it makes possible to switch send and sends within the encoder function.
v2: use GBE_ASSERT etc.
Signed-off-by: Guo, Yejun <yejun....@intel.com>
---
backend/src/backend/gen8_context.cpp | 14 ++++++-------
backend/src/backend/gen8_encoder.cpp | 2 +-
backend/src/backend/gen8_encoder.hpp | 2 +-
backend/src/backend/gen9_encoder.cpp | 22 ++++++++++++---------
backend/src/backend/gen9_encoder.hpp | 4 ++--
backend/src/backend/gen_context.cpp | 38 ++++++++++++++++--------------------
backend/src/backend/gen_encoder.cpp | 4 ++--
backend/src/backend/gen_encoder.hpp | 4 ++--
8 files changed, 44 insertions(+), 46 deletions(-)
diff --git a/backend/src/backend/gen8_context.cpp
b/backend/src/backend/gen8_context.cpp
index 95b1013..a3045ce 100644
--- a/backend/src/backend/gen8_context.cpp
+++ b/backend/src/backend/gen8_context.cpp
@@ -969,8 +969,6 @@ namespace gbe
const GenRegister addr = ra->genReg(insn.src(elemNum));
const GenRegister bti = ra->genReg(insn.src(elemNum*2+1));
GenRegister data = ra->genReg(insn.src(elemNum+1));
- if (!insn.extra.splitSend)
- data = addr;
/* Because BDW's store and load send instructions for 64 bits require the
bti to be surfaceless,
which we can not accept. We just fallback to 2 DW untypewrite here. */
@@ -981,7 +979,7 @@ namespace gbe
}
if (bti.file == GEN_IMMEDIATE_VALUE) {
- p->UNTYPED_WRITE(addr, data, bti, elemNum*2);
+ p->UNTYPED_WRITE(addr, data, bti, elemNum*2, insn.extra.splitSend);
} else {
const GenRegister tmp = ra->genReg(insn.dst(elemNum));
const GenRegister btiTmp = ra->genReg(insn.dst(elemNum + 1));
@@ -997,7 +995,7 @@ namespace gbe
p->push();
p->curr.predicate = GEN_PREDICATE_NORMAL;
p->curr.useFlag(insn.state.flag, insn.state.subFlag);
- p->UNTYPED_WRITE(addr, data, GenRegister::addr1(0), elemNum*2);
+ p->UNTYPED_WRITE(addr, data, GenRegister::addr1(0), elemNum*2,
insn.extra.splitSend);
p->pop();
afterMessage(insn, bti, tmp, btiTmp, jip0);
}
@@ -1358,7 +1356,7 @@ namespace gbe
nextDst = GenRegister::Qn(tempDst, 1);
p->MOV(nextDst, nextSrc);
p->pop();
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
p->push();
@@ -1374,7 +1372,7 @@ namespace gbe
nextDst = GenRegister::Qn(tempDst, 1);
p->MOV(nextDst, nextSrc);
p->pop();
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
}
@@ -1801,7 +1799,7 @@ namespace gbe
p->curr.execWidth = 8;
p->MUL(msgAddr, threadId, GenRegister::immd(0x8));
p->ADD(msgAddr, msgAddr, msgSlmOff);
- p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 2);
+ p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 2, false);
}
else
{
@@ -1809,7 +1807,7 @@ namespace gbe
p->MOV(msgData, threadData);
p->MUL(msgAddr, threadId, GenRegister::immd(0x4));
p->ADD(msgAddr, msgAddr, msgSlmOff);
- p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 1);
+ p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 1, false);
}
/* init partialData register, it will hold the final result */
diff --git a/backend/src/backend/gen8_encoder.cpp
b/backend/src/backend/gen8_encoder.cpp
index 8f73346..2928943 100644
--- a/backend/src/backend/gen8_encoder.cpp
+++ b/backend/src/backend/gen8_encoder.cpp
@@ -268,7 +268,7 @@ namespace gbe
return insn->bits3.ud;
