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;
56 f : Execute1ToFetch1Type;
58 terminate: std_ulogic;
59 fp_exception_next : std_ulogic;
60 trace_next : std_ulogic;
61 prev_op : insn_type_t;
62 lr_update : std_ulogic;
63 next_lr : std_ulogic_vector(63 downto 0);
64 mul_in_progress : std_ulogic;
65 mul_finish : std_ulogic;
66 div_in_progress : std_ulogic;
67 cntz_in_progress : std_ulogic;
68 slow_op_insn : insn_type_t;
69 slow_op_dest : gpr_index_t;
70 slow_op_rc : std_ulogic;
71 slow_op_oe : std_ulogic;
72 slow_op_xerc : xer_common_t;
73 last_nia : std_ulogic_vector(63 downto 0);
74 log_addr_spr : std_ulogic_vector(31 downto 0);
76 constant reg_type_init : reg_type :=
77 (e => Execute1ToWritebackInit, f => Execute1ToFetch1Init,
78 busy => '0', lr_update => '0', terminate => '0',
79 fp_exception_next => '0', trace_next => '0', prev_op => OP_ILLEGAL,
80 mul_in_progress => '0', mul_finish => '0', div_in_progress => '0', cntz_in_progress => '0',
81 slow_op_insn => OP_ILLEGAL, slow_op_rc => '0', slow_op_oe => '0', slow_op_xerc => xerc_init,
82 next_lr => (others => '0'), last_nia => (others => '0'), others => (others => '0'));
84 signal r, rin : reg_type;
86 signal a_in, b_in, c_in : std_ulogic_vector(63 downto 0);
87 signal cr_in : std_ulogic_vector(31 downto 0);
89 signal valid_in : std_ulogic;
90 signal ctrl: ctrl_t := (irq_state => WRITE_SRR0, others => (others => '0'));
91 signal ctrl_tmp: ctrl_t := (irq_state => WRITE_SRR0, others => (others => '0'));
92 signal right_shift, rot_clear_left, rot_clear_right: std_ulogic;
93 signal rot_sign_ext: std_ulogic;
94 signal rotator_result: std_ulogic_vector(63 downto 0);
95 signal rotator_carry: std_ulogic;
96 signal logical_result: std_ulogic_vector(63 downto 0);
97 signal countzero_result: std_ulogic_vector(63 downto 0);
100 signal x_to_multiply: MultiplyInputType;
101 signal multiply_to_x: MultiplyOutputType;
104 signal x_to_divider: Execute1ToDividerType;
105 signal divider_to_x: DividerToExecute1Type;
107 -- random number generator signals
108 signal random_raw : std_ulogic_vector(63 downto 0);
109 signal random_cond : std_ulogic_vector(63 downto 0);
110 signal random_err : std_ulogic;
112 -- signals for logging
113 signal exception_log : std_ulogic;
114 signal irq_valid_log : std_ulogic;
116 type privilege_level is (USER, SUPER);
117 type op_privilege_array is array(insn_type_t) of privilege_level;
118 constant op_privilege: op_privilege_array := (
127 function instr_is_privileged(op: insn_type_t; insn: std_ulogic_vector(31 downto 0))
130 if op_privilege(op) = SUPER then
132 elsif op = OP_MFSPR or op = OP_MTSPR then
133 return insn(20) = '1';
139 procedure set_carry(e: inout Execute1ToWritebackType;
140 carry32 : in std_ulogic;
141 carry : in std_ulogic) is
143 e.xerc.ca32 := carry32;
145 e.write_xerc_enable := '1';
148 procedure set_ov(e: inout Execute1ToWritebackType;
150 ov32 : in std_ulogic) is
157 e.write_xerc_enable := '1';
160 function calc_ov(msb_a : std_ulogic; msb_b: std_ulogic;
161 ca: std_ulogic; msb_r: std_ulogic) return std_ulogic is
163 return (ca xor msb_r) and not (msb_a xor msb_b);
166 function decode_input_carry(ic : carry_in_t;
167 xerc : xer_common_t) return std_ulogic is
181 function msr_copy(msr: std_ulogic_vector(63 downto 0))
182 return std_ulogic_vector is
183 variable msr_out: std_ulogic_vector(63 downto 0);
186 -- Defined MSR bits are classified as either full func-
187 -- tion or partial function. Full function MSR bits are
188 -- saved in SRR1 or HSRR1 when an interrupt other
189 -- than a System Call Vectored interrupt occurs and
190 -- restored by rfscv, rfid, or hrfid, while partial func-
191 -- tion MSR bits are not saved or restored.
192 -- Full function MSR bits lie in the range 0:32, 37:41, and
193 -- 48:63, and partial function MSR bits lie in the range
194 -- 33:36 and 42:47. (Note this is IBM bit numbering).
195 msr_out := (others => '0');
196 msr_out(63 downto 31) := msr(63 downto 31);
197 msr_out(26 downto 22) := msr(26 downto 22);
198 msr_out(15 downto 0) := msr(15 downto 0);
202 -- Tell vivado to keep the hierarchy for the random module so that the
203 -- net names in the xdc file match.
