From f3141174dd990672e28db0d952e13ae19f08e825 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Mon, 16 Oct 2023 14:35:05 -0700 Subject: [PATCH] target/sparc: Implement UDIVX and SDIVX inline Tested-by: Mark Cave-Ayland Acked-by: Mark Cave-Ayland Signed-off-by: Richard Henderson --- target/sparc/helper.c | 24 --------- target/sparc/helper.h | 4 -- target/sparc/insns.decode | 4 +- target/sparc/translate.c | 109 +++++++++++++++++++++++++++++++++----- 4 files changed, 97 insertions(+), 44 deletions(-) diff --git a/target/sparc/helper.c b/target/sparc/helper.c index 6117e99b55..bd10b60e4b 100644 --- a/target/sparc/helper.c +++ b/target/sparc/helper.c @@ -129,30 +129,6 @@ uint64_t helper_sdiv(CPUSPARCState *env, target_ulong a, target_ulong b) return (uint32_t)r; } -#ifdef TARGET_SPARC64 -int64_t helper_sdivx(CPUSPARCState *env, int64_t a, int64_t b) -{ - if (b == 0) { - /* Raise divide by zero trap. */ - cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC()); - } else if (b == -1) { - /* Avoid overflow trap with i386 divide insn. */ - return -a; - } else { - return a / b; - } -} - -uint64_t helper_udivx(CPUSPARCState *env, uint64_t a, uint64_t b) -{ - if (b == 0) { - /* Raise divide by zero trap. */ - cpu_raise_exception_ra(env, TT_DIV_ZERO, GETPC()); - } - return a / b; -} -#endif - target_ulong helper_taddcctv(CPUSPARCState *env, target_ulong src1, target_ulong src2) { diff --git a/target/sparc/helper.h b/target/sparc/helper.h index decd94c0d6..55eff66283 100644 --- a/target/sparc/helper.h +++ b/target/sparc/helper.h @@ -31,10 +31,6 @@ DEF_HELPER_FLAGS_3(udiv, TCG_CALL_NO_WG, i64, env, tl, tl) DEF_HELPER_FLAGS_3(sdiv, TCG_CALL_NO_WG, i64, env, tl, tl) DEF_HELPER_3(taddcctv, tl, env, tl, tl) DEF_HELPER_3(tsubcctv, tl, env, tl, tl) -#ifdef TARGET_SPARC64 -DEF_HELPER_FLAGS_3(sdivx, TCG_CALL_NO_WG, s64, env, s64, s64) -DEF_HELPER_FLAGS_3(udivx, TCG_CALL_NO_WG, i64, env, i64, i64) -#endif #if !defined(CONFIG_USER_ONLY) || defined(TARGET_SPARC64) DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32) DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32) diff --git a/target/sparc/insns.decode b/target/sparc/insns.decode index 0552f1447d..52f54b87cc 100644 --- a/target/sparc/insns.decode +++ b/target/sparc/insns.decode @@ -182,8 +182,8 @@ UMUL 10 ..... 0.1010 ..... . ............. @r_r_ri_cc SMUL 10 ..... 0.1011 ..... . ............. @r_r_ri_cc MULScc 10 ..... 100100 ..... . ............. @r_r_ri_cc1 -UDIVX 10 ..... 001101 ..... . ............. @r_r_ri_cc0 -SDIVX 10 ..... 101101 ..... . ............. @r_r_ri_cc0 +UDIVX 10 ..... 001101 ..... . ............. @r_r_ri +SDIVX 10 ..... 101101 ..... . ............. @r_r_ri UDIV 10 ..... 0.1110 ..... . ............. @r_r_ri_cc SDIV 10 ..... 0.1111 ..... . ............. @r_r_ri_cc diff --git a/target/sparc/translate.c b/target/sparc/translate.c index 3564c6032e..95cc4c71f4 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -51,12 +51,10 @@ # define gen_helper_restored(E) qemu_build_not_reached() # define gen_helper_retry(E) qemu_build_not_reached() # define gen_helper_saved(E) qemu_build_not_reached() -# define gen_helper_sdivx(D, E, A, B) qemu_build_not_reached() # define gen_helper_set_softint(E, S) qemu_build_not_reached() # define gen_helper_tick_get_count(D, E, T, C) qemu_build_not_reached() # define gen_helper_tick_set_count(P, S) qemu_build_not_reached() # define gen_helper_tick_set_limit(P, S) qemu_build_not_reached() -# define