2 use ieee.std_logic_1164.all;
3 use ieee.numeric_std.all;
6 use work.decode_types.all;
9 use work.crhelpers.all;
10 use work.insn_helpers.all;
11 use work.ppc_fx_insns.all;
15 EX1_BYPASS : boolean := true;
16 HAS_FPU : boolean := true;
17 -- Non-zero to enable log data collection
18 LOG_LENGTH : natural := 0
25 flush_out : out std_ulogic;
26 busy_out : out std_ulogic;
28 e_in : in Decode2ToExecute1Type;
29 l_in : in Loadstore1ToExecute1Type;
30 fp_in : in FPUToExecute1Type;
32 ext_irq_in : std_ulogic;
35 l_out : out Execute1ToLoadstore1Type;
36 f_out : out Execute1ToFetch1Type;
37 fp_out : out Execute1ToFPUType;
39 e_out : out Execute1ToWritebackType;
41 dbg_msr_out : out std_ulogic_vector(63 downto 0);
43 icache_inval : out std_ulogic;
44 terminate_out : out std_ulogic;
46 log_out : out std_ulogic_vector(14 downto 0);
47 log_rd_addr : out std_ulogic_vector(31 downto 0);
48 log_rd_data : in std_ulogic_vector(63 downto 0);
49 log_wr_addr : in std_ulogic_vector(31 downto 0)
53 architecture behaviour of execute1 is
54 type reg_type is record
55 e : Execute1ToWritebackType;
57 terminate: std_ulogic;
58 fp_exception_next : std_ulogic;
59 trace_next : std_ulogic;
60 prev_op : insn_type_t;
61 lr_update : std_ulogic;
62 next_lr : std_ulogic_vector(63 downto 0);
63 mul_in_progress : std_ulogic;
64 mul_finish : std_ulogic;
65 div_in_progress : std_ulogic;
66 cntz_in_progress : std_ulogic;
67 slow_op_insn : insn_type_t;
68 slow_op_dest : gpr_index_t;
69 slow_op_rc : std_ulogic;
70 slow_op_oe : std_ulogic;
71 slow_op_xerc : xer_common_t;
72 last_nia : std_ulogic_vector(63 downto 0);
73 redirect : std_ulogic;
76 vector : integer range 0 to 16#fff#;
77 br_offset : std_ulogic_vector(63 downto 0);
78 redir_mode : std_ulogic_vector(3 downto 0);
79 log_addr_spr : std_ulogic_vector(31 downto 0);
81 constant reg_type_init : reg_type :=
82 (e => Execute1ToWritebackInit,
83 busy => '0', lr_update => '0', terminate => '0',
84 fp_exception_next => '0', trace_next => '0', prev_op => OP_ILLEGAL,
85 mul_in_progress => '0', mul_finish => '0', div_in_progress => '0', cntz_in_progress => '0',
86 slow_op_insn => OP_ILLEGAL, slow_op_rc => '0', slow_op_oe => '0', slow_op_xerc => xerc_init,
87 next_lr => (others => '0'), last_nia => (others => '0'),
88 redirect => '0', abs_br => '0', do_intr => '0', vector => 0,
89 br_offset => (others => '0'), redir_mode => "0000",
90 others => (others => '0'));
92 signal r, rin : reg_type;
94 signal a_in, b_in, c_in : std_ulogic_vector(63 downto 0);
95 signal cr_in : std_ulogic_vector(31 downto 0);
97 signal valid_in : std_ulogic;
98 signal ctrl: ctrl_t := (irq_state => WRITE_SRR0, others => (others => '0'));
99 signal ctrl_tmp: ctrl_t := (irq_state => WRITE_SRR0, others => (others => '0'));
100 signal right_shift, rot_clear_left, rot_clear_right: std_ulogic;
101 signal rot_sign_ext: std_ulogic;
102 signal rotator_result: std_ulogic_vector(63 downto 0);
103 signal rotator_carry: std_ulogic;
104 signal logical_result: std_ulogic_vector(63 downto 0);
105 signal countzero_result: std_ulogic_vector(63 downto 0);
108 signal x_to_multiply: MultiplyInputType;
109 signal multiply_to_x: MultiplyOutputType;
112 signal x_to_divider: Execute1ToDividerType;
113 signal divider_to_x: DividerToExecute1Type;
115 -- random number generator signals
116 signal random_raw : std_ulogic_vector(63 downto 0);
117 signal random_cond : std_ulogic_vector(63 downto 0);
118 signal random_err : std_ulogic;
120 -- signals for logging
121 signal exception_log : std_ulogic;
122 signal irq_valid_log : std_ulogic;
124 type privilege_level is (USER, SUPER);
125 type op_privilege_array is array(insn_type_t) of privilege_level;
126 constant op_privilege: op_privilege_array := (
135 function instr_is_privileged(op: insn_type_t; insn: std_ulogic_vector(31 downto 0))
138 if op_privilege(op) = SUPER then
140 elsif op = OP_MFSPR or op = OP_MTSPR then
141 return insn(20) = '1';
147 procedure set_carry(e: inout Execute1ToWritebackType;
148 carry32 : in std_ulogic;
149 carry : in std_ulogic) is
151 e.xerc.ca32 := carry32;
153 e.write_xerc_enable := '1';
156 procedure set_ov(e: inout Execute1ToWritebackType;
158 ov32 : in std_ulogic) is
165 e.write_xerc_enable := '1';
168 function calc_ov(msb_a : std_ulogic; msb_b: std_ulogic;
169 ca: std_ulogic; msb_r: std_ulogic) return std_ulogic is
171 return (ca xor msb_r) and not (msb_a xor msb_b);
174 function decode_input_carry(ic : carry_in_t;
175 xerc : xer_common_t) return std_ulogic is
189 function msr_copy(msr: std_ulogic_vector(63 downto 0))
190 return std_ulogic_vector is
191 variable msr_out: std_ulogic_vector(63 downto 0);
194 -- Defined MSR bits are classified as either full func-
195 -- tion or partial function. Full function MSR bits are
196 -- saved in SRR1 or HSRR1 when an interrupt other
197 -- than a System Call Vectored interrupt occurs and
198 -- restored by rfscv, rfid, or hrfid, while partial func-
199 -- tion MSR bits are not saved or restored.