}
- void Gen8Encoder::UNTYPED_WRITE(GenRegister msg, GenRegister data,
GenRegister bti, uint32_t elemNum) {
+ void Gen8Encoder::UNTYPED_WRITE(GenRegister msg, GenRegister data,
GenRegister bti, uint32_t elemNum, bool useSends) {
GenNativeInstruction *insn = this->next(GEN_OPCODE_SEND);
assert(elemNum >= 1 || elemNum <= 4);
this->setHeader(insn);
diff --git a/backend/src/backend/gen8_encoder.hpp
b/backend/src/backend/gen8_encoder.hpp
index f6a91a0..4afec0c 100644
--- a/backend/src/backend/gen8_encoder.hpp
+++ b/backend/src/backend/gen8_encoder.hpp
@@ -47,7 +47,7 @@ namespace gbe
virtual void ATOMIC(GenRegister dst, uint32_t function, GenRegister src,
GenRegister bti, uint32_t srcNum);
virtual void ATOMICA64(GenRegister dst, uint32_t function, GenRegister
src, GenRegister bti, uint32_t srcNum);
virtual void UNTYPED_READ(GenRegister dst, GenRegister src, GenRegister
bti, uint32_t elemNum);
- virtual void UNTYPED_WRITE(GenRegister src, GenRegister data, GenRegister
bti, uint32_t elemNum);
+ virtual void UNTYPED_WRITE(GenRegister src, GenRegister data, GenRegister
bti, uint32_t elemNum, bool useSends);
virtual void UNTYPED_READA64(GenRegister dst, GenRegister src, uint32_t
elemNum);
virtual void UNTYPED_WRITEA64(GenRegister src, uint32_t elemNum);
virtual void BYTE_GATHERA64(GenRegister dst, GenRegister src, uint32_t
elemSize);
diff --git a/backend/src/backend/gen9_encoder.cpp
b/backend/src/backend/gen9_encoder.cpp
index b5be852..47175f6 100644
--- a/backend/src/backend/gen9_encoder.cpp
+++ b/backend/src/backend/gen9_encoder.cpp
@@ -84,7 +84,7 @@ namespace gbe
else if (dst.file == GEN_GENERAL_REGISTER_FILE)
gen9_insn->bits1.sends.dest_reg_file_0 = 1;
else
- assert(!"should not reach here");
+ NOT_SUPPORTED;
gen9_insn->bits1.sends.src1_reg_file_0 = 1;
gen9_insn->bits1.sends.src1_reg_nr = src1.nr;
@@ -116,11 +116,13 @@ namespace gbe
return insn->bits3.ud;
}
- void Gen9Encoder::UNTYPED_WRITE(GenRegister addr, GenRegister data,
GenRegister bti, uint32_t elemNum)
+ void Gen9Encoder::UNTYPED_WRITE(GenRegister addr, GenRegister data,
GenRegister bti, uint32_t elemNum, bool useSends)
{
- if (addr.reg() == data.reg())
- Gen8Encoder::UNTYPED_WRITE(addr, data, bti, elemNum);
+ if (!useSends)
+ Gen8Encoder::UNTYPED_WRITE(addr, data, bti, elemNum, false);
else {
+ GBE_ASSERT(addr.reg() != data.reg());
+
GenNativeInstruction *insn = this->next(GEN_OPCODE_SENDS);
Gen9NativeInstruction *gen9_insn = &insn->gen9_insn;
assert(elemNum >= 1 || elemNum <= 4);
@@ -134,7 +136,7 @@ namespace gbe
else if (this->curr.execWidth == 16)
gen9_insn->bits2.sends.src1_length = 2 * elemNum;
else
- assert(!"unsupported");
+ NOT_SUPPORTED;
if (bti.file == GEN_IMMEDIATE_VALUE) {
gen9_insn->bits2.sends.sel_reg32_desc = 0;
@@ -164,11 +166,13 @@ namespace gbe
return insn->bits3.ud;
}
- void Gen9Encoder::BYTE_SCATTER(GenRegister addr, GenRegister data,
GenRegister bti, uint32_t elemSize)
+ void Gen9Encoder::BYTE_SCATTER(GenRegister addr, GenRegister data,
GenRegister bti, uint32_t elemSize, bool useSends)
{
- if (addr.reg() == data.reg())
- Gen8Encoder::BYTE_SCATTER(addr, data, bti, elemSize);
+ if (!useSends)
