From b88ce6f246f40fd293803562ac60f29ff1ce9f6f Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Thu, 5 Oct 2023 03:12:12 -0700 Subject: [PATCH] target/sparc: Move EDGE* to decodetree Tested-by: Mark Cave-Ayland Acked-by: Mark Cave-Ayland Signed-off-by: Richard Henderson --- target/sparc/insns.decode | 22 +++- target/sparc/translate.c | 271 ++++++++++++++++---------------------- 2 files changed, 133 insertions(+), 160 deletions(-) diff --git a/target/sparc/insns.decode b/target/sparc/insns.decode index 5df3b1add4..a9630509bd 100644 --- a/target/sparc/insns.decode +++ b/target/sparc/insns.decode @@ -35,6 +35,9 @@ CALL 01 i:s30 @r_r_ri_cc0 .. rd:5 ...... rs1:5 imm:1 rs2_or_imm:s13 &r_r_ri_cc cc=0 @r_r_ri_cc1 .. rd:5 ...... rs1:5 imm:1 rs2_or_imm:s13 &r_r_ri_cc cc=1 +&r_r_r rd rs1 rs2 +@r_r_r .. rd:5 ...... rs1:5 . ........ rs2:5 &r_r_r + { [ STBAR 10 00000 101000 01111 0 0000000000000 @@ -231,7 +234,24 @@ RESTORE 10 ..... 111101 ..... . ............. @r_r_ri DONE 10 00000 111110 00000 0 0000000000000 RETRY 10 00001 111110 00000 0 0000000000000 -NCP 10 ----- 110110 ----- --------- ----- # v8 CPop1 +{ + [ + EDGE8cc 10 ..... 110110 ..... 0 0000 0000 ..... @r_r_r + EDGE8N 10 ..... 110110 ..... 0 0000 0001 ..... @r_r_r + EDGE8Lcc 10 ..... 110110 ..... 0 0000 0010 ..... @r_r_r + EDGE8LN 10 ..... 110110 ..... 0 0000 0011 ..... @r_r_r + EDGE16cc 10 ..... 110110 ..... 0 0000 0100 ..... @r_r_r + EDGE16N 10 ..... 110110 ..... 0 0000 0101 ..... @r_r_r + EDGE16Lcc 10 ..... 110110 ..... 0 0000 0110 ..... @r_r_r + EDGE16LN 10 ..... 110110 ..... 0 0000 0111 ..... @r_r_r + EDGE32cc 10 ..... 110110 ..... 0 0000 1000 ..... @r_r_r + EDGE32N 10 ..... 110110 ..... 0 0000 1001 ..... @r_r_r + EDGE32Lcc 10 ..... 110110 ..... 0 0000 1010 ..... @r_r_r + EDGE32LN 10 ..... 110110 ..... 0 0000 1011 ..... @r_r_r + ] + NCP 10 ----- 110110 ----- --------- ----- # v8 CPop1 +} + NCP 10 ----- 110111 ----- --------- ----- # v8 CPop2 ## diff --git a/target/sparc/translate.c b/target/sparc/translate.c index bf8c5a16b6..2f38453dc6 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -2728,93 +2728,6 @@ static void gen_load_trap_state_at_tl(TCGv_ptr r_tsptr) } } -static void gen_edge(DisasContext *dc, TCGv dst, TCGv s1, TCGv s2, - int width, bool cc, bool left) -{ - TCGv lo1, lo2; - uint64_t amask, tabl, tabr; - int shift, imask, omask; - - if (cc) { - tcg_gen_mov_tl(cpu_cc_src, s1); - tcg_gen_mov_tl(cpu_cc_src2, s2); - tcg_gen_sub_tl(cpu_cc_dst, s1, s2); - tcg_gen_movi_i32(cpu_cc_op, CC_OP_SUB); - dc->cc_op = CC_OP_SUB; - } - - /* Theory of operation: there are two tables, left and right (not to - be confused with the left and right versions of the opcode). These - are indexed by the low 3 bits of the inputs. To make things "easy", - these tables are loaded into two constants, TABL and TABR below. - The operation index = (input & imask) << shift calculates the index - into the constant, while val = (table >> index) & omask calculates - the value we're looking for. */ - switch (width) { - case 8: - imask = 0x7; - shift = 3; - omask = 0xff; - if (left) { - tabl = 0x80c0e0f0f8fcfeffULL; - tabr = 0xff7f3f1f0f070301ULL; - } else { - tabl = 0x0103070f1f3f7fffULL; - tabr = 0xfffefcf8f0e0c080ULL; - } - break; - case 16: - imask = 0x6; - shift = 1; - omask = 0xf; - if (left) { - tabl = 0x8cef; - tabr = 0xf731; - } else { - tabl = 0x137f; - tabr = 