[Beignet] [PATCH 7/7] refine pass debug info from llvm ir to gen insn
He Junyan
junyan.he at inbox.com
Thu Oct 8 07:28:54 PDT 2015
On Fri, Sep 18, 2015 at 05:01:57PM +0800, Bai Yannan wrote:
> Date: Fri, 18 Sep 2015 17:01:57 +0800
> From: Bai Yannan <yannan.bai at intel.com>
> To: beignet at lists.freedesktop.org
> Cc: Bai Yannan <yannan.bai at intel.com>, Lv Meng <meng.lv at intel.com>
> Subject: [Beignet] [PATCH 7/7] refine pass debug info from llvm ir to gen
> insn
> X-Mailer: git-send-email 1.9.1
>
> Add line and col to ctx to pass debug infomation
>
> Signed-off-by: Bai Yannan <yannan.bai at intel.com>
> Signed-off-by: Lv Meng <meng.lv at intel.com>
> ---
> backend/src/backend/gen_context.cpp | 93 +++++++++++-----------
> backend/src/backend/gen_encoder.cpp | 24 +++---
> backend/src/backend/gen_encoder.hpp | 7 +-
> backend/src/ir/context.cpp | 2 +
> backend/src/ir/context.hpp | 3 +
> backend/src/llvm/llvm_gen_backend.cpp | 142 ++++++----------------------------
> backend/src/llvm/llvm_to_gen.cpp | 4 +-
> 7 files changed, 93 insertions(+), 182 deletions(-)
>
> diff --git a/backend/src/backend/gen_context.cpp b/backend/src/backend/gen_context.cpp
> index 43c0b25..18a02a0 100644
> --- a/backend/src/backend/gen_context.cpp
> +++ b/backend/src/backend/gen_context.cpp
> @@ -92,11 +92,6 @@ namespace gbe
> }
>
> #define OCL_PROFILING (bool)(getenv("OCL_PROFILING")[0]-48)
> -#define SET_GENINSN_DBGINFO(INSN) \
> - if(INSN.dbginfo.hasdbginfo) \
> - p->setDbginfo(INSN.dbginfo.line,INSN.dbginfo.col);\
> - else p->setDbginfo(0,0)
> -
> void GenContext::emitInstructionStream(void) {
> // Emit Gen ISA
> for (auto &block : *sel->blockList)
> @@ -106,6 +101,9 @@ namespace gbe
> // no more virtual register here in that part of the code generation
> GBE_ASSERT(insn.state.physicalFlag);
> p->curr = insn.state;
> + //meng
> + p->line = insn.dbginfo.line;
> + p->col = insn.dbginfo.col;
> switch (opcode) {
> #define DECL_SELECTION_IR(OPCODE, FAMILY) \
> case SEL_OP_##OPCODE: this->emit##FAMILY(insn); break;
> @@ -119,7 +117,6 @@ namespace gbe
> instruction prefetcher prefetch into an invalide page */
> for(int i = 0; i < 8; i++)
> p->NOP();
> - p->setDbginfo(0,0);
> }
>
> bool GenContext::patchBranches(void) {
> @@ -255,7 +252,7 @@ namespace gbe
> void GenContext::emitLabelInstruction(const SelectionInstruction &insn) {
> const ir::LabelIndex label(insn.index);
> this->labelPos.insert(std::make_pair(label, p->store.size()));
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
it seems to be a tmp comment, and should not be in the patches.
