Remove single-issue constraint for most loads and stores

This removes the constraint that loads and stores are single-issue,
at the expense of a stall of at least 2 cycles for every load and
store.

To do this, we plumb the existing stall signal that was generated
in dcache to core, where it gets ORed with the stall signal from
execute1.  Execute1 generates a stall signal for the first two
cycles of each load and store, and dcache generates the stall
signal in the 3rd and subsequent cycles if it needs to.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

diff --git a/core.vhdl b/core.vhdl
index bc0b16fecaefd93d06a7161f0496bb70e3ce47bf..87e73a4c1c0f5529282122f1702f2778be722a49 100644 (file)
--- a/core.vhdl
+++ b/core.vhdl
@@ -73,6 +73,7 @@ architecture behave of core is
     signal decode2_stall_out : std_ulogic;
     signal ex1_icache_inval: std_ulogic;
     signal ex1_stall_out: std_ulogic;
+    signal dcache_stall_out: std_ulogic;
 
     signal flush: std_ulogic;
 
@@ -195,7 +196,7 @@ begin
             c_in => cr_file_to_decode2,
             c_out => decode2_to_cr_file
             );
-    decode2_stall_in <= ex1_stall_out;
+    decode2_stall_in <= ex1_stall_out or dcache_stall_out;
 
     register_file_0: entity work.register_file
         generic map (
@@ -257,6 +258,7 @@ begin
            rst => core_rst,
             d_in => loadstore1_to_dcache,
             d_out => dcache_to_writeback,
+            stall_out => dcache_stall_out,
             wishbone_in => wishbone_data_in,
             wishbone_out => wishbone_data_out
             );