204 attribute keep_hierarchy : string;
205 attribute keep_hierarchy of random_0 : label is "yes";
209 rotator_0: entity work.rotator
213 shift => b_in(6 downto 0),
215 is_32bit => e_in.is_32bit,
216 right_shift => right_shift,
217 arith => e_in.is_signed,
218 clear_left => rot_clear_left,
219 clear_right => rot_clear_right,
220 sign_ext_rs => rot_sign_ext,
221 result => rotator_result,
222 carry_out => rotator_carry
225 logical_0: entity work.logical
229 op => e_in.insn_type,
230 invert_in => e_in.invert_a,
231 invert_out => e_in.invert_out,
232 result => logical_result,
233 datalen => e_in.data_len
236 countzero_0: entity work.zero_counter
240 count_right => e_in.insn(10),
241 is_32bit => e_in.is_32bit,
242 result => countzero_result
245 multiply_0: entity work.multiply
248 m_in => x_to_multiply,
249 m_out => multiply_to_x
252 divider_0: entity work.divider
256 d_in => x_to_divider,
257 d_out => divider_to_x
260 random_0: entity work.random
268 dbg_msr_out <= ctrl.msr;
269 log_rd_addr <= r.log_addr_spr;
271 a_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data1 = '1' else e_in.read_data1;
272 b_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data2 = '1' else e_in.read_data2;
273 c_in <= r.e.write_data when EX1_BYPASS and e_in.bypass_data3 = '1' else e_in.read_data3;
275 busy_out <= l_in.busy or r.busy or fp_in.busy;
276 valid_in <= e_in.valid and not busy_out;
278 terminate_out <= r.terminate;
280 execute1_0: process(clk)
282 if rising_edge(clk) then
285 ctrl.tb <= (others => '0');
286 ctrl.dec <= (others => '0');
287 ctrl.msr <= (MSR_SF => '1', MSR_LE => '1', others => '0');
288 ctrl.irq_state <= WRITE_SRR0;
292 assert not (r.lr_update = '1' and valid_in = '1')
293 report "LR update collision with valid in EX1"
295 if r.lr_update = '1' then
296 report "LR update to " & to_hstring(r.next_lr);
302 execute1_1: process(all)
303 variable v : reg_type;
304 variable a_inv : std_ulogic_vector(63 downto 0);
305 variable result : std_ulogic_vector(63 downto 0);
306 variable newcrf : std_ulogic_vector(3 downto 0);
307 variable sum_with_carry : std_ulogic_vector(64 downto 0);
308 variable result_en : std_ulogic;
309 variable crnum : crnum_t;
310 variable crbit : integer range 0 to 31;
311 variable scrnum : crnum_t;
312 variable lo, hi : integer;
313 variable sh, mb, me : std_ulogic_vector(5 downto 0);
314 variable sh32, mb32, me32 : std_ulogic_vector(4 downto 0);
315 variable bo, bi : std_ulogic_vector(4 downto 0);
316 variable bf, bfa : std_ulogic_vector(2 downto 0);
317 variable cr_op : std_ulogic_vector(9 downto 0);
318 variable cr_operands : std_ulogic_vector(1 downto 0);
319 variable bt, ba, bb : std_ulogic_vector(4 downto 0);
320 variable btnum, banum, bbnum : integer range 0 to 31;
321 variable crresult : std_ulogic;
322 variable l : std_ulogic;
323 variable next_nia : std_ulogic_vector(63 downto 0);
324 variable carry_32, carry_64 : std_ulogic;
325 variable sign1, sign2 : std_ulogic;
326 variable abs1, abs2 : signed(63 downto 0);
327 variable overflow : std_ulogic;
328 variable zerohi, zerolo : std_ulogic;
329 variable msb_a, msb_b : std_ulogic;
330 variable a_lt : std_ulogic;
331 variable lv : Execute1ToLoadstore1Type;
332 variable irq_valid : std_ulogic;
333 variable exception : std_ulogic;
334 variable exception_nextpc : std_ulogic;
335 variable trapval : std_ulogic_vector(4 downto 0);
336 variable illegal : std_ulogic;
337 variable is_branch : std_ulogic;
338 variable taken_branch : std_ulogic;
339 variable abs_branch : std_ulogic;
340 variable spr_val : std_ulogic_vector(63 downto 0);
341 variable addend : std_ulogic_vector(127 downto 0);
342 variable do_trace : std_ulogic;
343 variable fv : Execute1ToFPUType;
345 result := (others => '0');
346 sum_with_carry := (others => '0');
348 newcrf := (others => '0');
354 v.e := Execute1ToWritebackInit;
355 lv := Execute1ToLoadstore1Init;
357 fv := Execute1ToFPUInit;
359 -- XER forwarding. To avoid having to track XER hazards, we
360 -- use the previously latched value.
362 -- If the XER was modified by a multiply or a divide, those are
363 -- single issue, we'll get the up to date value from decode2 from
364 -- the register file.
366 -- If it was modified by an instruction older than the previous
367 -- one in EX1, it will have also hit writeback and will be up
368 -- to date in decode2.
370 -- That leaves us with the case where it was updated by the previous
371 -- instruction in EX1. In that case, we can forward it back here.
373 -- This will break if we allow pipelining of multiply and divide,
374 -- but ideally, those should go via EX1 anyway and run as a state
375 -- machine from here.
377 -- One additional hazard to beware of is an XER:SO modifying instruction
378 -- in EX1 followed immediately by a store conditional. Due to our
379 -- writeback latency, the store will go down the LSU with the previous
380 -- XER value, thus the stcx. will set CR0:SO using an obsolete SO value.
382 -- We will need to handle that if we ever make stcx. not single issue
384 -- We always pass a valid XER value downto writeback even when
385 -- we aren't updating it, in order for XER:SO -> CR0:SO transfer
386 -- to work for RC instructions.