200 -- Full function MSR bits lie in the range 0:32, 37:41, and
201 -- 48:63, and partial function MSR bits lie in the range
202 -- 33:36 and 42:47. (Note this is IBM bit numbering).
203 msr_out := (others => '0');
204 msr_out(63 downto 31) := msr(63 downto 31);
205 msr_out(26 downto 22) := msr(26 downto 22);
206 msr_out(15 downto 0) := msr(15 downto 0);
210 -- Tell vivado to keep the hierarchy for the random module so that the
211 -- net names in the xdc file match.
212 attribute keep_hierarchy : string;
213 attribute keep_hierarchy of random_0 : label is "yes";
217 rotator_0: entity work.rotator
221 shift => b_in(6 downto 0),
223 is_32bit => e_in.is_32bit,
224 right_shift => right_shift,
225 arith => e_in.is_signed,
226 clear_left => rot_clear_left,
227 clear_right => rot_clear_right,
228 sign_ext_rs => rot_sign_ext,
229 result => rotator_result,
230 carry_out => rotator_carry
233 logical_0: entity work.logical
237 op => e_in.insn_type,
238 invert_in => e_in.invert_a,
239 invert_out => e_in.invert_out,
240 result => logical_result,
241 datalen => e_in.data_len
244 countzero_0: entity work.zero_counter
248 count_right => e_in.insn(10),
249 is_32bit => e_in.is_32bit,
250 result => countzero_result
253 multiply_0: entity work.multiply
256 m_in => x_to_multiply,
257 m_out => multiply_to_x
260 divider_0: entity work.divider
264 d_in => x_to_divider,
265 d_out => divider_to_x
268 random_0: entity work.random
276 dbg_msr_out <= ctrl.msr;
277 log_rd_addr <= r.log_addr_spr;
279 a_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data1 = '1' else e_in.read_data1;
280 b_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data2 = '1' else e_in.read_data2;
281 c_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data3 = '1' else e_in.read_data3;
283 busy_out <= l_in.busy or r.busy or fp_in.busy;
284 valid_in <= e_in.valid and not busy_out;
286 terminate_out <= r.terminate;
288 execute1_0: process(clk)
290 if rising_edge(clk) then
293 ctrl.tb <= (others => '0');
294 ctrl.dec <= (others => '0');
295 ctrl.msr <= (MSR_SF => '1', MSR_LE => '1', others => '0');
296 ctrl.irq_state <= WRITE_SRR0;
300 assert not (r.lr_update = '1' and valid_in = '1')
301 report "LR update collision with valid in EX1"
303 if r.lr_update = '1' then
304 report "LR update to " & to_hstring(r.next_lr);
310 execute1_1: process(all)
311 variable v : reg_type;
312 variable a_inv : std_ulogic_vector(63 downto 0);
313 variable result : std_ulogic_vector(63 downto 0);
314 variable newcrf : std_ulogic_vector(3 downto 0);
315 variable sum_with_carry : std_ulogic_vector(64 downto 0);
316 variable result_en : std_ulogic;
317 variable crnum : crnum_t;
318 variable crbit : integer range 0 to 31;
319 variable scrnum : crnum_t;
320 variable lo, hi : integer;
321 variable sh, mb, me : std_ulogic_vector(5 downto 0);
322 variable sh32, mb32, me32 : std_ulogic_vector(4 downto 0);
323 variable bo, bi : std_ulogic_vector(4 downto 0);
324 variable bf, bfa : std_ulogic_vector(2 downto 0);
325 variable cr_op : std_ulogic_vector(9 downto 0);
326 variable cr_operands : std_ulogic_vector(1 downto 0);
327 variable bt, ba, bb : std_ulogic_vector(4 downto 0);
328 variable btnum, banum, bbnum : integer range 0 to 31;
329 variable crresult : std_ulogic;
330 variable l : std_ulogic;
331 variable next_nia : std_ulogic_vector(63 downto 0);
332 variable carry_32, carry_64 : std_ulogic;
333 variable sign1, sign2 : std_ulogic;
334 variable abs1, abs2 : signed(63 downto 0);
335 variable overflow : std_ulogic;
336 variable zerohi, zerolo : std_ulogic;
337 variable msb_a, msb_b : std_ulogic;
338 variable a_lt : std_ulogic;
339 variable lv : Execute1ToLoadstore1Type;
340 variable irq_valid : std_ulogic;
341 variable exception : std_ulogic;
342 variable exception_nextpc : std_ulogic;
343 variable trapval : std_ulogic_vector(4 downto 0);
344 variable illegal : std_ulogic;
345 variable is_branch : std_ulogic;
346 variable taken_branch : std_ulogic;
347 variable abs_branch : std_ulogic;
348 variable spr_val : std_ulogic_vector(63 downto 0);
349 variable addend : std_ulogic_vector(127 downto 0);
350 variable do_trace : std_ulogic;
351 variable f : Execute1ToFetch1Type;
352 variable fv : Execute1ToFPUType;
354 result := (others => '0');
355 sum_with_carry := (others => '0');
357 newcrf := (others => '0');
363 v.e := Execute1ToWritebackInit;
368 v.br_offset := (others => '0');
369 v.redir_mode := ctrl.msr(MSR_IR) & not ctrl.msr(MSR_PR) &
370 not ctrl.msr(MSR_LE) & not ctrl.msr(MSR_SF);
372 lv := Execute1ToLoadstore1Init;
373 fv := Execute1ToFPUInit;
375 -- XER forwarding. To avoid having to track XER hazards, we use
376 -- the previously latched value. Since the XER common bits
