diff options
author | Chih-Min Chao <chihmin.chao@sifive.com> | 2019-09-25 19:57:50 -0700 |
---|---|---|
committer | Chih-Min Chao <chihmin.chao@sifive.com> | 2019-11-11 19:02:34 -0800 |
commit | b9d9e1ebd02c62ad354195481d1136f5be3f54cd (patch) | |
tree | c0d333eae33185b189cf730719f7d007c98bddaa | |
parent | df85f7fbe4b82eccf594a0d0fb7b8f5e6150dea1 (diff) | |
download | spike-b9d9e1ebd02c62ad354195481d1136f5be3f54cd.zip spike-b9d9e1ebd02c62ad354195481d1136f5be3f54cd.tar.gz spike-b9d9e1ebd02c62ad354195481d1136f5be3f54cd.tar.bz2 |
rvv: remove configuable tail-zero
tail zero feature has been removed after v0.8-draft
Signed-off-by: Chih-Min Chao <chihmin.chao@sifive.com>
-rw-r--r-- | riscv/decode.h | 107 | ||||
-rw-r--r-- | riscv/insns/vcompress_vm.h | 2 | ||||
-rw-r--r-- | riscv/insns/vfmerge_vfm.h | 1 | ||||
-rw-r--r-- | riscv/insns/vfmv_s_f.h | 7 | ||||
-rw-r--r-- | riscv/insns/vfmv_v_f.h | 1 | ||||
-rw-r--r-- | riscv/insns/vid_v.h | 1 | ||||
-rw-r--r-- | riscv/insns/viota_m.h | 1 | ||||
-rw-r--r-- | riscv/insns/vleff_v.h | 17 | ||||
-rw-r--r-- | riscv/insns/vmsbf_m.h | 1 | ||||
-rw-r--r-- | riscv/insns/vmsif_m.h | 1 | ||||
-rw-r--r-- | riscv/insns/vmsof_m.h | 1 | ||||
-rw-r--r-- | riscv/insns/vmv_s_x.h | 20 | ||||
-rw-r--r-- | riscv/insns/vsuxb_v.h | 19 | ||||
-rw-r--r-- | riscv/insns/vsuxe_v.h | 21 | ||||
-rw-r--r-- | riscv/insns/vsuxh_v.h | 16 | ||||
-rw-r--r-- | riscv/insns/vsuxw_v.h | 11 |
16 files changed, 41 insertions, 186 deletions
diff --git a/riscv/decode.h b/riscv/decode.h index 0feb7ab..0f12262 100644 --- a/riscv/decode.h +++ b/riscv/decode.h @@ -65,29 +65,6 @@ const int NCSR = 4096; #define MAX_INSN_LENGTH 8 #define PC_ALIGN 2 -#ifndef TAIL_ZEROING - #define TAIL_ZEROING true -#else - #define TAIL_ZEROING false -#endif - -#ifdef WORDS_BIGENDIAN - // Elements are stored in opposite order, see comment in processor.h - #define TAIL_ZERO(x) \ - uint8_t *tail = &P.VU.elt<uint8_t>(rd_num, vl * (x) - 1); \ - memset(tail - (P.VU.vlmax - vl) * (x), 0, (P.VU.vlmax - vl) * (x)); - #define TAIL_ZERO_REDUCTION(x) \ - uint8_t *tail = (uint8_t *)&P.VU.elt<type_sew_t<x>::type>(rd_num, 0); \ - memset(tail - ((P.VU.get_vlen() - x) >> 3), 0, (P.VU.get_vlen() - x) >> 3); -#else - #define TAIL_ZERO(x) \ - uint8_t *tail = &P.VU.elt<uint8_t>(rd_num, vl * (x)); \ - memset(tail, 0, (P.VU.vlmax - vl) * (x)); - #define TAIL_ZERO_REDUCTION(x) \ - uint8_t *tail = (uint8_t *)&P.VU.elt<type_sew_t<x>::type>(rd_num, 