diff --git a/decode1.vhdl b/decode1.vhdl
index f1b5ad430512f313f3f9b9d9ae8e99df0c0b0444..bca7c2aceb3b12476636601465c84602281cf262 100644 (file)
--- a/decode1.vhdl
+++ b/decode1.vhdl
@@ -46,26 +46,26 @@ architecture behaviour of decode1 is
                16 =>       (ALU,    OP_BC,        SPR,        CONST_BD,    NONE, SPR , '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '1', '0'), -- bc
                11 =>       (ALU,    OP_CMP,       RA,         CONST_SI,    NONE, NONE, '0', '1', '1', '0', ONE,  '0', NONE, '0', '0', '0', '0', '0', '1', NONE, '0', '0'), -- cmpi
                10 =>       (ALU,    OP_CMP,       RA,         CONST_UI,    NONE, NONE, '0', '1', '1', '0', ONE,  '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- cmpli
-               34 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lbz
-               35 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lbzu
-               42 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '0', '0', '0', '0', NONE, '0', '1'), -- lha
-               43 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '1', '0', '0', '0', NONE, '0', '1'), -- lhau
-               40 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lhz
-               41 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lhzu
-               32 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lwz
-                33 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lwzu
+               34 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lbz
+               35 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lbzu
+               42 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '0', '0', '0', '0', NONE, '0', '0'), -- lha
+               43 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '1', '0', '0', '0', NONE, '0', '0'), -- lhau
+               40 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lhz
+               41 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lhzu
+               32 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lwz
+                33 =>       (LDST,   OP_LOAD,      RA_OR_ZERO, CONST_SI,    NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lwzu
                 7 =>       (ALU,    OP_MUL_L64,   RA,         CONST_SI,    NONE, RT,   '0', '1', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '1', NONE, '0', '0'), -- mulli
                24 =>       (ALU,    OP_OR,        NONE,       CONST_UI,    RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- ori
                25 =>       (ALU,    OP_OR,        NONE,       CONST_UI_HI, RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- oris
                20 =>       (ALU,    OP_RLC,       RA,         CONST_SH32,  RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwimi
                21 =>       (ALU,    OP_RLC,       NONE,       CONST_SH32,  RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwinm
                23 =>       (ALU,    OP_RLC,       NONE,       RB,          RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- rlwnm
-               38 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', RC,   '0', '1'), -- stb
-               39 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', RC,   '0', '1'), -- stbu
-               44 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- sth
-               45 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- sthu
-               36 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- stw
-               37 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- stwu
+               38 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- stb
+               39 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', RC,   '0', '0'), -- stbu
+               44 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- sth
+               45 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- sthu
+               36 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stw
+               37 =>       (LDST,   OP_STORE,     RA_OR_ZERO, CONST_SI,    RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- stwu
                 8 =>       (ALU,    OP_ADD,       RA,         CONST_SI,    NONE, RT,   '0', '0', '1', '0', ONE,  '1', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- subfic
                 2 =>       (ALU,    OP_TDI,       RA,         CONST_SI,    NONE, NONE, '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- tdi
                --PPC_TWI 3
@@ -215,25 +215,25 @@ architecture behaviour of decode1 is
                2#1110101111#  =>       (ALU,    OP_ISEL,      RA_OR_ZERO, RB,          NONE, RT,   '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- isel
                2#1111001111#  =>       (ALU,    OP_ISEL,      RA_OR_ZERO, RB,          NONE, RT,   '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- isel
                2#1111101111#  =>       (ALU,    OP_ISEL,      RA_OR_ZERO, RB,          NONE, RT,   '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- isel
-               2#0000110100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- lbarx
-               2#0001110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lbzux
-               2#0001010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lbzx
-               2#0001010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- ldarx
-               2#1000010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- ldbrx
-               2#0000110101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- ldux
-               2#0000010101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- ldx
-               2#0001110100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- lharx
-               2#0101110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '1', '0', '0', '0', NONE, '0', '1'), -- lhaux
-               2#0101010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '0', '0', '0', '0', NONE, '0', '1'), -- lhax
-               2#1100010110#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lhbrx
-               2#0100110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lhzux
-               2#0100010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lhzx
-               2#0000010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- lwarx
-               2#0101110101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '1', '1', '0', '0', '0', NONE, '0', '1'), -- lwaux
-               2#0101010101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '1', '0', '0', '0', '0', NONE, '0', '1'), -- lwax
-               2#1000010110#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lwbrx
-               2#0000110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- lwzux
-               2#0000010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- lwzx
+               2#0000110100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- lbarx
+               2#0001110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lbzux
+               2#0001010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lbzx
+               2#0001010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- ldarx
+               2#1000010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- ldbrx
+               2#0000110101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- ldux
+               2#0000010101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- ldx
+               2#0001110100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- lharx
+               2#0101110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '1', '0', '0', '0', NONE, '0', '0'), -- lhaux
+               2#0101010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '1', '0', '0', '0', '0', NONE, '0', '0'), -- lhax
+               2#1100010110#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lhbrx
+               2#0100110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lhzux
+               2#0100010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lhzx
+               2#0000010100#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- lwarx
+               2#0101110101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '1', '1', '0', '0', '0', NONE, '0', '0'), -- lwaux
+               2#0101010101#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '1', '0', '0', '0', '0', NONE, '0', '0'), -- lwax
+               2#1000010110#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lwbrx
+               2#0000110111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- lwzux
+               2#0000010111#  =>       (LDST,   OP_LOAD,      RA_OR_ZERO, RB,          NONE, RT,   '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- lwzx
                -- 2#1000000000# mcrxr
                -- 2#1001000000# mcrxrx
                2#0000010011#  =>       (ALU,    OP_MFCR,      NONE,       NONE,        NONE, RT,   '1', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- mfcr/mfocrf
@@ -278,21 +278,21 @@ architecture behaviour of decode1 is
                2#1100111000#  =>       (ALU,    OP_SHR,       NONE,       CONST_SH32,  RS,   RA,   '0', '0', '0', '0', ZERO, '1', NONE, '0', '0', '0', '0', '1', '1', RC,   '0', '0'), -- srawi
                2#1000011011#  =>       (ALU,    OP_SHR,       NONE,       RB,          RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- srd
                2#1000011000#  =>       (ALU,    OP_SHR,       NONE,       RB,          RS,   RA,   '0', '0', '0', '0', ZERO, '0', NONE, '0', '0', '0', '0', '1', '0', RC,   '0', '0'), -- srw
-               2#1010110110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '1', '0', '0', RC,   '0', '1'), -- stbcx
-               2#0011110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', RC,   '0', '1'), -- stbux
-               2#0011010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', RC,   '0', '1'), -- stbx
-               2#1010010100#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- stdbrx
-               2#0011010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- stdcx
-               2#0010110101#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- stdux
-               2#0010010101#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- stdx
-               2#1110010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- sthbrx
-               2#1011010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- sthcx
-               2#0110110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- sthux
-               2#0110010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- sthx
-               2#1010010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '1', '0', '0', '0', '0', '0', NONE, '0', '1'), -- stwbrx
-               2#0010010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '1', '0', '0', NONE, '0', '1'), -- stwcx
-               2#0010110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '1'), -- stwux
-               2#0010010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '1'), -- stwx
+               2#1010110110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '1', '0', '0', RC,   '0', '0'), -- stbcx
+               2#0011110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '1', '0', '0', '0', RC,   '0', '0'), -- stbux
+               2#0011010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is1B, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- stbx
+               2#1010010100#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stdbrx
+               2#0011010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- stdcx
+               2#0010110101#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- stdux
+               2#0010010101#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is8B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stdx
+               2#1110010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- sthbrx
+               2#1011010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- sthcx
+               2#0110110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- sthux
+               2#0110010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is2B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- sthx
+               2#1010010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '1', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stwbrx
+               2#0010010110#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '1', '0', '0', NONE, '0', '0'), -- stwcx
+               2#0010110111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '1', '0', '0', '0', NONE, '0', '0'), -- stwux
+               2#0010010111#  =>       (LDST,   OP_STORE,     RA_OR_ZERO, RB,          RS,   NONE, '0', '0', '0', '0', ZERO, '0', is4B, '0', '0', '0', '0', '0', '0', NONE, '0', '0'), -- stwx
                2#0000101000#  =>       (ALU,    OP_ADD,       RA,         RB,          NONE, RT,   '0', '0', '1', '0', ONE,  '0', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- subf
                2#1000101000#  =>       (ALU,    OP_ADD,       RA,         RB,          NONE, RT,   '0', '0', '1', '0', ONE,  '0', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- subfo
                2#0000001000#  =>       (ALU,    OP_ADD,       RA,         RB,          NONE, RT,   '0', '0', '1', '0', ONE,  '1', NONE, '0', '0', '0', '0', '0', '0', RC,   '0', '0'), -- subfc

