From 1c35bd706f38198f65b0001552cd6a37c9bb066c Mon Sep 17 00:00:00 2001 From: Luke Kenneth Casson Leighton Date: Sat, 6 Oct 2018 10:24:25 +0100 Subject: [PATCH] create ldsp immediate offset overrides this allows C.LWSP/C.LDSP to do predicated load/stores from contiguous blocks of memory --- id_regs.py | 6 ++++++ riscv/sv.cc | 24 +++++++++++++++++++++++- riscv/sv_decode.h | 5 +++++ 3 files changed, 34 insertions(+), 1 deletion(-) diff --git a/id_regs.py b/id_regs.py index 782cbcb..d6259cd 100644 --- a/id_regs.py +++ b/id_regs.py @@ -129,6 +129,12 @@ def find_registers(fname, insn, twin_predication, immed_offset): if immed_offset: # C.LWSP predargs.append('&src_pred') fsrc = insn in ['c_flwsp', 'c_fldsp'] + c_sp_width = {'c_lwsp': 4, 'c_ldsp': 8, 'c_lqsp': 16, + 'c_flwsp': 4, 'c_fldsp': 8, + 'c_swsp': 4, 'c_sdsp': 8, 'c_sqsp': 16, + 'c_fswsp': 4, 'c_fsdsp': 8} + iwidth = c_sp_width[insn] + res.append('#define IMMEDWIDTH %d' % (iwidth)) res.append('#define SRC_PREDINT %d' % (0 if fsrc else 1)) if twin_predication: diff --git a/riscv/sv.cc b/riscv/sv.cc index a6bf42a..8088ef4 100644 --- a/riscv/sv.cc +++ b/riscv/sv.cc @@ -92,7 +92,7 @@ uint64_t sv_insn_t::remap(uint64_t reg, bool intreg, int voffs) // we return the re-mapped register... if (!r->isvec) // scalar { - return reg; + return reg; } vloop_continue = true; @@ -167,3 +167,25 @@ bool sv_insn_t::stop_vloop(void) } +/* c_lwsp's immediate offset is turned into a Vector "unit stride" if + * x2 (sp by convention) is marked as vectorised. + * + */ +uint64_t sv_insn_t::_rvc_spoffs_imm(uint64_t elwidth, uint64_t offs) +{ + sv_reg_entry *r = get_regentry(X_SP, 1); + if (!r->active) + { + return offs; + } + vloop_continue = true; + reg_t reg = r->regidx; + if (!r->isvec) + { + return offs; + } + offs += (*offs_imm) * elwidth; + + return offs; +} + diff --git a/riscv/sv_decode.h b/riscv/sv_decode.h index 8065013..d85f047 100644 --- a/riscv/sv_decode.h +++ b/riscv/sv_decode.h @@ -26,7 +26,11 @@ public: int *o_rd, int *o_rs1, int *o_rs2, int *o_rs3, int *o_imm) : insn_t(bits), p(pr), vloop_continue(false), fimap(f), offs_rd(o_rd), offs_rs1(o_rs1), offs_rs2(o_rs2), offs_rs3(o_rs3), + offs_imm(o_imm), prd(p_rd), prs1(p_rs1), prs2(p_rs2), prs3(p_rs3) {} + uint64_t _rvc_spoffs_imm(uint64_t elwidth, uint64_t baseoffs); + uint64_t rvc_ldsp_imm() { return _rvc_spoffs_imm(4, insn_t::rvc_ldsp_imm()); } + uint64_t rvc_lwsp_imm() { return _rvc_spoffs_imm(8, insn_t::rvc_lwsp_imm()); } uint64_t rd () { return predicated(_rd (), *offs_rd, prd); } uint64_t rs1() { return predicated(_rs1(), *offs_rs1, prs1); } uint64_t rs2() { return predicated(_rs2(), *offs_rs2, prs2); } @@ -65,6 +69,7 @@ private: int *offs_rs1; int *offs_rs2; int *offs_rs3; + int *offs_imm; uint64_t &prd; uint64_t &prs1; uint64_t &prs2; -- 2.30.2