> }
>
> void GenContext::emitUnaryInstruction(const SelectionInstruction &insn) {
> @@ -330,7 +327,7 @@ namespace gbe
> break;
> default: NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitUnaryWithTempInstruction(const SelectionInstruction &insn) {
> @@ -489,7 +486,7 @@ namespace gbe
> default:
> NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitBinaryWithTempInstruction(const SelectionInstruction &insn) {
> @@ -590,7 +587,7 @@ namespace gbe
> default:
> NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitSimdShuffleInstruction(const SelectionInstruction &insn) {
> @@ -682,7 +679,7 @@ namespace gbe
> break;
> default: NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::collectShifter(GenRegister dest, GenRegister src) {
> @@ -777,7 +774,7 @@ namespace gbe
> }
> storeTopHalf(dest, e);
> storeBottomHalf(dest, f);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64MADSATInstruction(const SelectionInstruction &insn) {
> @@ -909,7 +906,7 @@ namespace gbe
> }
> storeTopHalf(dest, g);
> storeBottomHalf(dest, h);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64HADDInstruction(const SelectionInstruction &insn) {
> @@ -937,7 +934,7 @@ namespace gbe
> p->OR(c, c, d);
> storeBottomHalf(dest, a);
> storeTopHalf(dest, c);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64RHADDInstruction(const SelectionInstruction &insn) {
> @@ -968,7 +965,7 @@ namespace gbe
> p->OR(c, c, d);
> storeBottomHalf(dest, a);
> storeTopHalf(dest, c);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64ShiftInstruction(const SelectionInstruction &insn) {
> @@ -1075,7 +1072,7 @@ namespace gbe
> default:
> NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
> void GenContext::setFlag(GenRegister flagReg, GenRegister src) {
> p->push();
> @@ -1231,7 +1228,7 @@ namespace gbe
> p->OR(dest, dest, GenRegister::immud(0x80000000));
> p->pop();
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
>
> @@ -1271,7 +1268,7 @@ namespace gbe
> }
> storeTopHalf(dst, high);
> storeBottomHalf(dst, low);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64CompareInstruction(const SelectionInstruction &insn) {
> @@ -1352,7 +1349,7 @@ namespace gbe
> p->curr.execWidth = 1;
> p->MOV(GenRegister::flag(flag, subFlag), f1);
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64SATADDInstruction(const SelectionInstruction &insn) {
> @@ -1402,7 +1399,7 @@ namespace gbe
> p->pop();
> storeTopHalf(dst, a);
> storeBottomHalf(dst, b);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64SATSUBInstruction(const SelectionInstruction &insn) {
> @@ -1451,7 +1448,7 @@ namespace gbe
> p->pop();
> storeTopHalf(dst, a);
> storeBottomHalf(dst, b);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::loadTopHalf(GenRegister dest, GenRegister src) {
> @@ -1549,7 +1546,7 @@ namespace gbe
> p->pop();
> storeTopHalf(dest, e);
> storeBottomHalf(dest, a);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitI64DIVREMInstruction(const SelectionInstruction &insn) {
> @@ -1703,7 +1700,7 @@ namespace gbe
> storeTopHalf(dest, a);
> storeBottomHalf(dest, b);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitTernaryInstruction(const SelectionInstruction &insn) {
> @@ -1715,17 +1712,17 @@ namespace gbe
> case SEL_OP_MAD: p->MAD(dst, src0, src1, src2); break;
> default: NOT_IMPLEMENTED;
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitNoOpInstruction(const SelectionInstruction &insn) {
> p->NOP();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitWaitInstruction(const SelectionInstruction &insn) {
