[Beignet] [PATCH 09/15] Backend: Add WORKGROUP_OP instruction selection.
Pan Xiuli
xiuli.pan at intel.com
Wed Jan 20 22:51:49 PST 2016
From: Junyan He <junyan.he at linux.intel.com>
Signed-off-by: Junyan He <junyan.he at linux.intel.com>
Reviewed-by: Yang Rong <rong.r.yang at intel.com>
---
backend/src/backend/gen_context.cpp | 3 ++
backend/src/backend/gen_context.hpp | 1 +
.../src/backend/gen_insn_gen7_schedule_info.hxx | 1 +
backend/src/backend/gen_insn_selection.cpp | 34 ++++++++++++++++++++++
backend/src/backend/gen_insn_selection.hpp | 1 +
backend/src/backend/gen_insn_selection.hxx | 1 +
6 files changed, 41 insertions(+)
diff --git a/backend/src/backend/gen_context.cpp b/backend/src/backend/gen_context.cpp
index 05359af..ed6c9f0 100644
--- a/backend/src/backend/gen_context.cpp
+++ b/backend/src/backend/gen_context.cpp
@@ -2345,6 +2345,9 @@ namespace gbe
p->TYPED_WRITE(header, true, bti);
}
+ void GenContext::emitWorkGroupOpInstruction(const SelectionInstruction &insn) {
+ }
+
void GenContext::setA0Content(uint16_t new_a0[16], uint16_t max_offset, int sz) {
if (sz == 0)
sz = 8;
diff --git a/backend/src/backend/gen_context.hpp b/backend/src/backend/gen_context.hpp
index f050548..d1294fc 100644
--- a/backend/src/backend/gen_context.hpp
+++ b/backend/src/backend/gen_context.hpp
@@ -181,6 +181,7 @@ namespace gbe
virtual void emitI64MULInstruction(const SelectionInstruction &insn);
virtual void emitI64DIVREMInstruction(const SelectionInstruction &insn);
virtual void emitF64DIVInstruction(const SelectionInstruction &insn);
+ void emitWorkGroupOpInstruction(const SelectionInstruction &insn);
void scratchWrite(const GenRegister header, uint32_t offset, uint32_t reg_num, uint32_t reg_type, uint32_t channel_mode);
void scratchRead(const GenRegister dst, const GenRegister header, uint32_t offset, uint32_t reg_num, uint32_t reg_type, uint32_t channel_mode);
unsigned beforeMessage(const SelectionInstruction &insn, GenRegister bti, GenRegister flagTemp, GenRegister btiTmp, unsigned desc);
diff --git a/backend/src/backend/gen_insn_gen7_schedule_info.hxx b/backend/src/backend/gen_insn_gen7_schedule_info.hxx
index 792014f..49e328f 100644
--- a/backend/src/backend/gen_insn_gen7_schedule_info.hxx
+++ b/backend/src/backend/gen_insn_gen7_schedule_info.hxx
@@ -51,3 +51,4 @@ DECL_GEN7_SCHEDULE(I64MUL, 20, 40, 20)
DECL_GEN7_SCHEDULE(I64SATADD, 20, 40, 20)
DECL_GEN7_SCHEDULE(I64SATSUB, 20, 40, 20)
DECL_GEN7_SCHEDULE(F64DIV, 20, 40, 20)
+DECL_GEN7_SCHEDULE(WorkGroupOp, 80, 1, 1)
diff --git a/backend/src/backend/gen_insn_selection.cpp b/backend/src/backend/gen_insn_selection.cpp
index 6511f56..89adca4 100644
--- a/backend/src/backend/gen_insn_selection.cpp
+++ b/backend/src/backend/gen_insn_selection.cpp
@@ -697,6 +697,9 @@ namespace gbe
void I64REM(Reg dst, Reg src0, Reg src1, GenRegister *tmp, int tmp_int);
/*! double division */
void F64DIV(Reg dst, Reg src0, Reg src1, GenRegister* tmp, int tmpNum);
+ /*! Work Group Operations */
+ void WORKGROUP_OP(uint32_t wg_op, Reg dst, GenRegister src, GenRegister nextThreadID,
+ GenRegister threadID, GenRegister threadn, GenRegister tmp);
/* common functions for both binary instruction and sel_cmp and compare instruction.