+ Gen8Encoder::BYTE_SCATTER(addr, data, bti, elemSize, false);
else {
+ GBE_ASSERT(addr.reg() != data.reg());
+
GenNativeInstruction *insn = this->next(GEN_OPCODE_SENDS);
Gen9NativeInstruction *gen9_insn = &insn->gen9_insn;
@@ -181,7 +185,7 @@ namespace gbe
else if (this->curr.execWidth == 16)
gen9_insn->bits2.sends.src1_length = 2;
else
- assert(!"unsupported");
+ NOT_SUPPORTED;
if (bti.file == GEN_IMMEDIATE_VALUE) {
gen9_insn->bits2.sends.sel_reg32_desc = 0;
diff --git a/backend/src/backend/gen9_encoder.hpp
b/backend/src/backend/gen9_encoder.hpp
index 1c40b92..4eb36e4 100644
--- a/backend/src/backend/gen9_encoder.hpp
+++ b/backend/src/backend/gen9_encoder.hpp
@@ -48,9 +48,9 @@ namespace gbe
bool isLD,
bool isUniform);
void setSendsOperands(Gen9NativeInstruction *gen9_insn, GenRegister dst,
GenRegister src0, GenRegister src1);
- virtual void UNTYPED_WRITE(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemNum);
+ virtual void UNTYPED_WRITE(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemNum, bool useSends);
virtual unsigned setUntypedWriteSendsMessageDesc(GenNativeInstruction
*insn, unsigned bti, unsigned elemNum);
- virtual void BYTE_SCATTER(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemSize);
+ virtual void BYTE_SCATTER(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemSize, bool useSends);
virtual unsigned setByteScatterSendsMessageDesc(GenNativeInstruction
*insn, unsigned bti, unsigned elemSize);
};
}
diff --git a/backend/src/backend/gen_context.cpp
b/backend/src/backend/gen_context.cpp
index 8288fa5..4341677 100644
--- a/backend/src/backend/gen_context.cpp
+++ b/backend/src/backend/gen_context.cpp
@@ -2146,7 +2146,7 @@ namespace gbe
const GenRegister bti = ra->genReg(insn.src(elemNum+1));
if (bti.file == GEN_IMMEDIATE_VALUE) {
- p->UNTYPED_WRITE(src, src, bti, elemNum*2);
+ p->UNTYPED_WRITE(src, src, bti, elemNum*2, false);
} else {
const GenRegister tmp = ra->genReg(insn.dst(0));
const GenRegister btiTmp = ra->genReg(insn.dst(1));
@@ -2158,7 +2158,7 @@ namespace gbe
p->push();
p->curr.predicate = GEN_PREDICATE_NORMAL;
p->curr.useFlag(insn.state.flag, insn.state.subFlag);
- p->UNTYPED_WRITE(src, src, GenRegister::addr1(0), elemNum*2);
+ p->UNTYPED_WRITE(src, src, GenRegister::addr1(0), elemNum*2, false);
p->pop();
afterMessage(insn, bti, tmp, btiTmp, jip0);
}
@@ -2167,12 +2167,10 @@ namespace gbe
void GenContext::emitUntypedWriteInstruction(const SelectionInstruction
&insn) {
const GenRegister addr = ra->genReg(insn.src(0));
GenRegister data = ra->genReg(insn.src(1));
- if (!insn.extra.splitSend)
- data = addr;
const uint32_t elemNum = insn.extra.elem;
const GenRegister bti = ra->genReg(insn.src(elemNum+1));
if (bti.file == GEN_IMMEDIATE_VALUE) {
- p->UNTYPED_WRITE(addr, data, bti, elemNum);
+ p->UNTYPED_WRITE(addr, data, bti, elemNum, insn.extra.splitSend);
} else {
const GenRegister tmp = ra->genReg(insn.dst(0));
const GenRegister btiTmp = ra->genReg(insn.dst(1));
@@ -2188,7 +2186,7 @@ namespace gbe
p->push();
p->curr.predicate = GEN_PREDICATE_NORMAL;
p->curr.useFlag(insn.state.flag, insn.state.subFlag);
- p->UNTYPED_WRITE(addr, data, GenRegister::addr1(0), elemNum);