> p->WAIT();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitBarrierInstruction(const SelectionInstruction &insn) {
> @@ -1752,14 +1749,14 @@ namespace gbe
> // Now we wait for the other threads
> p->WAIT();
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitFenceInstruction(const SelectionInstruction &insn) {
> const GenRegister dst = ra->genReg(insn.dst(0));
> p->FENCE(dst);
> p->MOV(dst, dst);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitMathInstruction(const SelectionInstruction &insn) {
> @@ -1771,7 +1768,7 @@ namespace gbe
> p->MATH(dst, function, src0, src1);
> } else
> p->MATH(dst, function, src0);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitCompareInstruction(const SelectionInstruction &insn) {
> @@ -1785,7 +1782,7 @@ namespace gbe
> const GenRegister dst = ra->genReg(insn.dst(0));
> p->SEL_CMP(insn.extra.function, dst, src0, src1);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitAtomicInstruction(const SelectionInstruction &insn) {
> @@ -1811,7 +1808,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, flagTemp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitIndirectMoveInstruction(const SelectionInstruction &insn) {
> @@ -1855,7 +1852,7 @@ namespace gbe
> p->MOV(nextDst, indirect_src);
> p->pop();
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::insertJumpPos(const SelectionInstruction &insn) {
> @@ -1867,7 +1864,7 @@ namespace gbe
> insertJumpPos(insn);
> const GenRegister src = ra->genReg(insn.src(0));
> p->JMPI(src, insn.extra.longjmp);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitEotInstruction(const SelectionInstruction &insn) {
> @@ -1878,7 +1875,7 @@ namespace gbe
> p->curr.execWidth = 8;
> p->EOT(112);
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitSpillRegInstruction(const SelectionInstruction &insn) {
> @@ -1915,7 +1912,7 @@ namespace gbe
> this->scratchWrite(msg, scratchOffset + 4*simdWidth, regNum, GEN_TYPE_UD, GEN_SCRATCH_CHANNEL_MODE_DWORD);
> }
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitUnSpillRegInstruction(const SelectionInstruction &insn) {
> @@ -1944,7 +1941,7 @@ namespace gbe
> storeTopHalf(dst, payload);
> }
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitRead64Instruction(const SelectionInstruction &insn) {
> @@ -1969,7 +1966,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
> unsigned GenContext::beforeMessage(const SelectionInstruction &insn, GenRegister bti, GenRegister tmp, unsigned desc) {
> const GenRegister flagReg = GenRegister::flag(insn.state.flag, insn.state.subFlag);
> @@ -2038,7 +2035,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitWrite64Instruction(const SelectionInstruction &insn) {
> @@ -2062,7 +2059,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitUntypedWriteInstruction(const SelectionInstruction &insn) {
> @@ -2085,7 +2082,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitByteGatherInstruction(const SelectionInstruction &insn) {
> @@ -2110,7 +2107,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitByteScatterInstruction(const SelectionInstruction &insn) {
> @@ -2134,7 +2131,7 @@ namespace gbe
> p->pop();
> afterMessage(insn, bti, tmp, jip0);
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
>
> }
>
> @@ -2143,7 +2140,7 @@ namespace gbe
> for(uint32_t i = 0; i < insn.dstNum; i++) {
> p->MOV(ra->genReg(insn.dst(i)), GenRegister::splitReg(src, insn.extra.elem, i));
> }
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitPackByteInstruction(const SelectionInstruction &insn) {