It will handle the IMM or normal register assignment, and will try to avoid LOADI
as much as possible. */
@@ -2036,6 +2039,20 @@ namespace gbe
insn->dst(i + 1) = tmp[i];
}
+ void Selection::Opaque::WORKGROUP_OP(uint32_t wg_op, Reg dst, GenRegister src, GenRegister nextThreadID,
+ GenRegister threadID, GenRegister threadn, GenRegister tmp) {
+ SelectionInstruction *insn = this->appendInsn(SEL_OP_WORKGROUP_OP, 3, 4);
+ insn->extra.workgroupOp = wg_op;
+ insn->dst(0) = dst;
+ insn->dst(1) = nextThreadID;
+ insn->dst(2) = tmp;
+
+ insn->src(0) = src;
+ insn->src(1) = nextThreadID;
+ insn->src(2) = threadID;
+ insn->src(3) = threadn;
+ }
+
// Boiler plate to initialize the selection library at c++ pre-main
static SelectionLibrary *selLib = NULL;
static void destroySelectionLibrary(void) { GBE_DELETE(selLib); }
@@ -6553,9 +6570,26 @@ namespace gbe
/* Third, get the next thread ID which we will Forward MSG to. */
GenRegister nextThreadID = getNextThreadID(sel, slmAddr);
+ GenRegister threadID = sel.selReg(ocl::threadid, ir::TYPE_U32);
+ GenRegister threadNum = sel.selReg(ocl::threadn, ir::TYPE_U32);
+ GenRegister tmp = GenRegister::retype(sel.selReg(sel.reg(FAMILY_DWORD)), GEN_TYPE_UD);
+
+ const Type type = insn.getType();
+ const GenRegister dst = sel.selReg(insn.getDst(0), type);
+ const uint32_t srcNum = insn.getSrcNum();
+ GBE_ASSERT(srcNum == 3);
+ GBE_ASSERT(insn.getSrc(0) == ir::ocl::threadn);
+ GBE_ASSERT(insn.getSrc(1) == ir::ocl::threadid);
+ GenRegister src = sel.selReg(insn.getSrc(2), type);
+ sel.push(); {
+ sel.curr.flag = 0;
+ sel.curr.subFlag = 1;
+ sel.WORKGROUP_OP(workGroupOp, dst, src, nextThreadID, threadID, threadNum, tmp);
+ } sel.pop();
} else {
GBE_ASSERT(0);
}
+
return true;
}
DECL_CTOR(WorkGroupInstruction, 1, 1);
diff --git a/backend/src/backend/gen_insn_selection.hpp b/backend/src/backend/gen_insn_selection.hpp
index f51c905..9e4aee9 100644
--- a/backend/src/backend/gen_insn_selection.hpp
+++ b/backend/src/backend/gen_insn_selection.hpp
@@ -132,6 +132,7 @@ namespace gbe
uint32_t barrierType;
bool longjmp;
uint32_t indirect_offset;
+ uint32_t workgroupOp;
} extra;
/*! Gen opcode */
uint8_t opcode;
diff --git a/backend/src/backend/gen_insn_selection.hxx b/backend/src/backend/gen_insn_selection.hxx
index f6ed284..7c4991a 100644
--- a/backend/src/backend/gen_insn_selection.hxx
+++ b/backend/src/backend/gen_insn_selection.hxx
@@ -98,3 +98,4 @@ DECL_SELECTION_IR(ELSE, UnaryInstruction)
DECL_SELECTION_IR(READ_ARF, UnaryInstruction)
DECL_SELECTION_IR(WHILE, UnaryInstruction)
DECL_SELECTION_IR(F64DIV, F64DIVInstruction)
+DECL_SELECTION_IR(WORKGROUP_OP, WorkGroupOpInstruction)
--
2.5.0
More information about the Beignet
mailing list