0xfec8; - } - break; - case 32: - imask = 0x4; - shift = 0; - omask = 0x3; - if (left) { - tabl = (2 << 2) | 3; - tabr = (3 << 2) | 1; - } else { - tabl = (1 << 2) | 3; - tabr = (3 << 2) | 2; - } - break; - default: - abort(); - } - - lo1 = tcg_temp_new(); - lo2 = tcg_temp_new(); - tcg_gen_andi_tl(lo1, s1, imask); - tcg_gen_andi_tl(lo2, s2, imask); - tcg_gen_shli_tl(lo1, lo1, shift); - tcg_gen_shli_tl(lo2, lo2, shift); - - tcg_gen_shr_tl(lo1, tcg_constant_tl(tabl), lo1); - tcg_gen_shr_tl(lo2, tcg_constant_tl(tabr), lo2); - tcg_gen_andi_tl(lo1, lo1, omask); - tcg_gen_andi_tl(lo2, lo2, omask); - - amask = -8; - if (AM_CHECK(dc)) { - amask &= 0xffffffffULL; - } - tcg_gen_andi_tl(s1, s1, amask); - tcg_gen_andi_tl(s2, s2, amask); - - /* Compute dst = (s1 == s2 ? lo1 : lo1 & lo2). */ - tcg_gen_and_tl(lo2, lo2, lo1); - tcg_gen_movcond_tl(TCG_COND_EQ, dst, s1, s2, lo1, lo2); -} - static void gen_alignaddr(TCGv dst, TCGv s1, TCGv s2, bool left) { TCGv tmp = tcg_temp_new(); @@ -2877,6 +2790,8 @@ static int extract_qfpreg(DisasContext *dc, int x) # define avail_64(C) true # define avail_GL(C) ((C)->def->features & CPU_FEATURE_GL) # define avail_HYPV(C) ((C)->def->features & CPU_FEATURE_HYPV) +# define avail_VIS1(C) ((C)->def->features & CPU_FEATURE_VIS1) +# define avail_VIS2(C) ((C)->def->features & CPU_FEATURE_VIS2) #else # define avail_32(C) true # define avail_ASR17(C) ((C)->def->features & CPU_FEATURE_ASR17) @@ -2887,6 +2802,8 @@ static int extract_qfpreg(DisasContext *dc, int x) # define avail_64(C) false # define avail_GL(C) false # define avail_HYPV(C) false +# define avail_VIS1(C) false +# define avail_VIS2(C) false #endif /* Default case for non jump instructions. */ @@ -4187,6 +4104,113 @@ static bool trans_MULScc(DisasContext *dc, arg_r_r_ri_cc *a) return do_arith(dc, a, CC_OP_ADD, NULL, NULL, gen_op_mulscc); } +static bool gen_edge(DisasContext *dc, arg_r_r_r *a, + int width, bool cc, bool left) +{ + TCGv dst, s1, s2, lo1, lo2; + uint64_t amask, tabl, tabr; + int shift, imask, omask; + + dst = gen_dest_gpr(dc, a->rd); + s1 = gen_load_gpr(dc, a->rs1); + s2 = gen_load_gpr(dc, a->rs2); + + if (cc) { + tcg_gen_mov_tl(cpu_cc_src, s1); + tcg_gen_mov_tl(cpu_cc_src2, s2); + tcg_gen_sub_tl(cpu_cc_dst, s1, s2); + tcg_gen_movi_i32(cpu_cc_op, CC_OP_SUB); + dc->cc_op = CC_OP_SUB; + } + + /* + * Theory of operation: there are two tables, left and right (not to + * be confused with the left and right versions of the opcode). These + * are indexed by the low 3 bits of the inputs. To make things "easy", + * these tables are loaded into two constants, TABL and TABR below. + * The operation index = (input & imask) << shift calculates the index + * into the constant, while val = (table >> index) & omask calculates + * the value we're looking for. + */ + switch (width) { + case 8: + imask = 0x7; + shift = 3; + omask = 0xff; + if (left) { + tabl = 0x80c0e0f0f8fcfeffULL; + tabr = 0xff7f3f1f0f070301ULL; + } else { + tabl = 0x0103070f1f3f7fffULL; + tabr = 0xfffefcf8f0e0c080ULL; + } + break; + case 16: + imask = 0x6; + shift = 1; + omask = 0xf; + if (left) { + tabl = 0x8cef; + tabr = 0xf731; + } else { + tabl = 0x137f; + tabr = 0xfec8; + } + break; + case 32: + imask = 0x4; + shift = 0; + omask = 0x3; + if (left) { + tabl = (2 << 2) | 3; + tabr = (3 << 2) | 1; + } else { + tabl = (1 << 2) | 3; + tabr = (3 << 2) | 2; + } + break; + default: + abort(); + } + + lo1 = tcg_temp_new(); + lo2 = tcg_temp_new(); + tcg_gen_andi_tl(lo1, s1, imask); + tcg_gen_andi_tl(lo2, s2, imask); + tcg_gen_shli_tl(lo1, lo1, shift); + tcg_gen_shli_tl(lo2, lo2, shift); + + tcg_gen_shr_tl(lo1, tcg_constant_tl(tabl), lo1); + tcg_gen_shr_tl(lo2, tcg_constant_tl(tabr), lo2); + tcg_gen_andi_tl(lo1, lo1, omask); + tcg_gen_andi_tl(lo2, lo2, omask); + + amask = address_mask_i(dc, -8); + tcg_gen_andi_tl(s1, s1, amask); + tcg_gen_andi_tl(s2, s2, amask); + + /* Compute dst = (s1 == s2 ? lo1 : lo1 & lo2). */ + tcg_gen_and_tl(lo2, lo2, lo1); + tcg_gen_movcond_tl(TCG_COND_EQ, dst, s1, s2, lo1, lo2); + + gen_store_gpr(dc, a->rd, dst); + return advance_pc(dc); +} + +TRANS(EDGE8cc, VIS1, gen_edge, a, 8, 1, 0) +TRANS(EDGE8Lcc, VIS1, gen_edge, a, 8, 1, 1) +TRANS(EDGE16cc, VIS1, gen_edge, a, 16, 1, 0) +TRANS(EDGE16Lcc, VIS1, gen_edge, a, 16, 1, 1) +TRANS(EDGE32cc, VIS1, gen_edge, a, 32, 1, 0) +TRANS(EDGE32Lcc, VIS1, gen_edge, a, 32, 1, 1) + +TRANS(EDGE8N, VIS2, gen_edge, a, 8, 0, 0) +TRANS(EDGE8LN, VIS2, gen_edge, a, 8, 0, 1) +TRANS(EDGE16N, VIS2, gen_edge, a, 16, 0, 0) +TRANS(EDGE16LN, VIS2, gen_edge, a, 16, 0, 1) +TRANS(EDGE32N, VIS2, gen_edge, a, 32, 0, 0) +TRANS(EDGE32LN, VIS2, gen_edge, a, 32, 0, 1) + static bool do_shift_r(DisasContext *dc, arg_shiftr *a, bool l, bool u) { TCGv dst, src1, src2; @@ -5075,89 +5099,18 @@ static void disas_sparc_legacy(DisasContext *dc, unsigned int insn) switch (opf) { case 0x000: /* VIS I edge8cc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 8, 1, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x001: /* VIS II edge8n */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 8, 0, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x002: /* VIS I edge8lcc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 8, 1, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x003: /* VIS II edge8ln */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 8, 0, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x004: /* VIS I edge16cc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 16, 1, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x005: /* VIS II edge16n */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 16, 0, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x006: /* VIS I edge16lcc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 16, 1, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x007: /* VIS II edge16ln */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 16, 0, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x008: /* VIS I edge32cc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 32, 1, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x009: /* VIS II edge32n */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 32, 0, 0); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x00a: /* VIS I edge32lcc */ - CHECK_FPU_FEATURE(dc, VIS1); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 32, 1, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; case 0x00b: /* VIS II edge32ln */ - CHECK_FPU_FEATURE(dc, VIS2); - cpu_src1 = gen_load_gpr(dc, rs1); - cpu_src2 = gen_load_gpr(dc, rs2); - gen_edge(dc, cpu_dst, cpu_src1, cpu_src2, 32, 0, 1); - gen_store_gpr(dc, rd, cpu_dst); - break; + g_assert_not_reached(); /* in decodetree */ case 0x010: /* VIS I array8 */ CHECK_FPU_FEATURE(dc, VIS1); cpu_src1 = gen_load_gpr(dc, rs1);