diff options
| author | Junyan He <junyan.he@linux.intel.com> | 2015-12-10 15:04:25 +0800 |
|---|---|---|
| committer | Yang Rong <rong.r.yang@intel.com> | 2015-12-14 15:11:44 +0800 |
| commit | 1f030e70ae8123b388076974aa2501bd98eb6b4b (patch) | |
| tree | 7d29b0f2a0fcb2f9fa2a2c0fe70a19ee741dee22 | |
| parent | 99f968ddd378e2c957c37d99c2b9f408ceb7ce38 (diff) | |
Backend: Add state register into schedule consideration.
Because the workgroup OP has forwarding msg and wait functions,
it needs all the threads to sync with each other. It has very
similar behavior as BARRIER, so we add it into schedule consideration
accordingly.
Signed-off-by: Junyan He <junyan.he@linux.intel.com>
Reviewed-by: Yang Rong <rong.r.yang@intel.com>
| -rw-r--r-- | backend/src/backend/gen_insn_scheduling.cpp | 16 |
1 files changed, 12 insertions, 4 deletions
diff --git a/backend/src/backend/gen_insn_scheduling.cpp b/backend/src/backend/gen_insn_scheduling.cpp index c413bac0..245d17a5 100644 --- a/backend/src/backend/gen_insn_scheduling.cpp +++ b/backend/src/backend/gen_insn_scheduling.cpp @@ -199,8 +199,10 @@ namespace gbe static const uint32_t MAX_ACC_REGISTER = 1u; /*! Maximum number of *physical* tm registers */ static const uint32_t MAX_TM_REGISTER = 1u; + /*! Maximum number of state registers */ + static const uint32_t MAX_ST_REGISTER = 2u; /*! Maximum number of *physical* arf registers */ - static const uint32_t MAX_ARF_REGISTER = MAX_FLAG_REGISTER + MAX_ACC_REGISTER + MAX_TM_REGISTER; + static const uint32_t MAX_ARF_REGISTER = MAX_FLAG_REGISTER + MAX_ACC_REGISTER + MAX_TM_REGISTER + MAX_ST_REGISTER; /*! Stores the last node that wrote to a register / memory ... */ vector<ScheduleDAGNode*> nodes; /*! store nodes each node depends on */ @@ -343,6 +345,9 @@ namespace gbe return grfNum + MAX_FLAG_REGISTER + nr; } else if (file == GEN_ARF_TM) { return grfNum + MAX_FLAG_REGISTER + MAX_ACC_REGISTER; + } else if (file == GEN_ARF_STATE) { + GBE_ASSERT(nr < MAX_ST_REGISTER); + return grfNum + MAX_FLAG_REGISTER + MAX_ACC_REGISTER + MAX_TM_REGISTER + nr; } else { NOT_SUPPORTED; return 0; @@ -510,7 +515,8 @@ namespace gbe // Consider barriers and wait are reading memory (local and global) if (insn.opcode == SEL_OP_BARRIER || insn.opcode == SEL_OP_FENCE || - insn.opcode == SEL_OP_WAIT) { + insn.opcode == SEL_OP_WAIT || + insn.opcode == SEL_OP_WORKGROUP_OP) { const uint32_t memIndex = tracker.getMemoryIndex(); tracker.addDependency(node, memIndex, READ_AFTER_WRITE); } @@ -572,7 +578,8 @@ namespace gbe // Consider barriers and wait are reading memory (local and global) if (insn.opcode == SEL_OP_BARRIER || insn.opcode == SEL_OP_FENCE || - insn.opcode == SEL_OP_WAIT) { + insn.opcode == SEL_OP_WAIT || + insn.opcode == SEL_OP_WORKGROUP_OP) { const uint32_t memIndex = tracker.getMemoryIndex(); tracker.addDependency(memIndex, node, WRITE_AFTER_READ); } @@ -602,7 +609,8 @@ namespace gbe || node->insn.opcode == SEL_OP_BARRIER || node->insn.opcode == SEL_OP_CALC_TIMESTAMP || node->insn.opcode == SEL_OP_STORE_PROFILING - || node->insn.opcode == SEL_OP_WAIT) + || node->insn.opcode == SEL_OP_WAIT + || node->insn.opcode == SEL_OP_WORKGROUP_OP) tracker.makeBarrier(insnID, insnNum); } |