377 -- (SO, OV[32] and CA[32]) are only modified by instructions that are
378 -- handled here, we can just forward the result being sent to
380 if r.e.write_xerc_enable = '1' then
381 v.e.xerc := r.e.xerc;
383 v.e.xerc := e_in.xerc;
388 if EX1_BYPASS and e_in.bypass_cr = '1' and r.e.write_cr_enable = '1' then
390 if r.e.write_cr_mask(i) = '1' then
391 cr_in(i * 4 + 3 downto i * 4) <= r.e.write_cr_data(i * 4 + 3 downto i * 4);
397 v.mul_in_progress := '0';
398 v.div_in_progress := '0';
399 v.cntz_in_progress := '0';
403 if e_in.invert_a = '0' then
408 sum_with_carry := ppc_adde(a_inv, b_in,
409 decode_input_carry(e_in.input_carry, v.e.xerc));
411 -- signals to multiply and divide units
414 if e_in.is_signed = '1' then
415 if e_in.is_32bit = '1' then
423 -- take absolute values
425 abs1 := signed(a_in);
427 abs1 := - signed(a_in);
430 abs2 := signed(b_in);
432 abs2 := - signed(b_in);
435 x_to_multiply <= MultiplyInputInit;
436 x_to_multiply.is_32bit <= e_in.is_32bit;
438 x_to_divider <= Execute1ToDividerInit;
439 x_to_divider.is_signed <= e_in.is_signed;
440 x_to_divider.is_32bit <= e_in.is_32bit;
441 if e_in.insn_type = OP_MOD then
442 x_to_divider.is_modulus <= '1';
445 addend := (others => '0');
446 if e_in.insn(26) = '0' then
447 -- integer multiply-add, major op 4 (if it is a multiply)
448 addend(63 downto 0) := c_in;
449 if e_in.is_signed = '1' then
450 addend(127 downto 64) := (others => c_in(63));
453 if (sign1 xor sign2) = '1' then
454 addend := not addend;
457 x_to_multiply.not_result <= sign1 xor sign2;
458 x_to_multiply.addend <= addend;
459 x_to_divider.neg_result <= sign1 xor (sign2 and not x_to_divider.is_modulus);
460 if e_in.is_32bit = '0' then
462 x_to_multiply.data1 <= std_ulogic_vector(abs1);
463 x_to_multiply.data2 <= std_ulogic_vector(abs2);
464 if e_in.insn_type = OP_DIVE then
465 x_to_divider.is_extended <= '1';
467 x_to_divider.dividend <= std_ulogic_vector(abs1);
468 x_to_divider.divisor <= std_ulogic_vector(abs2);
471 x_to_multiply.data1 <= x"00000000" & std_ulogic_vector(abs1(31 downto 0));
472 x_to_multiply.data2 <= x"00000000" & std_ulogic_vector(abs2(31 downto 0));
473 x_to_divider.is_extended <= '0';
474 if e_in.insn_type = OP_DIVE then -- extended forms
475 x_to_divider.dividend <= std_ulogic_vector(abs1(31 downto 0)) & x"00000000";
477 x_to_divider.dividend <= x"00000000" & std_ulogic_vector(abs1(31 downto 0));
479 x_to_divider.divisor <= x"00000000" & std_ulogic_vector(abs2(31 downto 0));
483 -- FIXME: run at 512MHz not core freq
484 ctrl_tmp.tb <= std_ulogic_vector(unsigned(ctrl.tb) + 1);
485 ctrl_tmp.dec <= std_ulogic_vector(unsigned(ctrl.dec) - 1);
488 if ctrl.msr(MSR_EE) = '1' then
489 if ctrl.dec(63) = '1' then
491 report "IRQ valid: DEC";
493 elsif ext_irq_in = '1' then
495 report "IRQ valid: External";
504 -- Next insn adder used in a couple of places
505 next_nia := std_ulogic_vector(unsigned(e_in.nia) + 4);
507 -- rotator control signals
508 right_shift <= '1' when e_in.insn_type = OP_SHR else '0';
509 rot_clear_left <= '1' when e_in.insn_type = OP_RLC or e_in.insn_type = OP_RLCL else '0';
510 rot_clear_right <= '1' when e_in.insn_type = OP_RLC or e_in.insn_type = OP_RLCR else '0';
511 rot_sign_ext <= '1' when e_in.insn_type = OP_EXTSWSLI else '0';
513 ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
514 ctrl_tmp.irq_state <= WRITE_SRR0;
517 exception_nextpc := '0';
518 v.e.exc_write_enable := '0';
519 v.e.exc_write_reg := fast_spr_num(SPR_SRR0);
520 if valid_in = '1' then
521 v.e.exc_write_data := e_in.nia;
522 v.last_nia := e_in.nia;
524 v.e.exc_write_data := r.last_nia;
527 v.e.mode_32bit := not ctrl.msr(MSR_SF);
529 do_trace := valid_in and ctrl.msr(MSR_SE);
530 if valid_in = '1' then
531 v.prev_op := e_in.insn_type;
534 if ctrl.irq_state = WRITE_SRR1 then
535 v.e.exc_write_reg := fast_spr_num(SPR_SRR1);
536 v.e.exc_write_data := ctrl.srr1;
537 v.e.exc_write_enable := '1';
538 ctrl_tmp.msr(MSR_SF) <= '1';
539 ctrl_tmp.msr(MSR_EE) <= '0';
540 ctrl_tmp.msr(MSR_PR) <= '0';
541 ctrl_tmp.msr(MSR_SE) <= '0';
542 ctrl_tmp.msr(MSR_BE) <= '0';
543 ctrl_tmp.msr(MSR_FP) <= '0';
544 ctrl_tmp.msr(MSR_FE0) <= '0';
545 ctrl_tmp.msr(MSR_FE1) <= '0';
546 ctrl_tmp.msr(MSR_IR) <= '0';
547 ctrl_tmp.msr(MSR_DR) <= '0';
548 ctrl_tmp.msr(MSR_RI) <= '0';
549 ctrl_tmp.msr(MSR_LE) <= '1';
552 v.fp_exception_next := '0';
553 report "Writing SRR1: " & to_hstring(ctrl.srr1);
555 elsif valid_in = '1' and e_in.second = '0' and
556 ((HAS_FPU and r.fp_exception_next = '1') or r.trace_next = '1') then
557 if HAS_FPU and r.fp_exception_next = '1' then
558 -- This is used for FP-type program interrupts that
559 -- become pending due to MSR[FE0,FE1] changing from 00 to non-zero.