> @@ -2166,7 +2163,7 @@ namespace gbe
> }
> }
> p->pop();
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitUnpackLongInstruction(const SelectionInstruction &insn) {
> @@ -2182,7 +2179,7 @@ namespace gbe
> const GenRegister src = ra->genReg(insn.src(0));
> const uint32_t bti = insn.getbti();
> p->DWORD_GATHER(dst, src, bti);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::emitSampleInstruction(const SelectionInstruction &insn) {
> @@ -2193,7 +2190,7 @@ namespace gbe
> const unsigned int msgLen = insn.extra.rdmsglen;
> uint32_t simdWidth = p->curr.execWidth;
> p->SAMPLE(dst, msgPayload, msgLen, false, bti, sampler, simdWidth, -1, 0, insn.extra.isLD, insn.extra.isUniform);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::scratchWrite(const GenRegister header, uint32_t offset, uint32_t reg_num, uint32_t reg_type, uint32_t channel_mode) {
> @@ -2231,7 +2228,7 @@ namespace gbe
> const GenRegister header = GenRegister::retype(ra->genReg(insn.src(0)), GEN_TYPE_UD);
> const uint32_t bti = insn.getbti();
> p->TYPED_WRITE(header, true, bti);
> - SET_GENINSN_DBGINFO(insn);
> + //SET_GENINSN_DBGINFO(insn);
> }
>
> void GenContext::setA0Content(uint16_t new_a0[16], uint16_t max_offset, int sz) {
> @@ -2370,6 +2367,8 @@ namespace gbe
> std::memcpy(genKernel->insns, &p->store[0], genKernel->insnNum * sizeof(GenInstruction));
> if (OCL_OUTPUT_ASM)
> outputAssembly(stdout, genKernel);
> + if (OCL_PROFILING)
> + outputAssembly(stdout, genKernel);
>
> if (this->asmFileName) {
> FILE *asmDumpStream = fopen(this->asmFileName, "a");
> diff --git a/backend/src/backend/gen_encoder.cpp b/backend/src/backend/gen_encoder.cpp
> index 0f50c35..8a19cb9 100644
> --- a/backend/src/backend/gen_encoder.cpp
> +++ b/backend/src/backend/gen_encoder.cpp
> @@ -50,7 +50,6 @@
>
> #include "backend/gen_encoder.hpp"
> #include <cstring>
> -#include <map>
>
>
> namespace gbe
> @@ -598,6 +597,11 @@ namespace gbe
> std::memset(&insn, 0, sizeof(GenCompactInstruction));
> insn.bits1.opcode = opcode;
> this->store.push_back(insn.low);
> + //meng
> + GenInsnDBGInfo dginfo;
> + dginfo.srcline = line;
> + dginfo.srccol = col;
> + storedbg.push_back(dginfo);
> return (GenCompactInstruction *)&this->store.back();
> }
>
> @@ -607,6 +611,12 @@ namespace gbe
> insn.header.opcode = opcode;
> this->store.push_back(insn.low);
> this->store.push_back(insn.high);
> + //meng
> + GenInsnDBGInfo dginfo;
> + dginfo.srcline = line;
> + dginfo.srccol = col;
> + storedbg.push_back(dginfo);
> + storedbg.push_back(dginfo);
> return (GenNativeInstruction *)(&this->store.back()-1);
> }
>
> @@ -1299,17 +1309,5 @@ namespace gbe
> insn->header.destreg_or_condmod = GEN_SFID_THREAD_SPAWNER;
> }
>
> - void GenEncoder::setDbginfo(uint32_t line,uint32_t col) {
> - GenInsnDBGInfo gd;
> - for(int i = 0; i<this->store.size(); i++)
> - {
> - auto it = this->storedbg.find(i);
> - if(it != this->storedbg.end())
> - continue;
> - gd.srcline = line;
> - gd.srccol = col;
> - storedbg[i] = gd;
> - }
> - }
> } /* namespace gbe */
>
> diff --git a/backend/src/backend/gen_encoder.hpp b/backend/src/backend/gen_encoder.hpp
> index 9797abc..d727720 100644
> --- a/backend/src/backend/gen_encoder.hpp
> +++ b/backend/src/backend/gen_encoder.hpp
> @@ -57,7 +57,6 @@
> #include "sys/vector.hpp"
> #include <cassert>
> #include "src/cl_device_data.h"
> -#include <map>
>
> namespace gbe
> {
> @@ -89,6 +88,9 @@ namespace gbe
> uint32_t deviceID;
> /*! simd width for this codegen */
> uint32_t simdWidth;
> + //meng
> + uint32_t line;
> + uint32_t col;