388 if r.e.write_xerc_enable = '1' then
389 v.e.xerc := r.e.xerc;
391 v.e.xerc := e_in.xerc;
396 if EX1_BYPASS and e_in.bypass_cr = '1' and r.e.write_cr_enable = '1' then
398 if r.e.write_cr_mask(i) = '1' then
399 cr_in(i * 4 + 3 downto i * 4) <= r.e.write_cr_data(i * 4 + 3 downto i * 4);
405 v.mul_in_progress := '0';
406 v.div_in_progress := '0';
407 v.cntz_in_progress := '0';
411 if e_in.invert_a = '0' then
416 sum_with_carry := ppc_adde(a_inv, b_in,
417 decode_input_carry(e_in.input_carry, v.e.xerc));
419 -- signals to multiply and divide units
422 if e_in.is_signed = '1' then
423 if e_in.is_32bit = '1' then
431 -- take absolute values
433 abs1 := signed(a_in);
435 abs1 := - signed(a_in);
438 abs2 := signed(b_in);
440 abs2 := - signed(b_in);
443 x_to_multiply <= MultiplyInputInit;
444 x_to_multiply.is_32bit <= e_in.is_32bit;
446 x_to_divider <= Execute1ToDividerInit;
447 x_to_divider.is_signed <= e_in.is_signed;
448 x_to_divider.is_32bit <= e_in.is_32bit;
449 if e_in.insn_type = OP_MOD then
450 x_to_divider.is_modulus <= '1';
453 addend := (others => '0');
454 if e_in.insn(26) = '0' then
455 -- integer multiply-add, major op 4 (if it is a multiply)
456 addend(63 downto 0) := c_in;
457 if e_in.is_signed = '1' then
458 addend(127 downto 64) := (others => c_in(63));
461 if (sign1 xor sign2) = '1' then
462 addend := not addend;
465 x_to_multiply.not_result <= sign1 xor sign2;
466 x_to_multiply.addend <= addend;
467 x_to_divider.neg_result <= sign1 xor (sign2 and not x_to_divider.is_modulus);
468 if e_in.is_32bit = '0' then
470 x_to_multiply.data1 <= std_ulogic_vector(abs1);
471 x_to_multiply.data2 <= std_ulogic_vector(abs2);
472 if e_in.insn_type = OP_DIVE then
473 x_to_divider.is_extended <= '1';
475 x_to_divider.dividend <= std_ulogic_vector(abs1);
476 x_to_divider.divisor <= std_ulogic_vector(abs2);
479 x_to_multiply.data1 <= x"00000000" & std_ulogic_vector(abs1(31 downto 0));
480 x_to_multiply.data2 <= x"00000000" & std_ulogic_vector(abs2(31 downto 0));
481 x_to_divider.is_extended <= '0';
482 if e_in.insn_type = OP_DIVE then -- extended forms
483 x_to_divider.dividend <= std_ulogic_vector(abs1(31 downto 0)) & x"00000000";
485 x_to_divider.dividend <= x"00000000" & std_ulogic_vector(abs1(31 downto 0));
487 x_to_divider.divisor <= x"00000000" & std_ulogic_vector(abs2(31 downto 0));
491 -- FIXME: run at 512MHz not core freq
492 ctrl_tmp.tb <= std_ulogic_vector(unsigned(ctrl.tb) + 1);
493 ctrl_tmp.dec <= std_ulogic_vector(unsigned(ctrl.dec) - 1);
496 if ctrl.msr(MSR_EE) = '1' then
497 if ctrl.dec(63) = '1' then
498 v.f.redirect_nia := std_logic_vector(to_unsigned(16#900#, 64));
499 report "IRQ valid: DEC";
501 elsif ext_irq_in = '1' then
502 v.f.redirect_nia := std_logic_vector(to_unsigned(16#500#, 64));
503 report "IRQ valid: External";
511 -- send MSR[IR], ~MSR[PR], ~MSR[LE] and ~MSR[SF] up to fetch1
512 v.f.virt_mode := ctrl.msr(MSR_IR);
513 v.f.priv_mode := not ctrl.msr(MSR_PR);
514 v.f.big_endian := not ctrl.msr(MSR_LE);
515 v.f.mode_32bit := not ctrl.msr(MSR_SF);
517 -- Next insn adder used in a couple of places
518 next_nia := std_ulogic_vector(unsigned(e_in.nia) + 4);
520 -- rotator control signals
521 right_shift <= '1' when e_in.insn_type = OP_SHR else '0';
522 rot_clear_left <= '1' when e_in.insn_type = OP_RLC or e_in.insn_type = OP_RLCL else '0';
523 rot_clear_right <= '1' when e_in.insn_type = OP_RLC or e_in.insn_type = OP_RLCR else '0';
524 rot_sign_ext <= '1' when e_in.insn_type = OP_EXTSWSLI else '0';
526 ctrl_tmp.srr1 <= msr_copy(ctrl.msr);
527 ctrl_tmp.irq_state <= WRITE_SRR0;
530 exception_nextpc := '0';
531 v.e.exc_write_enable := '0';
532 v.e.exc_write_reg := fast_spr_num(SPR_SRR0);
533 if valid_in = '1' then
534 v.e.exc_write_data := e_in.nia;
535 v.last_nia := e_in.nia;
537 v.e.exc_write_data := r.last_nia;
540 v.e.mode_32bit := not ctrl.msr(MSR_SF);
542 do_trace := valid_in and ctrl.msr(MSR_SE);
543 if valid_in = '1' then
544 v.prev_op := e_in.insn_type;
547 if ctrl.irq_state = WRITE_SRR1 then
548 v.e.exc_write_reg := fast_spr_num(SPR_SRR1);
549 v.e.exc_write_data := ctrl.srr1;
550 v.e.exc_write_enable := '1';
551 ctrl_tmp.msr(MSR_SF) <= '1';
552 ctrl_tmp.msr(MSR_EE) <= '0';
553 ctrl_tmp.msr(MSR_PR) <= '0';
554 ctrl_tmp.msr(MSR_SE) <= '0';
555 ctrl_tmp.msr(MSR_BE) <= '0';
556 ctrl_tmp.msr(MSR_FP) <= '0';
557 ctrl_tmp.msr(MSR_FE0) <= '0';
558 ctrl_tmp.msr(MSR_FE1) <= '0';
559 ctrl_tmp.msr(MSR_IR) <= '0';
560 ctrl_tmp.msr(MSR_DR) <= '0';
561 ctrl_tmp.msr(MSR_RI) <= '0';
562 ctrl_tmp.msr(MSR_LE) <= '1';
565 v.fp_exception_next := '0';
566 report "Writing SRR1: " & to_hstring(ctrl.srr1);
568 elsif valid_in = '1' and ((HAS_FPU and r.fp_exception_next = '1') or r.trace_next = '1') then
569 if HAS_FPU and r.fp_exception_next = '1' then
570 -- This is used for FP-type program interrupts that
571 -- become pending due to MSR[FE0,FE1] changing from 00 to non-zero.