561 ctrl_tmp.srr1(63 - 43) <= '1';
562 ctrl_tmp.srr1(63 - 47) <= '1';
564 -- Generate a trace interrupt rather than executing the next instruction
565 -- or taking any asynchronous interrupt
567 ctrl_tmp.srr1(63 - 33) <= '1';
568 if r.prev_op = OP_LOAD or r.prev_op = OP_ICBI or r.prev_op = OP_ICBT or
569 r.prev_op = OP_DCBT or r.prev_op = OP_DCBST or r.prev_op = OP_DCBF then
570 ctrl_tmp.srr1(63 - 35) <= '1';
571 elsif r.prev_op = OP_STORE or r.prev_op = OP_DCBZ or r.prev_op = OP_DCBTST then
572 ctrl_tmp.srr1(63 - 36) <= '1';
577 elsif irq_valid = '1' and valid_in = '1' and e_in.second = '0' then
578 -- we need two cycles to write srr0 and 1
579 -- will need more when we have to write HEIR
580 -- Don't deliver the interrupt until we have a valid instruction
581 -- coming in, so we have a valid NIA to put in SRR0.
584 elsif valid_in = '1' and ctrl.msr(MSR_PR) = '1' and
585 instr_is_privileged(e_in.insn_type, e_in.insn) then
586 -- generate a program interrupt
589 -- set bit 45 to indicate privileged instruction type interrupt
590 ctrl_tmp.srr1(63 - 45) <= '1';
591 report "privileged instruction";
593 elsif not HAS_FPU and valid_in = '1' and e_in.fac = FPU then
594 -- make lfd/stfd/lfs/stfs etc. illegal in no-FPU implementations
597 elsif HAS_FPU and valid_in = '1' and ctrl.msr(MSR_FP) = '0' and e_in.fac = FPU then
598 -- generate a floating-point unavailable interrupt
601 report "FP unavailable interrupt";
603 elsif valid_in = '1' and e_in.unit = ALU then
605 report "execute nia " & to_hstring(e_in.nia);
608 v.e.write_reg := e_in.write_reg;
609 v.slow_op_insn := e_in.insn_type;
610 v.slow_op_dest := gspr_to_gpr(e_in.write_reg);
611 v.slow_op_rc := e_in.rc;
612 v.slow_op_oe := e_in.oe;
613 v.slow_op_xerc := v.e.xerc;
615 case_0: case e_in.insn_type is
618 -- we need two cycles to write srr0 and 1
619 -- will need more when we have to write HEIR
622 -- check bit 1 of the instruction is 1 so we know this is sc;
623 -- 0 would mean scv, so generate an illegal instruction interrupt
624 -- we need two cycles to write srr0 and 1
625 if e_in.insn(1) = '1' then
627 exception_nextpc := '1';
634 -- check bits 1-10 of the instruction to make sure it's attn
635 -- if not then it is illegal
636 if e_in.insn(10 downto 1) = "0100000000" then
642 when OP_NOP | OP_DCBF | OP_DCBST | OP_DCBT | OP_DCBTST | OP_ICBT =>
644 when OP_ADD | OP_CMP | OP_TRAP =>
645 result := sum_with_carry(63 downto 0);
646 carry_32 := result(32) xor a_inv(32) xor b_in(32);
647 carry_64 := sum_with_carry(64);
648 if e_in.insn_type = OP_ADD then
649 if e_in.output_carry = '1' then
650 if e_in.input_carry /= OV then
651 set_carry(v.e, carry_32, carry_64);
653 v.e.xerc.ov := carry_64;
654 v.e.xerc.ov32 := carry_32;
655 v.e.write_xerc_enable := '1';
658 if e_in.oe = '1' then
660 calc_ov(a_inv(63), b_in(63), carry_64, sum_with_carry(63)),
661 calc_ov(a_inv(31), b_in(31), carry_32, sum_with_carry(31)));
665 -- trap, CMP and CMPL instructions
666 -- Note, we have done RB - RA, not RA - RB
667 if e_in.insn_type = OP_CMP then
668 l := insn_l(e_in.insn);
670 l := not e_in.is_32bit;
672 zerolo := not (or (a_in(31 downto 0) xor b_in(31 downto 0)));
673 zerohi := not (or (a_in(63 downto 32) xor b_in(63 downto 32)));
674 if zerolo = '1' and (l = '0' or zerohi = '1') then
687 if msb_a /= msb_b then
688 -- Subtraction might overflow, but
689 -- comparison is clear from MSB difference.