gen_helper_udivx(D, E, A, B) qemu_build_not_reached() # define gen_helper_wrccr(E, S) qemu_build_not_reached() # define gen_helper_wrcwp(E, S) qemu_build_not_reached() # define gen_helper_wrgl(E, S) qemu_build_not_reached() @@ -579,16 +577,6 @@ static void gen_op_smul(TCGv dst, TCGv src1, TCGv src2) gen_op_multiply(dst, src1, src2, 1); } -static void gen_op_udivx(TCGv dst, TCGv src1, TCGv src2) -{ - gen_helper_udivx(dst, tcg_env, src1, src2); -} - -static void gen_op_sdivx(TCGv dst, TCGv src1, TCGv src2) -{ - gen_helper_sdivx(dst, tcg_env, src1, src2); -} - static void gen_op_udiv(TCGv dst, TCGv src1, TCGv src2) { #ifdef TARGET_SPARC64 @@ -3580,8 +3568,6 @@ TRANS(UMUL, MUL, do_logic, a, gen_op_umul, NULL) TRANS(SMUL, MUL, do_logic, a, gen_op_smul, NULL) TRANS(MULScc, ALL, do_arith, a, NULL, NULL, gen_op_mulscc) -TRANS(UDIVX, 64, do_arith, a, gen_op_udivx, NULL, NULL) -TRANS(SDIVX, 64, do_arith, a, gen_op_sdivx, NULL, NULL) TRANS(UDIV, DIV, do_arith, a, gen_op_udiv, NULL, gen_op_udivcc) TRANS(SDIV, DIV, do_arith, a, gen_op_sdiv, NULL, gen_op_sdivcc) @@ -3605,6 +3591,101 @@ static bool trans_OR(DisasContext *dc, arg_r_r_ri_cc *a) return do_logic(dc, a, tcg_gen_or_tl, tcg_gen_ori_tl); } +static bool trans_UDIVX(DisasContext *dc, arg_r_r_ri *a) +{ + TCGv dst, src1, src2; + + if (!avail_64(dc)) { + return false; + } + /* For simplicity, we under-decoded the rs2 form. */ + if (!a->imm && a->rs2_or_imm & ~0x1f) { + return false; + } + + if (unlikely(a->rs2_or_imm == 0)) { + gen_exception(dc, TT_DIV_ZERO); + return true; + } + + if (a->imm) { + src2 = tcg_constant_tl(a->rs2_or_imm); + } else { + TCGLabel *lab; + + finishing_insn(dc); + flush_cond(dc); + + lab = delay_exception(dc, TT_DIV_ZERO); + src2 = cpu_regs[a->rs2_or_imm]; + tcg_gen_brcondi_tl(TCG_COND_EQ, src2, 0, lab); + } + + dst = gen_dest_gpr(dc, a->rd); + src1 = gen_load_gpr(dc, a->rs1); + + tcg_gen_divu_tl(dst, src1, src2); + gen_store_gpr(dc, a->rd, dst); + return advance_pc(dc); +} + +static bool trans_SDIVX(DisasContext *dc, arg_r_r_ri *a) +{ + TCGv dst, src1, src2; + + if (!avail_64(dc)) { + return false; + } + /* For simplicity, we under-decoded the rs2 form. */ + if (!a->imm && a->rs2_or_imm & ~0x1f) { + return false; + } + + if (unlikely(a->rs2_or_imm == 0)) { + gen_exception(dc, TT_DIV_ZERO); + return true; + } + + dst = gen_dest_gpr(dc, a->rd); + src1 = gen_load_gpr(dc, a->rs1); + + if (a->imm) { + if (unlikely(a->rs2_or_imm == -1)) { + tcg_gen_neg_tl(dst, src1); + gen_store_gpr(dc, a->rd, dst); + return advance_pc(dc); + } + src2 = tcg_constant_tl(a->rs2_or_imm); + } else { + TCGLabel *lab; + TCGv t1, t2; + + finishing_insn(dc); + flush_cond(dc); + + lab = delay_exception(dc, TT_DIV_ZERO); + src2 = cpu_regs[a->rs2_or_imm]; + tcg_gen_brcondi_tl(TCG_COND_EQ, src2, 0, lab); + + /* + * Need to avoid INT64_MIN / -1, which will trap on x86 host. + * Set SRC2 to 1 as a new divisor, to produce the correct result. + */ + t1 = tcg_temp_new(); + t2 = tcg_temp_new(); + tcg_gen_setcondi_tl(TCG_COND_EQ, t1, src1, (target_long)INT64_MIN); + tcg_gen_setcondi_tl(TCG_COND_EQ, t2, src2, -1); + tcg_gen_and_tl(t1, t1, t2); + tcg_gen_movcond_tl(TCG_COND_NE, t1, t1, tcg_constant_tl(0), + tcg_constant_tl(1), src2); + src2 = t1; + } + + tcg_gen_div_tl(dst, src1, src2); + gen_store_gpr(dc, a->rd, dst); + return advance_pc(dc); +} + static bool gen_edge(DisasContext *dc, arg_r_r_r *a, int width, bool cc, bool left) {