572 v.f.redirect_nia := std_logic_vector(to_unsigned(16#700#, 64));
573 ctrl_tmp.srr1(63 - 43) <= '1';
574 ctrl_tmp.srr1(63 - 47) <= '1';
576 -- Generate a trace interrupt rather than executing the next instruction
577 -- or taking any asynchronous interrupt
578 v.f.redirect_nia := std_logic_vector(to_unsigned(16#d00#, 64));
579 ctrl_tmp.srr1(63 - 33) <= '1';
580 if r.prev_op = OP_LOAD or r.prev_op = OP_ICBI or r.prev_op = OP_ICBT or
581 r.prev_op = OP_DCBT or r.prev_op = OP_DCBST or r.prev_op = OP_DCBF then
582 ctrl_tmp.srr1(63 - 35) <= '1';
583 elsif r.prev_op = OP_STORE or r.prev_op = OP_DCBZ or r.prev_op = OP_DCBTST then
584 ctrl_tmp.srr1(63 - 36) <= '1';
589 elsif irq_valid = '1' and valid_in = '1' then
590 -- we need two cycles to write srr0 and 1
591 -- will need more when we have to write HEIR
592 -- Don't deliver the interrupt until we have a valid instruction
593 -- coming in, so we have a valid NIA to put in SRR0.
596 elsif valid_in = '1' and ctrl.msr(MSR_PR) = '1' and
597 instr_is_privileged(e_in.insn_type, e_in.insn) then
598 -- generate a program interrupt
600 v.f.redirect_nia := std_logic_vector(to_unsigned(16#700#, 64));
601 -- set bit 45 to indicate privileged instruction type interrupt
602 ctrl_tmp.srr1(63 - 45) <= '1';
603 report "privileged instruction";
605 elsif not HAS_FPU and valid_in = '1' and
606 (e_in.insn_type = OP_FPLOAD or e_in.insn_type = OP_FPSTORE) then
607 -- make lfd/stfd/lfs/stfs etc. illegal in no-FPU implementations
610 elsif HAS_FPU and valid_in = '1' and ctrl.msr(MSR_FP) = '0' and
611 (e_in.unit = FPU or e_in.insn_type = OP_FPLOAD or e_in.insn_type = OP_FPSTORE) then
612 -- generate a floating-point unavailable interrupt
614 v.f.redirect_nia := std_logic_vector(to_unsigned(16#800#, 64));
615 report "FP unavailable interrupt";
617 elsif valid_in = '1' and e_in.unit = ALU then
619 report "execute nia " & to_hstring(e_in.nia);
622 v.e.write_reg := e_in.write_reg;
623 v.slow_op_insn := e_in.insn_type;
624 v.slow_op_dest := gspr_to_gpr(e_in.write_reg);
625 v.slow_op_rc := e_in.rc;
626 v.slow_op_oe := e_in.oe;
627 v.slow_op_xerc := v.e.xerc;
629 case_0: case e_in.insn_type is
632 -- we need two cycles to write srr0 and 1
633 -- will need more when we have to write HEIR
636 -- check bit 1 of the instruction is 1 so we know this is sc;
637 -- 0 would mean scv, so generate an illegal instruction interrupt
638 -- we need two cycles to write srr0 and 1
639 if e_in.insn(1) = '1' then
641 exception_nextpc := '1';
642 v.f.redirect_nia := std_logic_vector(to_unsigned(16#C00#, 64));
648 -- check bits 1-10 of the instruction to make sure it's attn
649 -- if not then it is illegal
650 if e_in.insn(10 downto 1) = "0100000000" then
656 when OP_NOP | OP_DCBF | OP_DCBST | OP_DCBT | OP_DCBTST | OP_ICBT =>
658 when OP_ADD | OP_CMP | OP_TRAP =>
659 result := sum_with_carry(63 downto 0);
660 carry_32 := result(32) xor a_inv(32) xor b_in(32);
661 carry_64 := sum_with_carry(64);
662 if e_in.insn_type = OP_ADD then
663 if e_in.output_carry = '1' then
664 if e_in.input_carry /= OV then
665 set_carry(v.e, carry_32, carry_64);
667 v.e.xerc.ov := carry_64;
668 v.e.xerc.ov32 := carry_32;
669 v.e.write_xerc_enable := '1';
672 if e_in.oe = '1' then
674 calc_ov(a_inv(63), b_in(63), carry_64, sum_with_carry(63)),
675 calc_ov(a_inv(31), b_in(31), carry_32, sum_with_carry(31)));
679 -- trap, CMP and CMPL instructions
680 -- Note, we have done RB - RA, not RA - RB
681 if e_in.insn_type = OP_CMP then
682 l := insn_l(e_in.insn);
684 l := not e_in.is_32bit;
686 zerolo := not (or (a_in(31 downto 0) xor b_in(31 downto 0)));
687 zerohi := not (or (a_in(63 downto 32) xor b_in(63 downto 32)));
688 if zerolo = '1' and (l = '0' or zerohi = '1') then
701 if msb_a /= msb_b then
702 -- Subtraction might overflow, but
703 -- comparison is clear from MSB difference.