690 -- for signed, 0 is greater; for unsigned, 1 is greater
691 trapval := msb_a & msb_b & '0' & msb_b & msb_a;
693 -- Subtraction cannot overflow since MSBs are equal.
694 -- carry = 1 indicates RA is smaller (signed or unsigned)
695 a_lt := (not l and carry_32) or (l and carry_64);
696 trapval := a_lt & not a_lt & '0' & a_lt & not a_lt;
699 if e_in.insn_type = OP_CMP then
700 if e_in.is_signed = '1' then
701 newcrf := trapval(4 downto 2) & v.e.xerc.so;
703 newcrf := trapval(1 downto 0) & trapval(2) & v.e.xerc.so;
705 bf := insn_bf(e_in.insn);
706 crnum := to_integer(unsigned(bf));
707 v.e.write_cr_enable := '1';
708 v.e.write_cr_mask := num_to_fxm(crnum);
712 v.e.write_cr_data(hi downto lo) := newcrf;
715 -- trap instructions (tw, twi, td, tdi)
717 -- set bit 46 to say trap occurred
718 ctrl_tmp.srr1(63 - 46) <= '1';
719 if or (trapval and insn_to(e_in.insn)) = '1' then
720 -- generate trap-type program interrupt
727 result := (others => '0');
728 for i in 0 to 14 loop
731 if (a_in(hi) xor b_in(hi) xor sum_with_carry(hi)) = '0' then
732 result(lo + 3 downto lo) := "0110";
735 if sum_with_carry(64) = '0' then
736 result(63 downto 60) := "0110";
740 newcrf := ppc_cmprb(a_in, b_in, insn_l(e_in.insn));
741 bf := insn_bf(e_in.insn);
742 crnum := to_integer(unsigned(bf));
743 v.e.write_cr_enable := '1';
744 v.e.write_cr_mask := num_to_fxm(crnum);
745 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
746 newcrf & newcrf & newcrf & newcrf;
748 newcrf := ppc_cmpeqb(a_in, b_in);
749 bf := insn_bf(e_in.insn);
750 crnum := to_integer(unsigned(bf));
751 v.e.write_cr_enable := '1';
752 v.e.write_cr_mask := num_to_fxm(crnum);
753 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
754 newcrf & newcrf & newcrf & newcrf;
755 when OP_AND | OP_OR | OP_XOR | OP_POPCNT | OP_PRTY | OP_CMPB | OP_EXTS |
757 result := logical_result;
762 abs_branch := insn_aa(e_in.insn);
763 if ctrl.msr(MSR_BE) = '1' then
768 bo := insn_bo(e_in.insn);
769 bi := insn_bi(e_in.insn);
770 if bo(4-2) = '0' then
771 result := std_ulogic_vector(unsigned(a_in) - 1);
773 v.e.write_reg := fast_spr_num(SPR_CTR);
776 taken_branch := ppc_bc_taken(bo, bi, cr_in, a_in);
777 abs_branch := insn_aa(e_in.insn);
778 if ctrl.msr(MSR_BE) = '1' then
783 -- read_data2 is target register (CTR, LR or TAR)
784 bo := insn_bo(e_in.insn);
785 bi := insn_bi(e_in.insn);
786 if bo(4-2) = '0' and e_in.insn(10) = '0' then
787 result := std_ulogic_vector(unsigned(a_in) - 1);
789 v.e.write_reg := fast_spr_num(SPR_CTR);
792 taken_branch := ppc_bc_taken(bo, bi, cr_in, a_in);
794 if ctrl.msr(MSR_BE) = '1' then
799 v.redir_mode := (a_in(MSR_IR) or a_in(MSR_PR)) & not a_in(MSR_PR) &
800 not a_in(MSR_LE) & not a_in(MSR_SF);
801 -- Can't use msr_copy here because the partial function MSR
802 -- bits should be left unchanged, not zeroed.
