(void *vd, uint32_t idx, void *host) \ > -{ \ > - ETYPE data = *((ETYPE *)vd + H(idx)); \ > - STSUF##_p(host, data); \ > +#define GEN_VEXT_HOST_ST_ELEM(NAME, ETYPE, H, STSUF) \ > +static inline QEMU_ALWAYS_INLINE \ > +void NAME##_host(void *vd, void *host, uint32_t idx, uint32_t evl) \ > +{ \ > + do { \ > + ETYPE data = *((ETYPE *)vd + H(idx)); \ > + STSUF##_p(host, data); \ > + host += sizeof(ETYPE); \ > + } while (++idx < evl); \ > } > > GEN_VEXT_TLB_ST_ELEM(ste_b, uint8_t, H1, stb) > @@ -267,7 +283,16 @@ GEN_VEXT_TLB_ST_ELEM(ste_h, uint16_t, H2, stw) > GEN_VEXT_TLB_ST_ELEM(ste_w, uint32_t, H4, stl) > GEN_VEXT_TLB_ST_ELEM(ste_d, uint64_t, H8, stq) > > +#if HOST_BIG_ENDIAN > GEN_VEXT_HOST_ST_ELEM(ste_b, uint8_t, H1, stb) > +#else > +static inline QEMU_ALWAYS_INLINE > +void ste_b_host(void *vd, void *host, uint32_t idx, uint32_t evl) > +{ > + memcpy(host, vd + idx, evl - idx); > +} > +#endif > + > GEN_VEXT_HOST_ST_ELEM(ste_h, uint16_t, H2, stw_le) > GEN_VEXT_HOST_ST_ELEM(ste_w, uint32_t, H4, stl_le) > GEN_VEXT_HOST_ST_ELEM(ste_d, uint64_t, H8, stq_le) > @@ -284,33 +309,6 @@ vext_continuous_ldst_tlb(CPURISCVState *env, > vext_ldst_elem_fn_tlb *ldst_tlb, > } > } > > -static inline QEMU_ALWAYS_INLINE void > -vext_continuous_ldst_host(CPURISCVState *env, vext_ldst_elem_fn_host > *ldst_host, > - void *vd, uint32_t evl, uint32_t reg_start, void > *host, > - uint32_t esz, bool is_load) > -{ > - if (HOST_BIG_ENDIAN) { > - for (; reg_start < evl; reg_start++, host += esz) { > - ldst_host(vd, reg_start, host); > - } > - } else { > - if (esz == 1) { > - uint32_t byte_offset = reg_start * esz; > - uint32_t size = (evl - reg_start) * esz; > - > - if (is_load) { > - memcpy(vd + byte_offset, host, size); > - } else { > - memcpy(host, vd + byte_offset, size); > - } > - } else { > - for (; reg_start < evl; reg_start++, host += esz) { > - ldst_host(vd, reg_start, host); > - } > - } > - } > -} > - > static void vext_set_tail_elems_1s(uint32_t vl, void *vd, uint32_t nf, > uint32_t esz, uint32_t max_elems) > { > @@ -334,7 +332,7 @@ static void vext_ldst_nf_host(void *vd, void *host, > uint32_t i, uint32_t nf, > vext_ldst_elem_fn_host *ldst_host) > { > for (uint32_t k = 0; k < nf; k++, host += esz) { > - ldst_host(vd, i + k * max_elems, host); > + ldst_host(vd + k * max_elems, host, i, i + 1);
The type of ldst_host vd parameter is void *, so vd + k * max_elems advances in bytes, not the target element width. I think here should be > + ldst_host(vd, host, i + k * max_elems, i + k * max_elems + 1); rnax