704 -- for signed, 0 is greater; for unsigned, 1 is greater
705 trapval := msb_a & msb_b & '0' & msb_b & msb_a;
707 -- Subtraction cannot overflow since MSBs are equal.
708 -- carry = 1 indicates RA is smaller (signed or unsigned)
709 a_lt := (not l and carry_32) or (l and carry_64);
710 trapval := a_lt & not a_lt & '0' & a_lt & not a_lt;
713 if e_in.insn_type = OP_CMP then
714 if e_in.is_signed = '1' then
715 newcrf := trapval(4 downto 2) & v.e.xerc.so;
717 newcrf := trapval(1 downto 0) & trapval(2) & v.e.xerc.so;
719 bf := insn_bf(e_in.insn);
720 crnum := to_integer(unsigned(bf));
721 v.e.write_cr_enable := '1';
722 v.e.write_cr_mask := num_to_fxm(crnum);
726 v.e.write_cr_data(hi downto lo) := newcrf;
729 -- trap instructions (tw, twi, td, tdi)
730 v.f.redirect_nia := std_logic_vector(to_unsigned(16#700#, 64));
731 -- set bit 46 to say trap occurred
732 ctrl_tmp.srr1(63 - 46) <= '1';
733 if or (trapval and insn_to(e_in.insn)) = '1' then
734 -- generate trap-type program interrupt
741 result := (others => '0');
742 for i in 0 to 14 loop
745 if (a_in(hi) xor b_in(hi) xor sum_with_carry(hi)) = '0' then
746 result(lo + 3 downto lo) := "0110";
749 if sum_with_carry(64) = '0' then
750 result(63 downto 60) := "0110";
754 newcrf := ppc_cmprb(a_in, b_in, insn_l(e_in.insn));
755 bf := insn_bf(e_in.insn);
756 crnum := to_integer(unsigned(bf));
757 v.e.write_cr_enable := '1';
758 v.e.write_cr_mask := num_to_fxm(crnum);
759 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
760 newcrf & newcrf & newcrf & newcrf;
762 newcrf := ppc_cmpeqb(a_in, b_in);
763 bf := insn_bf(e_in.insn);
764 crnum := to_integer(unsigned(bf));
765 v.e.write_cr_enable := '1';
766 v.e.write_cr_mask := num_to_fxm(crnum);
767 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
768 newcrf & newcrf & newcrf & newcrf;
769 when OP_AND | OP_OR | OP_XOR | OP_POPCNT | OP_PRTY | OP_CMPB | OP_EXTS |
771 result := logical_result;
776 abs_branch := insn_aa(e_in.insn);
777 if ctrl.msr(MSR_BE) = '1' then
782 bo := insn_bo(e_in.insn);
783 bi := insn_bi(e_in.insn);
784 if bo(4-2) = '0' then
785 result := std_ulogic_vector(unsigned(a_in) - 1);
787 v.e.write_reg := fast_spr_num(SPR_CTR);
790 taken_branch := ppc_bc_taken(bo, bi, cr_in, a_in);
791 abs_branch := insn_aa(e_in.insn);
792 if ctrl.msr(MSR_BE) = '1' then
797 -- read_data2 is target register (CTR, LR or TAR)
798 bo := insn_bo(e_in.insn);
799 bi := insn_bi(e_in.insn);
800 if bo(4-2) = '0' and e_in.insn(10) = '0' then
801 result := std_ulogic_vector(unsigned(a_in) - 1);
803 v.e.write_reg := fast_spr_num(SPR_CTR);
806 taken_branch := ppc_bc_taken(bo, bi, cr_in, a_in);
808 if ctrl.msr(MSR_BE) = '1' then
813 v.f.virt_mode := a_in(MSR_IR) or a_in(MSR_PR);
814 v.f.priv_mode := not a_in(MSR_PR);
815 v.f.big_endian := not a_in(MSR_LE);
816 v.f.mode_32bit := not a_in(MSR_SF);
817 -- Can't use msr_copy here because the partial function MSR
818 -- bits should be left unchanged, not zeroed.