803 ctrl_tmp.msr(63 downto 31) <= a_in(63 downto 31);
804 ctrl_tmp.msr(26 downto 22) <= a_in(26 downto 22);
805 ctrl_tmp.msr(15 downto 0) <= a_in(15 downto 0);
806 if a_in(MSR_PR) = '1' then
807 ctrl_tmp.msr(MSR_EE) <= '1';
808 ctrl_tmp.msr(MSR_IR) <= '1';
809 ctrl_tmp.msr(MSR_DR) <= '1';
811 -- mark this as a branch so CFAR gets updated
816 v.fp_exception_next := fp_in.exception and
817 (a_in(MSR_FE0) or a_in(MSR_FE1));
823 v.cntz_in_progress := '1';
826 crbit := to_integer(unsigned(insn_bc(e_in.insn)));
827 if cr_in(31-crbit) = '1' then
834 cr_op := insn_cr(e_in.insn);
835 report "CR OP " & to_hstring(cr_op);
836 if cr_op(0) = '0' then -- MCRF
837 bf := insn_bf(e_in.insn);
838 bfa := insn_bfa(e_in.insn);
839 v.e.write_cr_enable := '1';
840 crnum := to_integer(unsigned(bf));
841 scrnum := to_integer(unsigned(bfa));
842 v.e.write_cr_mask := num_to_fxm(crnum);
847 newcrf := cr_in(hi downto lo);
853 v.e.write_cr_data(hi downto lo) := newcrf;
856 v.e.write_cr_enable := '1';
857 bt := insn_bt(e_in.insn);
858 ba := insn_ba(e_in.insn);
859 bb := insn_bb(e_in.insn);
860 btnum := 31 - to_integer(unsigned(bt));
861 banum := 31 - to_integer(unsigned(ba));
862 bbnum := 31 - to_integer(unsigned(bb));
863 -- Bits 5-8 of cr_op give the truth table of the requested
865 cr_operands := cr_in(banum) & cr_in(bbnum);
866 crresult := cr_op(5 + to_integer(unsigned(cr_operands)));
867 v.e.write_cr_mask := num_to_fxm((31-btnum) / 4);
868 for i in 0 to 31 loop
870 v.e.write_cr_data(i) := crresult;
872 v.e.write_cr_data(i) := cr_in(i);
877 newcrf := v.e.xerc.ov & v.e.xerc.ca & v.e.xerc.ov32 & v.e.xerc.ca32;
878 bf := insn_bf(e_in.insn);
879 crnum := to_integer(unsigned(bf));
880 v.e.write_cr_enable := '1';
881 v.e.write_cr_mask := num_to_fxm(crnum);
882 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
883 newcrf & newcrf & newcrf & newcrf;
885 if random_err = '0' then
886 case e_in.insn(17 downto 16) is
888 result := x"00000000" & random_cond(31 downto 0);
890 result := random_raw;
892 result := random_cond;
895 result := (others => '1');
902 report "MFSPR to SPR " & integer'image(decode_spr_num(e_in.insn)) &
903 "=" & to_hstring(a_in);
905 if is_fast_spr(e_in.read_reg1) then
907 if decode_spr_num(e_in.insn) = SPR_XER then
908 -- bits 0:31 and 35:43 are treated as reserved and return 0s when read using mfxer
909 result(63 downto 32) := (others => '0');
910 result(63-32) := v.e.xerc.so;
911 result(63-33) := v.e.xerc.ov;
912 result(63-34) := v.e.xerc.ca;
913 result(63-35 downto 63-43) := "000000000";
914 result(63-44) := v.e.xerc.ov32;
915 result(63-45) := v.e.xerc.ca32;
919 case decode_spr_num(e_in.insn) is
923 spr_val(63 downto 32) := (others => '0');
924 spr_val(31 downto 0) := ctrl.tb(63 downto 32);
928 spr_val := ctrl.cfar;
930 spr_val(63 downto 32) := (others => '0');
931 spr_val(31 downto 0) := PVR_MICROWATT;
932 when 724 => -- LOG_ADDR SPR
933 spr_val := log_wr_addr & r.log_addr_spr;
934 when 725 => -- LOG_DATA SPR
935 spr_val := log_rd_data;
936 v.log_addr_spr := std_ulogic_vector(unsigned(r.log_addr_spr) + 1);
938 -- mfspr from unimplemented SPRs should be a nop in
939 -- supervisor mode and a program interrupt for user mode
940 if ctrl.msr(MSR_PR) = '1' then
947 if e_in.insn(20) = '0' then
949 result := x"00000000" & cr_in;
952 crnum := fxm_to_num(insn_fxm(e_in.insn));
953 result := (others => '0');
958 result(hi downto lo) := cr_in(hi downto lo);
964 v.e.write_cr_enable := '1';
965 if e_in.insn(20) = '0' then
967 v.e.write_cr_mask := insn_fxm(e_in.insn);
969 -- mtocrf: We require one hot priority encoding here
970 crnum := fxm_to_num(insn_fxm(e_in.insn));
971 v.e.write_cr_mask := num_to_fxm(crnum);
973 v.e.write_cr_data := c_in(31 downto 0);
975 if e_in.insn(16) = '1' then
976 -- just update EE and RI
977 ctrl_tmp.msr(MSR_EE) <= c_in(MSR_EE);
978 ctrl_tmp.msr(MSR_RI) <= c_in(MSR_RI);
980 -- Architecture says to leave out bits 3 (HV), 51 (ME)
981 -- and 63 (LE) (IBM bit numbering)
982 if e_in.is_32bit = '0' then
983 ctrl_tmp.msr(63 downto 61) <= c_in(63 downto 61);
984 ctrl_tmp.msr(59 downto 32) <= c_in(59 downto 32);
986 ctrl_tmp.msr(31 downto 13) <= c_in(31 downto 13);
987 ctrl_tmp.msr(11 downto 1) <= c_in(11 downto 1);
988 if c_in(MSR_PR) = '1' then
989 ctrl_tmp.msr(MSR_EE) <= '1';
990 ctrl_tmp.msr(MSR_IR) <= '1';
991 ctrl_tmp.msr(MSR_DR) <= '1';
994 v.fp_exception_next := fp_in.exception and
995 (c_in(MSR_FE0) or c_in(MSR_FE1));
999 report "MTSPR to SPR " & integer'image(decode_spr_num(e_in.insn)) &
1000 "=" & to_hstring(c_in);
1001 if is_fast_spr(e_in.write_reg) then
1004 if decode_spr_num(e_in.insn) = SPR_XER then
1005 v.e.xerc.so := c_in(63-32);
1006 v.e.xerc.ov := c_in(63-33);
1007 v.e.xerc.ca := c_in(63-34);
1008 v.e.xerc.ov32 := c_in(63-44);
1009 v.e.xerc.ca32 := c_in(63-45);
