From 82a636ff5560370ff9e606cc01d9171876bd9690 Mon Sep 17 00:00:00 2001 From: Rimas Avizienis Date: Tue, 15 Nov 2011 00:51:45 -0800 Subject: [PATCH] AMOADD, AMOAND, AMOOR, AMOSWAP working --- rocket/src/main/scala/ctrl.scala | 17 +++++++++-------- rocket/src/main/scala/dcache.scala | 24 ++++++++++++++++++++---- 2 files changed, 29 insertions(+), 12 deletions(-) diff --git a/rocket/src/main/scala/ctrl.scala b/rocket/src/main/scala/ctrl.scala index 9ee4c92d..5e7e6add 100644 --- a/rocket/src/main/scala/ctrl.scala +++ b/rocket/src/main/scala/ctrl.scala @@ -190,6 +190,15 @@ class rocketCtrl extends Component SW-> List(Y, BR_N, REN_Y,REN_Y,A2_SPLIT,A1_RS1,DW_XPR,FN_ADD, M_Y,M_XWR, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_X, WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), SD-> List(xpr64, BR_N, REN_Y,REN_Y,A2_SPLIT,A1_RS1,DW_XPR,FN_ADD, M_Y,M_XWR, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_X, WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOADD_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_ADD, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOSWAP_W->List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_SWAP,MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOAND_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_AND, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOOR_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_OR, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOADD_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_ADD, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOSWAP_D->List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_SWAP,MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOAND_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_AND, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + AMOOR_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_OR, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), + LUI-> List(Y, BR_N, REN_N,REN_Y,A2_0, A1_LUI,DW_XPR,FN_ADD, M_N,M_X, MT_X, N,MUL_X, N,DIV_X, WEN_Y,WA_RD,WB_ALU,REN_N,WEN_N,I_X ,SYNC_N,N,N,N), SLTI -> List(Y, BR_N, REN_N,REN_Y,A2_SEXT, A1_RS1,DW_XPR,FN_SLT, M_N,M_X, MT_X, N,MUL_X, N,DIV_X, WEN_Y,WA_RD,WB_ALU,REN_N,WEN_N,I_X ,SYNC_N,N,N,N), SLTIU-> List(Y, BR_N, REN_N,REN_Y,A2_SEXT, A1_RS1,DW_XPR,FN_SLTU,M_N,M_X, MT_X, N,MUL_X, N,DIV_X, WEN_Y,WA_RD,WB_ALU,REN_N,WEN_N,I_X ,SYNC_N,N,N,N), @@ -254,18 +263,10 @@ class rocketCtrl extends Component FSD-> List(FPU_Y, BR_N, REN_Y,REN_Y,A2_SPLIT,A1_RS1,DW_XPR,FN_ADD, M_Y,M_FWR, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_X, WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), // atomic memory operations - AMOADD_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_ADD, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOSWAP_W->List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_SWAP,MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOAND_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_AND, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOOR_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_OR, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMIN_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MIN, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMAX_W-> List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MAX, MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMINU_W->List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MINU,MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMAXU_W->List(Y, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MAXU,MT_W, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOADD_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_ADD, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOSWAP_D->List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_SWAP,MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOAND_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_AND, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), - AMOOR_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_OR, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMIN_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MIN, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMAX_D-> List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MAX, MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), AMOMINU_D->List(xpr64, BR_N, REN_Y,REN_Y,A2_0, A1_RS1,DW_XPR,FN_ADD, M_Y,M_XA_MINU,MT_D, N,MUL_X, N,DIV_X, WEN_N,WA_RD,WB_X, REN_N,WEN_N,I_X ,SYNC_N,N,N,N), diff --git a/rocket/src/main/scala/dcache.scala b/rocket/src/main/scala/dcache.scala index e7ac5744..d683ac14 100644 --- a/rocket/src/main/scala/dcache.scala +++ b/rocket/src/main/scala/dcache.scala @@ -219,7 +219,7 @@ class rocketDCacheDM(lines: Int) extends Component { otherwise { r_cpu_req_val <== Bool(false); } - when (((state === s_resolve_miss) && r_req_load) || (state === s_replay_load)) { + when (((state === s_resolve_miss) && (r_req_load || r_req_amo)) || (state === s_replay_load)) { r_cpu_resp_val <== Bool(true); } otherwise { @@ -339,11 +339,23 @@ class rocketDCacheDM(lines: Int) extends Component { Mux(r_cpu_req_type === MT_D, ~Bits(0,8), Mux(r_cpu_req_idx(2).toBool, Cat(~Bits(0,4), Bits(0,4)), Cat(Bits(0,4), ~Bits(0,4)))); - + + val amo_store_wmask_d = Cat(Fill(8, amo_wmask(7)), + Fill(8, amo_wmask(6)), + Fill(8, amo_wmask(5)), + Fill(8, amo_wmask(4)), + Fill(8, amo_wmask(3)), + Fill(8, amo_wmask(2)), + Fill(8, amo_wmask(1)), + Fill(8, amo_wmask(0))); + + val amo_store_idx_sel = r_cpu_req_idx(offsetlsb).toBool; + val amo_store_wmask = Mux(amo_store_idx_sel, Cat(amo_store_wmask_d, Bits(0,64)), Cat(Bits(0,64), amo_store_wmask_d)); + val amo_alu = new rocketDCacheAmoALU(); amo_alu.io.cmd := r_cpu_req_cmd; amo_alu.io.wmask := amo_wmask; - amo_alu.io.lhs := r_resp_data.toUFix; + amo_alu.io.lhs := Mux(r_cpu_resp_val, resp_data, r_resp_data).toUFix; amo_alu.io.rhs := r_amo_data.toUFix; val amo_alu_out = amo_alu.io.result; @@ -363,7 +375,11 @@ class rocketDCacheDM(lines: Int) extends Component { (state === s_write_amo) || drain_store || resolve_store; - data_array.io.bweb := Mux((state === s_refill), ~Bits(0,128), store_wmask); + data_array.io.bweb := + Mux((state === s_refill), ~Bits(0,128), + Mux((state === s_write_amo), amo_store_wmask, + store_wmask)); + data_array.io.ce := (io.cpu.req_val && io.cpu.req_rdy && (req_load || req_amo)) || (state === s_start_writeback) ||