1); \ - memset(tail, 0, (P.VU.get_vlen() - x) >> 3); -#endif - typedef uint64_t insn_bits_t; class insn_t { @@ -368,9 +345,7 @@ inline long double to_f(float128_t f){long double r; memcpy(&r, &f, sizeof(r)); } #define VI_ELEMENT_SKIP(inx) \ - if (inx >= vl && TAIL_ZEROING) { \ - is_valid = false; \ - } else if (inx >= vl && !TAIL_ZEROING) { \ + if (inx >= vl) { \ continue; \ } else if (inx < P.VU.vstart) { \ continue; \ @@ -436,52 +411,22 @@ static inline bool is_overlaped(const int astart, const int asize, reg_t rs2_num = insn.rs2(); \ for (reg_t i=P.VU.vstart; i<vl; ++i){ -#define VI_TAIL_ZERO(elm) \ - if (vl != 0 && vl < P.VU.vlmax && TAIL_ZEROING) { \ - TAIL_ZERO((sew >> 3) * elm); \ - } - -#define VI_TAIL_ZERO_MASK(dst) \ - if (vl != 0 && TAIL_ZEROING){ \ - for (reg_t i=vl; i<P.VU.vlmax; ++i){ \ - const int mlen = P.VU.vmlen; \ - const int midx = (mlen * i) / 64; \ - const int mpos = (mlen * i) % 64; \ - uint64_t mmask = (UINT64_MAX << (64 - mlen)) >> (64 - mlen - mpos); \ - uint64_t &vdi = P.VU.elt<uint64_t>(dst, midx); \ - vdi = (vdi & ~mmask);\ - }\ - }\ - #define VI_LOOP_BASE \ VI_GENERAL_LOOP_BASE \ VI_LOOP_ELEMENT_SKIP(); #define VI_LOOP_END \ } \ - if (vl != 0 && vl < P.VU.vlmax && TAIL_ZEROING){ \ - TAIL_ZERO((sew >> 3) * 1); \ - }\ - P.VU.vstart = 0; - -#define VI_LOOP_END_NO_TAIL_ZERO \ - } \ P.VU.vstart = 0; #define VI_LOOP_WIDEN_END \ } \ - if (vl != 0 && vl < P.VU.vlmax && TAIL_ZEROING){ \ - TAIL_ZERO((sew >> 3) * 2); \ - }\ P.VU.vstart = 0; #define VI_LOOP_REDUCTION_END(x) \ } \ if (vl > 0) { \ vd_0_des = vd_0_res; \ - if (TAIL_ZEROING) { \ - TAIL_ZERO_REDUCTION(x); \ - } \ } \ P.VU.vstart = 0; @@ -502,7 +447,6 @@ static inline bool is_overlaped(const int astart, const int asize, #define VI_LOOP_CMP_END \ vdi = (vdi & ~mmask) | (((res) << mpos) & mmask); \ } \ - VI_TAIL_ZERO_MASK(rd_num); \ P.VU.vstart = 0; #define VI_LOOP_MASK(op) \ @@ -518,17 +462,6 @@ static inline bool is_overlaped(const int astart, const int asize, uint64_t &res = P.VU.elt<uint64_t>(insn.rd(), midx); \ res = (res & ~mmask) | ((op) & (1ULL << mpos)); \ } \ - \ - if (TAIL_ZEROING) {\ - for (reg_t i = vl; i < P.VU.vlmax && i > 0; ++i) { \ - int mlen = P.VU.vmlen; \ - int midx = (mlen * i) / 64; \ - int mpos = (mlen * i) % 64; \ - uint64_t mmask = (UINT64_MAX << (64 - mlen)) >> (64 - mlen - mpos); \ - uint64_t &res = P.VU.elt<uint64_t>(insn.rd(), midx); \ - res = (res & ~mmask); \ - } \ - } \ P.VU.vstart = 0; #define VI_LOOP_NSHIFT_BASE \ @@ -1307,7 +1240,6 @@ VI_LOOP_END BODY; \ } \ } \ - VI_TAIL_ZERO_MASK(rd_num); #define VI_XI_LOOP_CARRY(BODY) \ VI_LOOP_BASE \ @@ -1325,7 +1257,6 @@ VI_LOOP_END BODY; \ } \ } \ - VI_TAIL_ZERO_MASK(rd_num); // average loop #define VI_VVX_LOOP_AVG(opd, op) \ @@ -1409,11 +1340,8 @@ for (reg_t i = 0; i < vlmax; ++i) { \ const reg_t vlmax = P.VU.vlmax; \ const reg_t vlmul = P.VU.vlmul; \ for (reg_t i = 0; i < vlmax && vl != 0; ++i) { \ - bool is_valid = true; \ VI_STRIP(i) \ VI_ELEMENT_SKIP(i); \ - if (!is_valid) \ - continue; \ for (reg_t fn = 0; fn < nf; ++fn) { \ st_width##_t val = 0; \ switch (P.VU.vsew) { \ @@ -1445,11 +1373,10 @@ for (reg_t i = 0; i < vlmax; ++i) { \ const reg_t vlmax = P.VU.vlmax; \ const reg_t vlmul = P.VU.vlmul; \ for (reg_t i = 0; i < vlmax && vl != 0; ++i) { \ - bool is_valid = true; \ VI_ELEMENT_SKIP(i); \ VI_STRIP(i); \ for (reg_t fn = 0; fn < nf; ++fn) { \ - ld_width##_t val = is_valid ? MMU.load_##ld_width(baseAddr + (stride) + (offset) * elt_byte) : 0; \ + ld_width##_t val = MMU.load_##ld_width(baseAddr + (stride) + (offset) * elt_byte); \ if (vd + fn >= NVPR){ \ P.VU.vstart = vreg_inx;\ require(false); \ @@ -1485,7 +1412,6 @@ for (reg_t i = 0; i < vlmax; ++i) { \ const reg_t vlmax = P.VU.vlmax; \ const reg_t vlmul = P.VU.vlmul; \ for (reg_t i = 0; i < vlmax && vl != 0; ++i) { \ - bool is_valid = true; \ VI_STRIP(i); \ VI_ELEMENT_SKIP(i); \ \ @@ -1494,20 +1420,20 @@ for (reg_t i = 0; i < vlmax; ++i) { \ \ switch (sew) { \ case e8: \ - p->VU.elt<uint8_t>(rd_num + fn * vlmul, vreg_inx) = is_valid ? val : 0; \ + p->VU.elt<uint8_t>(rd_num + fn * vlmul, vreg_inx) = val; \ break; \ case e16: \ - p->VU.elt<uint16_t>(rd_num + fn * vlmul, vreg_inx) = is_valid ? val : 0; \ + p->VU.elt<uint16_t>(rd_num + fn * vlmul, vreg_inx) = val; \ break; \ case e32: \ - p->VU.elt<uint32_t>(rd_num + fn * vlmul, vreg_inx) = is_valid ? val : 0; \ + p->VU.elt<uint32_t>(rd_num + fn * vlmul, vreg_inx) = val; \ break; \ case e64: \ - p->VU.elt<uint64_t>(rd_num + fn * vlmul, vreg_inx) = is_valid ? val : 0; \ + p->VU.elt<uint64_t>(rd_num + fn * vlmul, vreg_inx) = val; \ break; \ } \ \ - if (val == 0 && is_valid) { \ + if (val == 0) { \ p->VU.vl = i; \ early_stop = true; \ break; \ @@ -1568,16 +1494,10 @@ for (reg_t i = 0; i < vlmax; ++i) { \ #define VI_VFP_LOOP_END \ } \ - if (vl != 0 && vl < P.VU.vlmax && TAIL_ZEROING){ \ - TAIL_ZERO((P.VU.vsew >> 3) * 1); \ - }\ P.VU.vstart = 0; \ #define VI_VFP_LOOP_WIDE_END \ } \ - if (vl != 0 && vl < P.VU.vlmax && TAIL_ZEROING){ \ - TAIL_ZERO((P.VU.vsew >> 3) * 2); \ - }\ P.VU.vstart = 0; \ set_fp_exceptions; @@ -1587,9 +1507,6 @@ for (reg_t i = 0; i < vlmax; ++i) { \ set_fp_exceptions; \ if (vl > 0) { \ P.VU.elt<type_sew_t<x>::type>(rd_num, 0) = vd_0.v; \ - if (TAIL_ZEROING) { \ - TAIL_ZERO_REDUCTION(x); \ - } \ } #define VI_VFP_LOOP_CMP_END \ @@ -1605,16 +1522,6 @@ for (reg_t i = 0; i < vlmax; ++i) { \ break; \ }; \ } \ - if (vl != 0 && TAIL_ZEROING){ \ - for (reg_t i=vl; i<P.VU.vlmax; ++i){ \ - const int mlen = P.VU.vmlen; \ - const int midx = (mlen * i) / 64; \ - const int mpos = (mlen * i) % 64; \ - uint64_t mmask = (UINT64_MAX << (64 - mlen)) >> (64 - mlen - mpos); \ - uint64_t &vdi = P.VU.elt<uint64_t>(insn.rd(), midx); \ - vdi = (vdi & ~mmask);\ - }\ - }\ P.VU.vstart = 0; \ set_fp_exceptions; diff --git a/riscv/insns/vcompress_vm.h b/riscv/insns/vcompress_vm.h index b056b0e..91d6e90 100644 --- a/riscv/insns/vcompress_vm.h +++ b/riscv/insns/vcompress_vm.h @@ -34,7 +34,7 @@ for (reg_t i = P.VU.vstart ; i < vl; ++i) { } } -if (vl > 0 && TAIL_ZEROING) { +if (vl > 0 && P.VU.TZ) { uint8_t *tail = &P.VU.elt<uint8_t>(rd_num, pos * ((sew >> 3) * 1)); memset(tail, 0, (P.VU.vlmax - pos) * ((sew >> 3) * 1)); } diff --git a/riscv/insns/vfmerge_vfm.h b/riscv/insns/vfmerge_vfm.h index 6d12bce..e8601fe 100644 --- a/riscv/insns/vfmerge_vfm.h +++ b/riscv/insns/vfmerge_vfm.h @@ -20,6 +20,5 @@ for (reg_t i=P.VU.vstart; i<vl; ++i) { vd = use_first ? rs1 : vs2; } -VI_TAIL_ZERO(1); P.VU.vstart = 0; set_fp_exceptions; diff --git a/riscv/insns/vfmv_s_f.h b/riscv/insns/vfmv_s_f.h index a08282f..8ff6094 100644 --- a/riscv/insns/vfmv_s_f.h +++ b/riscv/insns/vfmv_s_f.h @@ -15,12 +15,5 @@ if (vl > 0) { else P.VU.elt<uint32_t>(rd_num, 0) = f32(FRS1).v; - if (TAIL_ZEROING) { - const reg_t max_len = P.VU.VLEN / sew; - for (reg_t i = 1; i < max_len; ++i) { - P.VU.elt<uint32_t>(rd_num, i) = 0; - } - } - vl = 0; } diff --git a/riscv/insns/vfmv_v_f.h b/riscv/insns/vfmv_v_f.h index c85a3e9..150298b 100644 --- a/riscv/insns/vfmv_v_f.h +++ b/riscv/insns/vfmv_v_f.h @@ -15,6 +15,5 @@ for (reg_t i=P.VU.vstart; i<vl; ++i) { vd = rs1; } -VI_TAIL_ZERO(1); P.VU.vstart = 0; set_fp_exceptions; diff --git a/riscv/insns/vid_v.h b/riscv/insns/vid_v.h index df6dd04..2291495 100644 --- a/riscv/insns/vid_v.h +++ b/riscv/insns/vid_v.h @@ -26,5 +26,4 @@ for (reg_t i = P.VU.vstart ; i < P.VU.vl; ++i) { } } -VI_TAIL_ZERO(1); P.VU.vstart = 0; diff --git a/riscv/insns/viota_m.h b/riscv/insns/viota_m.h index fde0291..55d8df1 100644 --- a/riscv/insns/viota_m.h +++ b/riscv/insns/viota_m.h @@ -49,4 +49,3 @@ for (reg_t i = 0; i < vl; ++i) { } } -VI_TAIL_ZERO(1); diff --git a/riscv/insns/vleff_v.h b/riscv/insns/vleff_v.h index ec2777a..7b870ca 100644 --- a/riscv/insns/vleff_v.h +++ b/riscv/insns/vleff_v.h @@ -9,7 +9,6 @@ const reg_t rd_num = insn.rd(); bool early_stop = false; const reg_t vlmul = P.VU.vlmul; for (reg_t i = 0; i < P.VU.vlmax && vl != 0; ++i) { - bool is_valid = true; bool is_zero = false; VI_STRIP(i); VI_ELEMENT_SKIP(i); @@ -20,23 +19,23 @@ for (reg_t i = 0; i < P.VU.vlmax && vl != 0; ++i) { switch (sew) { case e8: P.VU.elt<uint8_t>(rd_num + fn * vlmul, vreg_inx) = - is_valid ? MMU.load_uint8(baseAddr + (i * nf + fn) * 1) : 0; - is_zero = is_valid && P.VU.elt<uint8_t>(rd_num + fn * vlmul, vreg_inx) == 0; + MMU.load_uint8(baseAddr + (i * nf + fn) * 1); + is_zero = P.VU.elt<uint8_t>(rd_num + fn * vlmul, vreg_inx) == 0; break; case e16: P.VU.elt<uint16_t>(rd_num + fn * vlmul, vreg_inx) = - is_valid ? MMU.load_uint16(baseAddr + (i * nf + fn) * 2) : 0; - is_zero = is_valid && P.VU.elt<uint16_t>(rd_num + fn * vlmul, vreg_inx) == 0; + MMU.load_uint16(baseAddr + (i * nf + fn) * 2); + is_zero = P.VU.elt<uint16_t>(rd_num + fn * vlmul, vreg_inx) == 0; break; case e32: P.VU.elt<uint32_t>(rd_num + fn * vlmul, vreg_inx) = - is_valid ? MMU.load_uint32(baseAddr + (i * nf + fn) * 4) : 0; - is_zero = is_valid && P.VU.elt<uint32_t>(rd_num + fn * vlmul, vreg_inx) == 0; + MMU.load_uint32(baseAddr + (i * nf + fn) * 4); + is_zero = P.VU.elt<uint32_t>(rd_num + fn * vlmul, vreg_inx) == 0; break; case e64: P.VU.elt<uint64_t>(rd_num + fn * vlmul, vreg_inx) = - is_valid ? MMU.load_uint64(baseAddr + (i * nf + fn) * 8) : 0; - is_zero = is_valid && P.VU.elt<uint64_t>(rd_num + fn * vlmul, vreg_inx) == 0; + MMU.load_uint64(baseAddr + (i * nf + fn) * 8); + is_zero = P.VU.elt<uint64_t>(rd_num + fn * vlmul, vreg_inx) == 0; break; } diff --git a/riscv/insns/vmsbf_m.h b/riscv/insns/vmsbf_m.h index 3047cca..443fcbb 100644 --- a/riscv/insns/vmsbf_m.h +++ b/riscv/insns/vmsbf_m.h @@ -30,5 +30,4 @@ for (reg_t i = P.VU.vstart; i < vl; ++i) { } } -VI_TAIL_ZERO_MASK(rd_num); P.VU.vstart = 0; diff --git a/riscv/insns/vmsif_m.h b/riscv/insns/vmsif_m.h index 826e7cd..381088b 100644 --- a/riscv/insns/vmsif_m.h +++ b/riscv/insns/vmsif_m.h @@ -30,5 +30,4 @@ for (reg_t i = P.VU.vstart ; i < vl; ++i) { } } -VI_TAIL_ZERO_MASK(rd_num); P.VU.vstart = 0; diff --git a/riscv/insns/vmsof_m.h b/riscv/insns/vmsof_m.h index 48805f7..d66002d 100644 --- a/riscv/insns/vmsof_m.h +++ b/riscv/insns/vmsof_m.h @@ -28,5 +28,4 @@ for (reg_t i = P.VU.vstart ; i < vl; ++i) { } } -VI_TAIL_ZERO_MASK(rd_num); P.VU.vstart = 0; diff --git a/riscv/insns/vmv_s_x.h b/riscv/insns/vmv_s_x.h index f19fa61..948b5be 100644 --- a/riscv/insns/vmv_s_x.h +++ b/riscv/insns/vmv_s_x.h @@ -24,25 +24,5 @@ if (vl > 0) { break; } - if (TAIL_ZEROING) { - const reg_t max_len = P.VU.VLEN / sew; - for (reg_t i = 1; i < max_len; ++i) { - switch(sew) { - case e8: - P.VU.elt<uint8_t>(rd_num, i) = 0; - break; - case e16: - P.VU.elt<uint16_t>(rd_num, i) = 0; - break; - case e32: - P.VU.elt<uint32_t>(rd_num, i) = 0; - break; - default: - P.VU.elt<uint64_t>(rd_num, i) = 0; - break; - } - } - } - vl = 0; } diff --git a/riscv/insns/vsuxb_v.h b/riscv/insns/vsuxb_v.h index cf928f8..0dfe024 100644 --- a/riscv/insns/vsuxb_v.h +++ b/riscv/insns/vsuxb_v.h @@ -8,30 +8,25 @@ reg_t vs3 = insn.rd(); reg_t vlmax = P.VU.vlmax; VI_DUPLICATE_VREG(stride, vlmax); for (reg_t i = 0; i < vlmax && vl != 0; ++i) { - bool is_valid = true; VI_ELEMENT_SKIP(i); VI_STRIP(i) switch (P.VU.vsew) { case e8: - if (is_valid) - MMU.store_uint8(baseAddr + index[i], - P.VU.elt<uint8_t>(vs3, vreg_inx)); + MMU.store_uint8(baseAddr + index[i], + P.VU.elt<uint8_t>(vs3, vreg_inx)); break; case e16: - if (is_valid) - MMU.store_uint8(baseAddr + index[i], - P.VU.elt<uint16_t>(vs3, vreg_inx)); + MMU.store_uint8(baseAddr + index[i], + P.VU.elt<uint16_t>(vs3, vreg_inx)); break; case e32: - if (is_valid) - MMU.store_uint8(baseAddr + index[i], + MMU.store_uint8(baseAddr + index[i], P.VU.elt<uint32_t>(vs3, vreg_inx)); break; case e64: - if (is_valid) - MMU.store_uint8(baseAddr + index[i], - P.VU.elt<uint64_t>(vs3, vreg_inx)); + MMU.store_uint8(baseAddr + index[i], + P.VU.elt<uint64_t>(vs3, vreg_inx)); break; } } diff --git a/riscv/insns/vsuxe_v.h b/riscv/insns/vsuxe_v.h index 8bd7545..5e4d3a2 100644 --- a/riscv/insns/vsuxe_v.h +++ b/riscv/insns/vsuxe_v.h @@ -9,30 +9,25 @@ reg_t vs3 = insn.rd(); reg_t vlmax = P.VU.vlmax; VI_DUPLICATE_VREG(stride, vlmax); for (reg_t i = 0; i < vlmax && vl != 0; ++i) { - bool is_valid = true; VI_ELEMENT_SKIP(i); VI_STRIP(i) switch (sew) { case e8: - if (is_valid) - MMU.store_uint8(baseAddr + index[i], - P.VU.elt<uint8_t>(vs3, vreg_inx)); + MMU.store_uint8(baseAddr + index[i], + P.VU.elt<uint8_t>(vs3, vreg_inx)); break; case e16: - if (is_valid) - MMU.store_uint16(baseAddr + index[i], - P.VU.elt<uint16_t>(vs3, vreg_inx)); + MMU.store_uint16(baseAddr + index[i], + P.VU.elt<uint16_t>(vs3, vreg_inx)); break; case e32: - if (is_valid) - MMU.store_uint32(baseAddr + index[i], - P.VU.elt<uint32_t>(vs3, vreg_inx)); + MMU.store_uint32(baseAddr + index[i], + P.VU.elt<uint32_t>(vs3, vreg_inx)); break; case e64: - if (is_valid) - MMU.store_uint64(baseAddr + index[i], - P.VU.elt<uint64_t>(vs3, vreg_inx)); + MMU.store_uint64(baseAddr + index[i], + P.VU.elt<uint64_t>(vs3, vreg_inx)); break; } } diff --git a/riscv/insns/vsuxh_v.h b/riscv/insns/vsuxh_v.h index 1d5a1bd..c6f8be4 100644 --- a/riscv/insns/vsuxh_v.h +++ b/riscv/insns/vsuxh_v.h @@ -8,25 +8,21 @@ reg_t vs3 = insn.rd(); reg_t vlmax = P.VU.vlmax; VI_DUPLICATE_VREG(stride, vlmax); for (reg_t i = 0; i < vlmax && vl != 0; ++i) { - bool is_valid = true; VI_ELEMENT_SKIP(i); VI_STRIP(i) switch (P.VU.vsew) { case e16: - if (is_valid) - MMU.store_uint16(baseAddr + index[i], - P.VU.elt<uint16_t>(vs3, vreg_inx)); + MMU.store_uint16(baseAddr + index[i], + P.VU.elt<uint16_t>(vs3, vreg_inx)); break; case e32: - if (is_valid) - MMU.store_uint16(baseAddr + index[i], - P.VU.elt<uint32_t>(vs3, vreg_inx)); + MMU.store_uint16(baseAddr + index[i], + P.VU.elt<uint32_t>(vs3, vreg_inx)); break; case e64: - if (is_valid) - MMU.store_uint16(baseAddr + index[i], - P.VU.elt<uint64_t>(vs3, vreg_inx)); + MMU.store_uint16(baseAddr + index[i], + P.VU.elt<uint64_t>(vs3, vreg_inx)); break; } } diff --git a/riscv/insns/vsuxw_v.h b/riscv/insns/vsuxw_v.h index ec1a8fe..f133e77 100644 --- a/riscv/insns/vsuxw_v.h +++ b/riscv/insns/vsuxw_v.h @@ -8,20 +8,17 @@ reg_t vs3 = insn.rd(); reg_t vlmax = P.VU.vlmax; VI_DUPLICATE_VREG(stride, vlmax); for (reg_t i = 0; i < vlmax && vl != 0; ++i) { - bool is_valid = true; VI_ELEMENT_SKIP(i); VI_STRIP(i) switch (P.VU.vsew) { case e32: - if (is_valid) - MMU.store_uint32(baseAddr + index[i], - P.VU.elt<uint32_t>(vs3, vreg_inx)); + MMU.store_uint32(baseAddr + index[i], + P.VU.elt<uint32_t>(vs3, vreg_inx)); break; case e64: - if (is_valid) - MMU.store_uint32(baseAddr + index[i], - P.VU.elt<uint64_t>(vs3, vreg_inx)); + MMU.store_uint32(baseAddr + index[i], + P.VU.elt<uint64_t>(vs3, vreg_inx)); break; } } |