1010 v.e.write_xerc_enable := '1';
1014 case decode_spr_num(e_in.insn) is
1016 ctrl_tmp.dec <= c_in;
1017 when 724 => -- LOG_ADDR SPR
1018 v.log_addr_spr := c_in(31 downto 0);
1020 -- mtspr to unimplemented SPRs should be a nop in
1021 -- supervisor mode and a program interrupt for user mode
1022 if ctrl.msr(MSR_PR) = '1' then
1027 when OP_RLC | OP_RLCL | OP_RLCR | OP_SHL | OP_SHR | OP_EXTSWSLI =>
1028 result := rotator_result;
1029 if e_in.output_carry = '1' then
1030 set_carry(v.e, rotator_carry, rotator_carry);
1034 bfa := insn_bfa(e_in.insn);
1035 crbit := to_integer(unsigned(bfa)) * 4;
1036 result := (others => '0');
1037 if cr_in(31 - crbit) = '1' then
1038 result := (others => '1');
1039 elsif cr_in(30 - crbit) = '1' then
1045 v.br_offset := std_ulogic_vector(to_unsigned(4, 64));
1048 icache_inval <= '1';
1050 when OP_MUL_L64 | OP_MUL_H64 | OP_MUL_H32 =>
1052 v.mul_in_progress := '1';
1054 x_to_multiply.valid <= '1';
1056 when OP_DIV | OP_DIVE | OP_MOD =>
1058 v.div_in_progress := '1';
1060 x_to_divider.valid <= '1';
1067 v.e.rc := e_in.rc and valid_in;
1069 -- Mispredicted branches cause a redirect
1070 if is_branch = '1' then
1071 if taken_branch = '1' then
1072 ctrl_tmp.cfar <= e_in.nia;
1074 if e_in.br_pred = '0' then
1075 v.br_offset := b_in;
1076 v.abs_br := abs_branch;
1078 v.br_offset := std_ulogic_vector(to_unsigned(4, 64));
1080 if taken_branch /= e_in.br_pred then
1085 -- Update LR on the next cycle after a branch link
1086 -- If we're not writing back anything else, we can write back LR
1087 -- this cycle, otherwise we take an extra cycle. We use the
1088 -- exc_write path since next_nia is written through that path
1090 if e_in.lr = '1' then
1091 if result_en = '0' then
1092 v.e.exc_write_enable := '1';
1093 v.e.exc_write_data := next_nia;
1094 v.e.exc_write_reg := fast_spr_num(SPR_LR);
1097 v.next_lr := next_nia;
1099 report "Delayed LR update to " & to_hstring(next_nia);
1104 elsif valid_in = '1' then
1105 -- instruction for other units, i.e. LDST
1106 if e_in.unit = LDST then
1108 elsif e_in.unit = NONE then
1110 elsif HAS_FPU and e_in.unit = FPU then
1113 -- Handling an ITLB miss doesn't count as having executed an instruction
1114 if e_in.insn_type = OP_FETCH_FAILED then
1119 -- The following cases all occur when r.busy = 1 and therefore
1120 -- valid_in = 0. Hence they don't happen in the same cycle as any of
1121 -- the cases above which depend on valid_in = 1.
1123 if r.redirect = '1' then
1126 if r.lr_update = '1' then
1127 v.e.exc_write_enable := '1';
1128 v.e.exc_write_data := r.next_lr;
1129 v.e.exc_write_reg := fast_spr_num(SPR_LR);
1131 -- Keep r.e.write_data unchanged next cycle in case it is needed
1132 -- for a forwarded result (e.g. for CTR).
1133 result := r.e.write_data;
1134 elsif r.cntz_in_progress = '1' then
1135 -- cnt[lt]z always takes two cycles
1136 result := countzero_result;
1138 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1139 v.e.rc := r.slow_op_rc;
1140 v.e.xerc := r.slow_op_xerc;
1142 elsif r.mul_in_progress = '1' or r.div_in_progress = '1' then
1143 if (r.mul_in_progress = '1' and multiply_to_x.valid = '1') or
1144 (r.div_in_progress = '1' and divider_to_x.valid = '1') then
1145 if r.mul_in_progress = '1' then
1147 case r.slow_op_insn is
1149 result := multiply_to_x.result(63 downto 32) &
1150 multiply_to_x.result(63 downto 32);
1152 result := multiply_to_x.result(127 downto 64);
1155 result := multiply_to_x.result(63 downto 0);
1158 result := divider_to_x.write_reg_data;
1159 overflow := divider_to_x.overflow;
1161 if r.mul_in_progress = '1' and r.slow_op_oe = '1' then
1162 -- have to wait until next cycle for overflow indication
1163 v.mul_finish := '1';
1167 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1168 v.e.rc := r.slow_op_rc;
1169 v.e.xerc := r.slow_op_xerc;
1170 v.e.write_xerc_enable := r.slow_op_oe;
1171 -- We must test oe because the RC update code in writeback
1172 -- will use the xerc value to set CR0:SO so we must not clobber
1173 -- xerc if OE wasn't set.
1174 if r.slow_op_oe = '1' then
1175 v.e.xerc.ov := overflow;
1176 v.e.xerc.ov32 := overflow;
1177 v.e.xerc.so := r.slow_op_xerc.so or overflow;
1183 v.mul_in_progress := r.mul_in_progress;
1184 v.div_in_progress := r.div_in_progress;
1186 elsif r.mul_finish = '1' then
1187 result := r.e.write_data;
1189 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1190 v.e.rc := r.slow_op_rc;
1191 v.e.xerc := r.slow_op_xerc;
1192 v.e.write_xerc_enable := r.slow_op_oe;
1193 v.e.xerc.ov := multiply_to_x.overflow;
1194 v.e.xerc.ov32 := multiply_to_x.overflow;
1195 v.e.xerc.so := r.slow_op_xerc.so or multiply_to_x.overflow;
1199 -- Generate FP-type program interrupt. fp_in.interrupt will only
1200 -- be set during the execution of a FP instruction.