819 ctrl_tmp.msr(63 downto 31) <= a_in(63 downto 31);
820 ctrl_tmp.msr(26 downto 22) <= a_in(26 downto 22);
821 ctrl_tmp.msr(15 downto 0) <= a_in(15 downto 0);
822 if a_in(MSR_PR) = '1' then
823 ctrl_tmp.msr(MSR_EE) <= '1';
824 ctrl_tmp.msr(MSR_IR) <= '1';
825 ctrl_tmp.msr(MSR_DR) <= '1';
827 -- mark this as a branch so CFAR gets updated
832 v.fp_exception_next := fp_in.exception and
833 (a_in(MSR_FE0) or a_in(MSR_FE1));
839 v.cntz_in_progress := '1';
842 crbit := to_integer(unsigned(insn_bc(e_in.insn)));
843 if cr_in(31-crbit) = '1' then
850 cr_op := insn_cr(e_in.insn);
851 report "CR OP " & to_hstring(cr_op);
852 if cr_op(0) = '0' then -- MCRF
853 bf := insn_bf(e_in.insn);
854 bfa := insn_bfa(e_in.insn);
855 v.e.write_cr_enable := '1';
856 crnum := to_integer(unsigned(bf));
857 scrnum := to_integer(unsigned(bfa));
858 v.e.write_cr_mask := num_to_fxm(crnum);
863 newcrf := cr_in(hi downto lo);
869 v.e.write_cr_data(hi downto lo) := newcrf;
872 v.e.write_cr_enable := '1';
873 bt := insn_bt(e_in.insn);
874 ba := insn_ba(e_in.insn);
875 bb := insn_bb(e_in.insn);
876 btnum := 31 - to_integer(unsigned(bt));
877 banum := 31 - to_integer(unsigned(ba));
878 bbnum := 31 - to_integer(unsigned(bb));
879 -- Bits 5-8 of cr_op give the truth table of the requested
881 cr_operands := cr_in(banum) & cr_in(bbnum);
882 crresult := cr_op(5 + to_integer(unsigned(cr_operands)));
883 v.e.write_cr_mask := num_to_fxm((31-btnum) / 4);
884 for i in 0 to 31 loop
886 v.e.write_cr_data(i) := crresult;
888 v.e.write_cr_data(i) := cr_in(i);
893 newcrf := v.e.xerc.ov & v.e.xerc.ca & v.e.xerc.ov32 & v.e.xerc.ca32;
894 bf := insn_bf(e_in.insn);
895 crnum := to_integer(unsigned(bf));
896 v.e.write_cr_enable := '1';
897 v.e.write_cr_mask := num_to_fxm(crnum);
898 v.e.write_cr_data := newcrf & newcrf & newcrf & newcrf &
899 newcrf & newcrf & newcrf & newcrf;
901 if random_err = '0' then
902 case e_in.insn(17 downto 16) is
904 result := x"00000000" & random_cond(31 downto 0);
906 result := random_raw;
908 result := random_cond;
911 result := (others => '1');
918 report "MFSPR to SPR " & integer'image(decode_spr_num(e_in.insn)) &
919 "=" & to_hstring(a_in);
921 if is_fast_spr(e_in.read_reg1) then
923 if decode_spr_num(e_in.insn) = SPR_XER then
924 -- bits 0:31 and 35:43 are treated as reserved and return 0s when read using mfxer
925 result(63 downto 32) := (others => '0');
926 result(63-32) := v.e.xerc.so;
927 result(63-33) := v.e.xerc.ov;
928 result(63-34) := v.e.xerc.ca;
929 result(63-35 downto 63-43) := "000000000";
930 result(63-44) := v.e.xerc.ov32;
931 result(63-45) := v.e.xerc.ca32;
935 case decode_spr_num(e_in.insn) is
939 spr_val(63 downto 32) := (others => '0');
940 spr_val(31 downto 0) := ctrl.tb(63 downto 32);
944 spr_val := ctrl.cfar;
946 spr_val(63 downto 32) := (others => '0');
947 spr_val(31 downto 0) := PVR_MICROWATT;
948 when 724 => -- LOG_ADDR SPR
949 spr_val := log_wr_addr & r.log_addr_spr;
950 when 725 => -- LOG_DATA SPR
951 spr_val := log_rd_data;
952 v.log_addr_spr := std_ulogic_vector(unsigned(r.log_addr_spr) + 1);
954 -- mfspr from unimplemented SPRs should be a nop in
955 -- supervisor mode and a program interrupt for user mode
956 if ctrl.msr(MSR_PR) = '1' then
963 if e_in.insn(20) = '0' then
965 result := x"00000000" & cr_in;
968 crnum := fxm_to_num(insn_fxm(e_in.insn));
969 result := (others => '0');
974 result(hi downto lo) := cr_in(hi downto lo);
980 v.e.write_cr_enable := '1';
981 if e_in.insn(20) = '0' then
983 v.e.write_cr_mask := insn_fxm(e_in.insn);
985 -- mtocrf: We require one hot priority encoding here
986 crnum := fxm_to_num(insn_fxm(e_in.insn));
987 v.e.write_cr_mask := num_to_fxm(crnum);
989 v.e.write_cr_data := c_in(31 downto 0);
991 if e_in.insn(16) = '1' then
992 -- just update EE and RI
993 ctrl_tmp.msr(MSR_EE) <= c_in(MSR_EE);
994 ctrl_tmp.msr(MSR_RI) <= c_in(MSR_RI);
996 -- Architecture says to leave out bits 3 (HV), 51 (ME)
997 -- and 63 (LE) (IBM bit numbering)
998 if e_in.is_32bit = '0' then
999 ctrl_tmp.msr(63 downto 61) <= c_in(63 downto 61);
1000 ctrl_tmp.msr(59 downto 32) <= c_in(59 downto 32);
1002 ctrl_tmp.msr(31 downto 13) <= c_in(31 downto 13);
1003 ctrl_tmp.msr(11 downto 1) <= c_in(11 downto 1);
1004 if c_in(MSR_PR) = '1' then
1005 ctrl_tmp.msr(MSR_EE) <= '1';
1006 ctrl_tmp.msr(MSR_IR) <= '1';
1007 ctrl_tmp.msr(MSR_DR) <= '1';
1010 v.fp_exception_next := fp_in.exception and
1011 (c_in(MSR_FE0) or c_in(MSR_FE1));
