Drop the usage in vext_page_ldst_us, merging the nf == 1 case into the vext_ldst_nf_tlb loop. Inline it into the last use in vext_ldst_us.
Signed-off-by: Richard Henderson <[email protected]> --- target/riscv/tcg/vector_helper.c | 34 ++++++++++---------------------- 1 file changed, 10 insertions(+), 24 deletions(-) diff --git a/target/riscv/tcg/vector_helper.c b/target/riscv/tcg/vector_helper.c index da1e704c56..63a51f2b48 100644 --- a/target/riscv/tcg/vector_helper.c +++ b/target/riscv/tcg/vector_helper.c @@ -272,17 +272,6 @@ 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) -static inline QEMU_ALWAYS_INLINE void -vext_continuous_ldst_tlb(CPURISCVState *env, vext_ldst_elem_fn_tlb *ldst_tlb, - void *vd, uint32_t evl, target_ulong addr, - uint32_t reg_start, uintptr_t ra, uint32_t esz) -{ - uint32_t i; - for (i = env->vstart; i < evl; env->vstart = ++i, addr += esz) { - ldst_tlb(env, adjust_addr(env, addr), i, vd, ra); - } -} - 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, @@ -465,15 +454,10 @@ vext_page_ldst_us(CPURISCVState *env, void *vd, target_ulong addr, } env->vstart += elems; } else { - if (nf == 1) { - vext_continuous_ldst_tlb(env, ldst_tlb, vd, evl, addr, - env->vstart, ra, esz); - } else { - for (uint32_t i = env->vstart; i < evl; env->vstart = ++i) { - vext_ldst_nf_tlb(env, vd, addr, i, nf, esz, - max_elems, ldst_tlb, ra); - addr += msize; - } + for (uint32_t i = env->vstart; i < evl; env->vstart = ++i) { + vext_ldst_nf_tlb(env, vd, addr, i, nf, esz, + max_elems, ldst_tlb, ra); + addr += msize; } } } @@ -495,13 +479,15 @@ vext_ldst_us(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc, #if defined(CONFIG_USER_ONLY) /* - * For data sizes <= 6 bytes we get better performance by simply calling - * vext_continuous_ldst_tlb + * For data sizes <= 6 bytes we get better performance + * by simply calling ldst_tlb. */ if (nf == 1 && (evl << log2_esz) <= 6) { addr = base + (env->vstart << log2_esz); - vext_continuous_ldst_tlb(env, ldst_tlb, vd, evl, addr, - env->vstart, ra, esz); + for (uint32_t i = env->vstart; i < evl; + env->vstart = ++i, addr += esz) { + ldst_tlb(env, adjust_addr(env, addr), i, vd, ra); + } env->vstart = 0; vext_set_tail_elems_1s(evl, vd, desc, nf, esz, max_elems); return; -- 2.43.0