1201 -- The case where MSR[FE0,FE1] goes from zero to non-zero is
1202 -- handled above by mtmsrd and rfid setting v.fp_exception_next.
1203 if HAS_FPU and fp_in.interrupt = '1' then
1204 v.vector := 16#700#;
1205 ctrl_tmp.srr1(63 - 43) <= '1';
1209 if illegal = '1' or (HAS_FPU and fp_in.illegal = '1') then
1211 v.vector := 16#700#;
1212 -- Since we aren't doing Hypervisor emulation assist (0xe40) we
1213 -- set bit 44 to indicate we have an illegal
1214 ctrl_tmp.srr1(63 - 44) <= '1';
1217 if exception = '1' then
1218 v.e.exc_write_enable := '1';
1219 if exception_nextpc = '1' then
1220 v.e.exc_write_data := next_nia;
1224 if do_trace = '1' then
1225 v.trace_next := '1';
1228 v.e.write_data := result;
1229 v.e.write_enable := result_en and not exception;
1231 -- generate DSI or DSegI for load/store exceptions
1232 -- or ISI or ISegI for instruction fetch exceptions
1233 if l_in.exception = '1' then
1234 if l_in.alignment = '1' then
1235 v.vector := 16#600#;
1236 elsif l_in.instr_fault = '0' then
1237 if l_in.segment_fault = '0' then
1238 v.vector := 16#300#;
1240 v.vector := 16#380#;
1243 if l_in.segment_fault = '0' then
1244 ctrl_tmp.srr1(63 - 33) <= l_in.invalid;
1245 ctrl_tmp.srr1(63 - 35) <= l_in.perm_error; -- noexec fault
1246 ctrl_tmp.srr1(63 - 44) <= l_in.badtree;
1247 ctrl_tmp.srr1(63 - 45) <= l_in.rc_error;
1248 v.vector := 16#400#;
1250 v.vector := 16#480#;
1253 v.e.exc_write_enable := '1';
1254 v.e.exc_write_reg := fast_spr_num(SPR_SRR0);
1255 report "ldst exception writing srr0=" & to_hstring(r.last_nia);
1258 if exception = '1' or l_in.exception = '1' then
1259 ctrl_tmp.irq_state <= WRITE_SRR1;
1264 if v.redirect = '1' then
1269 -- Outputs to fetch1
1270 f.redirect := r.redirect;
1271 if r.do_intr = '1' then
1272 f.redirect_nia := std_ulogic_vector(to_unsigned(r.vector, 64));
1275 -- XXX need an interrupt LE bit here, e.g. from LPCR
1276 f.big_endian := '0';
1277 f.mode_32bit := '0';
1279 if r.abs_br = '1' then
1280 f.redirect_nia := r.br_offset;
1282 f.redirect_nia := std_ulogic_vector(unsigned(r.last_nia) + unsigned(r.br_offset));
1284 -- send MSR[IR], ~MSR[PR], ~MSR[LE] and ~MSR[SF] up to fetch1
1285 f.virt_mode := r.redir_mode(3);
1286 f.priv_mode := r.redir_mode(2);
1287 f.big_endian := r.redir_mode(1);
1288 f.mode_32bit := r.redir_mode(0);
1291 -- Outputs to loadstore1 (async)
1292 lv.op := e_in.insn_type;
1297 lv.write_reg := e_in.write_reg;
1298 lv.length := e_in.data_len;
1299 lv.byte_reverse := e_in.byte_reverse xnor ctrl.msr(MSR_LE);
1300 lv.sign_extend := e_in.sign_extend;
1301 lv.update := e_in.update;
1302 lv.update_reg := gspr_to_gpr(e_in.read_reg1);
1303 lv.xerc := v.e.xerc;
1304 lv.reserve := e_in.reserve;
1306 lv.insn := e_in.insn;
1307 -- decode l*cix and st*cix instructions here
1308 if e_in.insn(31 downto 26) = "011111" and e_in.insn(10 downto 9) = "11" and
1309 e_in.insn(5 downto 1) = "10101" then
1312 lv.virt_mode := ctrl.msr(MSR_DR);
1313 lv.priv_mode := not ctrl.msr(MSR_PR);
1314 lv.mode_32bit := not ctrl.msr(MSR_SF);
1315 lv.is_32bit := e_in.is_32bit;
1316 lv.repeat := e_in.repeat;
1317 lv.second := e_in.second;
1320 fv.op := e_in.insn_type;
1322 fv.insn := e_in.insn;
1323 fv.single := e_in.is_32bit;
1324 fv.fe_mode := ctrl.msr(MSR_FE0) & ctrl.msr(MSR_FE1);
1328 fv.frt := e_in.write_reg;
1330 fv.out_cr := e_in.output_cr;
1340 flush_out <= f_out.redirect;
1342 exception_log <= exception;
1343 irq_valid_log <= irq_valid;
1346 e1_log: if LOG_LENGTH > 0 generate
1347 signal log_data : std_ulogic_vector(14 downto 0);
1349 ex1_log : process(clk)
1351 if rising_edge(clk) then
1352 log_data <= ctrl.msr(MSR_EE) & ctrl.msr(MSR_PR) &
1353 ctrl.msr(MSR_IR) & ctrl.msr(MSR_DR) &
1356 std_ulogic_vector(to_unsigned(irq_state_t'pos(ctrl.irq_state), 1)) &
1365 log_out <= log_data;
1367 end architecture behaviour;