1015 report "MTSPR to SPR " & integer'image(decode_spr_num(e_in.insn)) &
1016 "=" & to_hstring(c_in);
1017 if is_fast_spr(e_in.write_reg) then
1020 if decode_spr_num(e_in.insn) = SPR_XER then
1021 v.e.xerc.so := c_in(63-32);
1022 v.e.xerc.ov := c_in(63-33);
1023 v.e.xerc.ca := c_in(63-34);
1024 v.e.xerc.ov32 := c_in(63-44);
1025 v.e.xerc.ca32 := c_in(63-45);
1026 v.e.write_xerc_enable := '1';
1030 case decode_spr_num(e_in.insn) is
1032 ctrl_tmp.dec <= c_in;
1033 when 724 => -- LOG_ADDR SPR
1034 v.log_addr_spr := c_in(31 downto 0);
1036 -- mtspr to unimplemented SPRs should be a nop in
1037 -- supervisor mode and a program interrupt for user mode
1038 if ctrl.msr(MSR_PR) = '1' then
1043 when OP_RLC | OP_RLCL | OP_RLCR | OP_SHL | OP_SHR | OP_EXTSWSLI =>
1044 result := rotator_result;
1045 if e_in.output_carry = '1' then
1046 set_carry(v.e, rotator_carry, rotator_carry);
1050 bfa := insn_bfa(e_in.insn);
1051 crbit := to_integer(unsigned(bfa)) * 4;
1052 result := (others => '0');
1053 if cr_in(31 - crbit) = '1' then
1054 result := (others => '1');
1055 elsif cr_in(30 - crbit) = '1' then
1060 v.f.redirect := '1';
1061 v.f.redirect_nia := next_nia;
1064 icache_inval <= '1';
1066 when OP_MUL_L64 | OP_MUL_H64 | OP_MUL_H32 =>
1068 v.mul_in_progress := '1';
1070 x_to_multiply.valid <= '1';
1072 when OP_DIV | OP_DIVE | OP_MOD =>
1074 v.div_in_progress := '1';
1076 x_to_divider.valid <= '1';
1083 v.e.rc := e_in.rc and valid_in;
1085 -- Mispredicted branches cause a redirect
1086 if is_branch = '1' then
1087 if taken_branch = '1' then
1088 ctrl_tmp.cfar <= e_in.nia;
1090 if e_in.br_pred = '0' then
1091 if abs_branch = '1' then
1092 v.f.redirect_nia := b_in;
1094 v.f.redirect_nia := std_ulogic_vector(signed(e_in.nia) + signed(b_in));
1097 v.f.redirect_nia := next_nia;
1099 if taken_branch /= e_in.br_pred then
1100 v.f.redirect := '1';
1104 -- Update LR on the next cycle after a branch link
1105 -- If we're not writing back anything else, we can write back LR
1106 -- this cycle, otherwise we take an extra cycle. We use the
1107 -- exc_write path since next_nia is written through that path
1109 if e_in.lr = '1' then
1110 if result_en = '0' then
1111 v.e.exc_write_enable := '1';
1112 v.e.exc_write_data := next_nia;
1113 v.e.exc_write_reg := fast_spr_num(SPR_LR);
1116 v.next_lr := next_nia;
1118 report "Delayed LR update to " & to_hstring(next_nia);
1123 elsif valid_in = '1' then
1124 -- instruction for other units, i.e. LDST
1125 if e_in.unit = LDST then
1127 elsif e_in.unit = NONE then
1129 elsif HAS_FPU and e_in.unit = FPU then
1132 -- Handling an ITLB miss doesn't count as having executed an instruction
1133 if e_in.insn_type = OP_FETCH_FAILED then
1138 -- The following cases all occur when r.busy = 1 and therefore
1139 -- valid_in = 0. Hence they don't happen in the same cycle as any of
1140 -- the cases above which depend on valid_in = 1.
1142 if r.f.redirect = '1' then
1145 if r.lr_update = '1' then
1146 v.e.exc_write_enable := '1';
1147 v.e.exc_write_data := r.next_lr;
1148 v.e.exc_write_reg := fast_spr_num(SPR_LR);
1150 -- Keep r.e.write_data unchanged next cycle in case it is needed
1151 -- for a forwarded result (e.g. for CTR).
1152 result := r.e.write_data;
1153 elsif r.cntz_in_progress = '1' then
1154 -- cnt[lt]z always takes two cycles
1155 result := countzero_result;
1157 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1158 v.e.rc := r.slow_op_rc;
1159 v.e.xerc := r.slow_op_xerc;
1161 elsif r.mul_in_progress = '1' or r.div_in_progress = '1' then
1162 if (r.mul_in_progress = '1' and multiply_to_x.valid = '1') or
1163 (r.div_in_progress = '1' and divider_to_x.valid = '1') then
1164 if r.mul_in_progress = '1' then
1166 case r.slow_op_insn is
1168 result := multiply_to_x.result(63 downto 32) &
1169 multiply_to_x.result(63 downto 32);
1171 result := multiply_to_x.result(127 downto 64);
1174 result := multiply_to_x.result(63 downto 0);
1177 result := divider_to_x.write_reg_data;
1178 overflow := divider_to_x.overflow;
1180 if r.mul_in_progress = '1' and r.slow_op_oe = '1' then
1181 -- have to wait until next cycle for overflow indication
1182 v.mul_finish := '1';
1186 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1187 v.e.rc := r.slow_op_rc;
1188 v.e.xerc := r.slow_op_xerc;
1189 v.e.write_xerc_enable := r.slow_op_oe;
1190 -- We must test oe because the RC update code in writeback
1191 -- will use the xerc value to set CR0:SO so we must not clobber
1192 -- xerc if OE wasn't set.
