riscv/rocket-chip
1
0
Fork 0
Code Releases Activity

Improve back-to-back integer multiplication performance

Browse Source 
More exact hazard checking in the decode stage avoids a pipeline flush.
This commit is contained in:
Andrew Waterman
2016-10-04 00:04:46 -07:00
parent 7b69f1f261
commit 516481b68b
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/main/scala/rocket/rocket.scala
View File

@@ -465,6 +465,7 @@ class Rocket(implicit p: Parameters) extends CoreModule()(p) {
wb_reg_pc := mem_reg_pc wb_reg_pc := mem_reg_pc
} }
val wb_wxd = wb_reg_valid && wb_ctrl.wxd
val wb_set_sboard = wb_ctrl.div || wb_dcache_miss || wb_ctrl.rocc val wb_set_sboard = wb_ctrl.div || wb_dcache_miss || wb_ctrl.rocc
val replay_wb_common = io.dmem.s2_nack || wb_reg_replay val replay_wb_common = io.dmem.s2_nack || wb_reg_replay
val replay_wb_rocc = wb_reg_valid && wb_ctrl.rocc && !io.rocc.cmd.ready val replay_wb_rocc = wb_reg_valid && wb_ctrl.rocc && !io.rocc.cmd.ready
@@ -479,12 +480,12 @@ class Rocket(implicit p: Parameters) extends CoreModule()(p) {
val dmem_resp_valid = io.dmem.resp.valid && io.dmem.resp.bits.has_data val dmem_resp_valid = io.dmem.resp.valid && io.dmem.resp.bits.has_data
val dmem_resp_replay = dmem_resp_valid && io.dmem.resp.bits.replay val dmem_resp_replay = dmem_resp_valid && io.dmem.resp.bits.replay
div.io.resp.ready := !(wb_reg_valid && wb_ctrl.wxd) div.io.resp.ready := !wb_wxd
val ll_wdata = Wire(init = div.io.resp.bits.data) val ll_wdata = Wire(init = div.io.resp.bits.data)
val ll_waddr = Wire(init = div.io.resp.bits.tag) val ll_waddr = Wire(init = div.io.resp.bits.tag)
val ll_wen = Wire(init = div.io.resp.fire()) val ll_wen = Wire(init = div.io.resp.fire())
if (usingRoCC) { if (usingRoCC) {
io.rocc.resp.ready := !(wb_reg_valid && wb_ctrl.wxd) io.rocc.resp.ready := !wb_wxd
when (io.rocc.resp.fire()) { when (io.rocc.resp.fire()) {
div.io.resp.ready := Bool(false) div.io.resp.ready := Bool(false)
ll_wdata := io.rocc.resp.bits.data ll_wdata := io.rocc.resp.bits.data
@@ -581,6 +582,7 @@ class Rocket(implicit p: Parameters) extends CoreModule()(p) {
id_ctrl.fp && id_stall_fpu || id_ctrl.fp && id_stall_fpu ||
id_ctrl.mem && dcache_blocked || // reduce activity during D$ misses id_ctrl.mem && dcache_blocked || // reduce activity during D$ misses
id_ctrl.rocc && rocc_blocked || // reduce activity while RoCC is busy id_ctrl.rocc && rocc_blocked || // reduce activity while RoCC is busy
id_ctrl.div && (!(div.io.req.ready || (div.io.resp.valid && !wb_wxd)) || div.io.req.valid) || // reduce odds of replay
id_do_fence || id_do_fence ||
csr.io.csr_stall csr.io.csr_stall
ctrl_killd := !ibuf.io.inst(0).valid || ibuf.io.inst(0).bits.replay || take_pc_mem_wb || ctrl_stalld || csr.io.interrupt ctrl_killd := !ibuf.io.inst(0).valid || ibuf.io.inst(0).bits.replay || take_pc_mem_wb || ctrl_stalld || csr.io.interrupt