diff --git a/execute1.vhdl b/execute1.vhdl
index e41a743a5dbbbbb32fc20247ca7861039359e2d6..c536a27cfc93be12f17511e87240e6b991834136 100644 (file)
--- a/execute1.vhdl
+++ b/execute1.vhdl
@@ -42,6 +42,7 @@ architecture behaviour of execute1 is
        next_lr : std_ulogic_vector(63 downto 0);
        mul_in_progress : std_ulogic;
         div_in_progress : std_ulogic;
+        ldst_in_progress : std_ulogic;
         cntz_in_progress : std_ulogic;
        slow_op_dest : gpr_index_t;
        slow_op_rc : std_ulogic;
@@ -263,6 +264,7 @@ begin
        v.mul_in_progress := '0';
         v.div_in_progress := '0';
         v.cntz_in_progress := '0';
+        v.ldst_in_progress := '0';
 
        -- signals to multiply unit
        x_to_multiply <= Execute1ToMultiplyInit;
@@ -660,6 +662,8 @@ begin
             when OP_LOAD | OP_STORE =>
                 -- loadstore/dcache has its own port to writeback
                 v.e.valid := '0';
+                stall_out <= '1';
+                v.ldst_in_progress := '1';
 
             when others =>
                terminate_out <= '1';
@@ -699,6 +703,10 @@ begin
             v.e.rc := v.slow_op_rc;
             v.e.xerc := v.slow_op_xerc;
             v.e.valid := '1';
+        elsif r.ldst_in_progress = '1' then
+            -- assert stall for 2 cycles on load/store, then
+            -- the stall output from dcache takes over
+            stall_out <= '1';
        elsif r.mul_in_progress = '1' or r.div_in_progress = '1' then
            if (r.mul_in_progress = '1' and multiply_to_x.valid = '1') or
               (r.div_in_progress = '1' and divider_to_x.valid = '1') then