1193 if r.slow_op_oe = '1' then
1194 v.e.xerc.ov := overflow;
1195 v.e.xerc.ov32 := overflow;
1196 v.e.xerc.so := r.slow_op_xerc.so or overflow;
1202 v.mul_in_progress := r.mul_in_progress;
1203 v.div_in_progress := r.div_in_progress;
1205 elsif r.mul_finish = '1' then
1206 result := r.e.write_data;
1208 v.e.write_reg := gpr_to_gspr(r.slow_op_dest);
1209 v.e.rc := r.slow_op_rc;
1210 v.e.xerc := r.slow_op_xerc;
1211 v.e.write_xerc_enable := r.slow_op_oe;
1212 v.e.xerc.ov := multiply_to_x.overflow;
1213 v.e.xerc.ov32 := multiply_to_x.overflow;
1214 v.e.xerc.so := r.slow_op_xerc.so or multiply_to_x.overflow;
1218 -- Generate FP-type program interrupt. fp_in.interrupt will only
1219 -- be set during the execution of a FP instruction.
1220 -- The case where MSR[FE0,FE1] goes from zero to non-zero is
1221 -- handled above by mtmsrd and rfid setting v.fp_exception_next.
1222 if HAS_FPU and fp_in.interrupt = '1' then
1223 v.f.redirect_nia := std_logic_vector(to_unsigned(16#700#, 64));
1224 ctrl_tmp.srr1(63 - 43) <= '1';
1228 if illegal = '1' or (HAS_FPU and fp_in.illegal = '1') then
1230 v.f.redirect_nia := std_logic_vector(to_unsigned(16#700#, 64));
1231 -- Since we aren't doing Hypervisor emulation assist (0xe40) we
1232 -- set bit 44 to indicate we have an illegal
1233 ctrl_tmp.srr1(63 - 44) <= '1';
1236 if exception = '1' then
1237 v.e.exc_write_enable := '1';
1238 if exception_nextpc = '1' then
1239 v.e.exc_write_data := next_nia;
1243 if do_trace = '1' then
1244 v.trace_next := '1';
1247 v.e.write_data := result;
1248 v.e.write_enable := result_en and not exception;
1250 -- generate DSI or DSegI for load/store exceptions
1251 -- or ISI or ISegI for instruction fetch exceptions
1252 if l_in.exception = '1' then
1253 if l_in.alignment = '1' then
1254 v.f.redirect_nia := std_logic_vector(to_unsigned(16#600#, 64));
1255 elsif l_in.instr_fault = '0' then
1256 if l_in.segment_fault = '0' then
1257 v.f.redirect_nia := std_logic_vector(to_unsigned(16#300#, 64));
1259 v.f.redirect_nia := std_logic_vector(to_unsigned(16#380#, 64));
1262 if l_in.segment_fault = '0' then
1263 ctrl_tmp.srr1(63 - 33) <= l_in.invalid;
1264 ctrl_tmp.srr1(63 - 35) <= l_in.perm_error; -- noexec fault
1265 ctrl_tmp.srr1(63 - 44) <= l_in.badtree;
1266 ctrl_tmp.srr1(63 - 45) <= l_in.rc_error;
1267 v.f.redirect_nia := std_logic_vector(to_unsigned(16#400#, 64));
1269 v.f.redirect_nia := std_logic_vector(to_unsigned(16#480#, 64));
1272 v.e.exc_write_enable := '1';
1273 v.e.exc_write_reg := fast_spr_num(SPR_SRR0);
1274 report "ldst exception writing srr0=" & to_hstring(r.last_nia);
1277 if exception = '1' or l_in.exception = '1' then
1278 ctrl_tmp.irq_state <= WRITE_SRR1;
1279 v.f.redirect := '1';
1280 v.f.virt_mode := '0';
1281 v.f.priv_mode := '1';
1282 -- XXX need an interrupt LE bit here, e.g. from LPCR
1283 v.f.big_endian := '0';
1284 v.f.mode_32bit := '0';
1287 if v.f.redirect = '1' then
1292 -- Outputs to loadstore1 (async)
1293 lv.op := e_in.insn_type;
1298 lv.write_reg := e_in.write_reg;
1299 lv.length := e_in.data_len;
1300 lv.byte_reverse := e_in.byte_reverse xnor ctrl.msr(MSR_LE);
1301 lv.sign_extend := e_in.sign_extend;
1302 lv.update := e_in.update;
1303 lv.update_reg := gspr_to_gpr(e_in.read_reg1);
1304 lv.xerc := v.e.xerc;
1305 lv.reserve := e_in.reserve;
1307 lv.insn := e_in.insn;
1308 -- decode l*cix and st*cix instructions here
1309 if e_in.insn(31 downto 26) = "011111" and e_in.insn(10 downto 9) = "11" and
1310 e_in.insn(5 downto 1) = "10101" then
1313 lv.virt_mode := ctrl.msr(MSR_DR);
1314 lv.priv_mode := not ctrl.msr(MSR_PR);
1315 lv.mode_32bit := not ctrl.msr(MSR_SF);
1316 lv.is_32bit := e_in.is_32bit;
1319 fv.op := e_in.insn_type;
1321 fv.insn := e_in.insn;
1322 fv.single := e_in.is_32bit;
1323 fv.fe_mode := ctrl.msr(MSR_FE0) & ctrl.msr(MSR_FE1);
1327 fv.frt := e_in.write_reg;
1329 fv.out_cr := e_in.output_cr;
1339 flush_out <= f_out.redirect;
1341 exception_log <= exception;
1342 irq_valid_log <= irq_valid;
1345 e1_log: if LOG_LENGTH > 0 generate
1346 signal log_data : std_ulogic_vector(14 downto 0);
1348 ex1_log : process(clk)
1350 if rising_edge(clk) then
1351 log_data <= ctrl.msr(MSR_EE) & ctrl.msr(MSR_PR) &
1352 ctrl.msr(MSR_IR) & ctrl.msr(MSR_DR) &
1355 std_ulogic_vector(to_unsigned(irq_state_t'pos(ctrl.irq_state), 1)) &
1364 log_out <= log_data;
1366 end architecture behaviour;