+ p->UNTYPED_WRITE(addr, data, GenRegister::addr1(0), elemNum,
insn.extra.splitSend);
p->pop();
afterMessage(insn, bti, tmp, btiTmp, jip0);
}
@@ -2222,13 +2220,11 @@ namespace gbe
void GenContext::emitByteScatterInstruction(const SelectionInstruction
&insn) {
const GenRegister addr = ra->genReg(insn.src(0));
GenRegister data = ra->genReg(insn.src(1));
- if (!insn.extra.splitSend)
- data = addr;
const uint32_t elemSize = insn.extra.elem;
const GenRegister bti = ra->genReg(insn.src(2));
if (bti.file == GEN_IMMEDIATE_VALUE) {
- p->BYTE_SCATTER(addr, data, bti, elemSize);
+ p->BYTE_SCATTER(addr, data, bti, elemSize, insn.extra.splitSend);
} else {
const GenRegister tmp = ra->genReg(insn.dst(0));
const GenRegister btiTmp = ra->genReg(insn.dst(1));
@@ -2244,7 +2240,7 @@ namespace gbe
p->push();
p->curr.predicate = GEN_PREDICATE_NORMAL;
p->curr.useFlag(insn.state.flag, insn.state.subFlag);
- p->BYTE_SCATTER(addr, data, GenRegister::addr1(0), elemSize);
+ p->BYTE_SCATTER(addr, data, GenRegister::addr1(0), elemSize,
insn.extra.splitSend);
p->pop();
afterMessage(insn, bti, tmp, btiTmp, jip0);
}
@@ -2895,14 +2891,14 @@ namespace gbe
// Write it out.
p->curr.execWidth = 8;
p->curr.noMask = 1;
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(32));
// time stamps
for (int i = 0; i < 3; i++) {
p->curr.execWidth = 8;
p->MOV(data, GenRegister::retype(profilingReg[i], GEN_TYPE_UD));
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(32));
}
} p->pop();
@@ -3308,7 +3304,7 @@ namespace gbe
p->curr.execWidth = 8;
p->MUL(msgAddr, threadId, GenRegister::immd(0x8));
p->ADD(msgAddr, msgAddr, msgSlmOff);
- p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 2);
+ p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 2, false);
}
else
{
@@ -3316,7 +3312,7 @@ namespace gbe
p->MOV(msgData, threadData);
p->MUL(msgAddr, threadId, GenRegister::immd(0x4));
p->ADD(msgAddr, msgAddr, msgSlmOff);
- p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 1);
+ p->UNTYPED_WRITE(msg, msg, GenRegister::immw(0xFE), 1, false);
}
/* init partialData register, it will hold the final result */
@@ -3474,11 +3470,11 @@ namespace gbe
void GenContext::emitPrintfLongInstruction(GenRegister& addr, GenRegister&
data,
GenRegister& src, uint32_t bti) {
p->MOV(GenRegister::retype(data, GEN_TYPE_UD), src.bottom_half());
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
p->MOV(GenRegister::retype(data, GEN_TYPE_UD),
src.top_half(this->simdWidth));
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(bti), 1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
}
@@ -3503,15 +3499,15 @@ namespace gbe
p->ATOMIC(addr, GEN_ATOMIC_OP_ADD, addr,
GenRegister::immud(insn.extra.printfBTI), 2);
/* Write out the header. */
p->MOV(data, GenRegister::immud(0xAABBCCDD));
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
p->MOV(data, GenRegister::immud(insn.extra.printfSize + 12));
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
p->MOV(data, GenRegister::immud(insn.extra.printfNum));
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
}
@@ -3521,11 +3517,11 @@ namespace gbe
src = ra->genReg(insn.src(i));
if (src.type == GEN_TYPE_UD || src.type == GEN_TYPE_D || src.type ==
GEN_TYPE_F) {
p->MOV(GenRegister::retype(data, src.type), src);
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
} else if (src.type == GEN_TYPE_B || src.type == GEN_TYPE_UB ) {
p->MOV(GenRegister::retype(data, GEN_TYPE_UD), src);
- p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1);
+ p->UNTYPED_WRITE(addr, addr, GenRegister::immud(insn.extra.printfBTI),
1, false);
p->ADD(addr, addr, GenRegister::immud(sizeof(uint32_t)));
} else if (src.type == GEN_TYPE_L || src.type == GEN_TYPE_UL ) {
emitPrintfLongInstruction(addr, data, src, insn.extra.printfBTI);
diff --git a/backend/src/backend/gen_encoder.cpp
b/backend/src/backend/gen_encoder.cpp
index 49d93e8..9d23df3 100644
--- a/backend/src/backend/gen_encoder.cpp
+++ b/backend/src/backend/gen_encoder.cpp
@@ -433,7 +433,7 @@ namespace gbe
assert(0);
}
- void GenEncoder::UNTYPED_WRITE(GenRegister msg, GenRegister data,
GenRegister bti, uint32_t elemNum) {
+ void GenEncoder::UNTYPED_WRITE(GenRegister msg, GenRegister data,
GenRegister bti, uint32_t elemNum, bool useSends) {
GenNativeInstruction *insn = this->next(GEN_OPCODE_SEND);
assert(elemNum >= 1 || elemNum <= 4);
this->setHeader(insn);
@@ -534,7 +534,7 @@ namespace gbe
return insn->bits3.ud;
}
- void GenEncoder::BYTE_SCATTER(GenRegister msg, GenRegister data, GenRegister
bti, uint32_t elemSize) {
+ void GenEncoder::BYTE_SCATTER(GenRegister msg, GenRegister data, GenRegister
bti, uint32_t elemSize, bool useSends) {
GenNativeInstruction *insn = this->next(GEN_OPCODE_SEND);
this->setHeader(insn);
diff --git a/backend/src/backend/gen_encoder.hpp
b/backend/src/backend/gen_encoder.hpp
index e5eb2e2..31b6e92 100644
--- a/backend/src/backend/gen_encoder.hpp
+++ b/backend/src/backend/gen_encoder.hpp
@@ -177,7 +177,7 @@ namespace gbe
/*! Untyped read (upto 4 channels) */
virtual void UNTYPED_READ(GenRegister dst, GenRegister src, GenRegister
bti, uint32_t elemNum);
/*! Untyped write (upto 4 channels) */
- virtual void UNTYPED_WRITE(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemNum);
+ virtual void UNTYPED_WRITE(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemNum, bool useSends);
/*! Untyped read A64(upto 4 channels) */
virtual void UNTYPED_READA64(GenRegister dst, GenRegister src, uint32_t
elemNum);
/*! Untyped write (upto 4 channels) */
@@ -185,7 +185,7 @@ namespace gbe
/*! Byte gather (for unaligned bytes, shorts and ints) */
void BYTE_GATHER(GenRegister dst, GenRegister src, GenRegister bti,
uint32_t elemSize);
/*! Byte scatter (for unaligned bytes, shorts and ints) */
- virtual void BYTE_SCATTER(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemSize);
+ virtual void BYTE_SCATTER(GenRegister addr, GenRegister data, GenRegister
bti, uint32_t elemSize, bool useSends);
/*! Byte gather a64 (for unaligned bytes, shorts and ints) */
virtual void BYTE_GATHERA64(GenRegister dst, GenRegister src, uint32_t
elemSize);
/*! Byte scatter a64 (for unaligned bytes, shorts and ints) */
--
1.9.1
_______________________________________________
Beignet mailing list
Beignet@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/beignet