> ////////////////////////////////////////////////////////////////////////
> // Encoding functions
> ////////////////////////////////////////////////////////////////////////
> @@ -258,8 +260,7 @@ namespace gbe
> virtual void alu3(uint32_t opcode, GenRegister dst,
> GenRegister src0, GenRegister src1, GenRegister src2) = 0;
> public:
> - void setDbginfo(uint32_t line, uint32_t col);
> - std::map<int, GenInsnDBGInfo> storedbg;
> + std::vector<GenInsnDBGInfo> storedbg;
> };
>
> void alu1(GenEncoder *p, uint32_t opcode, GenRegister dst,
> diff --git a/backend/src/ir/context.cpp b/backend/src/ir/context.cpp
> index 2412fe9..8e18be3 100644
> --- a/backend/src/ir/context.cpp
> +++ b/backend/src/ir/context.cpp
> @@ -159,6 +159,8 @@ namespace ir {
> // Append the instruction in the stream
> Instruction *insnPtr = fn->newInstruction(insn);
> bb->append(*insnPtr);
> + //meng
> + insnPtr->setDBGInfo(line,col);
> #if GBE_DEBUG
> std::string whyNot;
> if(getUnit().getValid())
> diff --git a/backend/src/ir/context.hpp b/backend/src/ir/context.hpp
> index 54265d0..c38c99f 100644
> --- a/backend/src/ir/context.hpp
> +++ b/backend/src/ir/context.hpp
> @@ -209,6 +209,9 @@ namespace ir {
> }
> void appendSurface(uint8_t bti, Register reg) { fn->appendSurface(bti, reg); }
>
> + //meng
> + uint32_t line;
> + uint32_t col;
> protected:
> /*! A block must be started with a label */
> void startBlock(void);
> diff --git a/backend/src/llvm/llvm_gen_backend.cpp b/backend/src/llvm/llvm_gen_backend.cpp
> index e3b6b13..b426f82 100644
> --- a/backend/src/llvm/llvm_gen_backend.cpp
> +++ b/backend/src/llvm/llvm_gen_backend.cpp
> @@ -202,11 +202,7 @@ using namespace llvm;
> if(OCL_PROFILING){ \
> llvm::DebugLoc dg; \
> GET_INSN_DBGLOC##FLAG(I); \
> - gbe::ir::BasicBlock * bb = ctx.getBlock(); \
> - if(bb)\
> - for(gbe::ir::Instruction *in = bb->getFirstInstruction(); in!=bb->getLastInstruction(); in=bb->getSuccessorInstruction(in) )\
> - if(!in->DBGInfo.hasDBGInfo) in->setDBGInfo(dg.getLine(), dg.getCol());\
> - }
> + ctx.line = dg.getLine(), ctx.col = dg.getCol();}
> // end define SET_GENIR_DBGINFO
> #define GET_PRE_LOADI_DBGINFO \
> if(OCL_PROFILING) { \
> @@ -1869,6 +1865,7 @@ namespace gbe
>
> void GenWriter::emitMovForPHI(BasicBlock *curr, BasicBlock *succ) {
> for (BasicBlock::iterator I = succ->begin(); isa<PHINode>(I); ++I) {
> + SET_GENIR_DBGINFO(I,2);
> PHINode *PN = cast<PHINode>(I);
> Value *IV = PN->getIncomingValueForBlock(curr);
> Type *llvmType = PN->getType();
> @@ -1888,11 +1885,9 @@ namespace gbe
> isa<UndefValue>(extractConstantElem(CPV, 0)))
> continue;
> ctx.MOV(type, dst, getRegister(CP));
> - SET_GENIR_DBGINFO(PN,2);
> } else if (regTranslator.valueExists(IV,0) || dyn_cast<Constant>(IV)) {
> const ir::Register src = this->getRegister(IV);
> ctx.MOV(type, dst, src);
> - SET_GENIR_DBGINFO(PN,2);
> }
> assert(!ctx.getBlock()->undefPhiRegs.contains(dst));
> ctx.getBlock()->definedPhiRegs.insert(dst);
> @@ -2827,6 +2822,7 @@ namespace gbe
> void GenWriter::regAllocateReturnInst(ReturnInst &I) {}
>
> void GenWriter::emitReturnInst(ReturnInst &I) {
> + SET_GENIR_DBGINFO(I,1);
> const ir::Function &fn = ctx.getFunction();
> GBE_ASSERTM(fn.outputNum() <= 1, "no more than one value can be returned");
> if (fn.outputNum() == 1 && I.getNumOperands() > 0) {
> @@ -2834,10 +2830,8 @@ namespace gbe
> const ir::Register src = this->getRegister(I.getOperand(0));
> const ir::RegisterFamily family = fn.getRegisterFamily(dst);
> ctx.MOV(ir::getType(family), dst, src);
> - SET_GENIR_DBGINFO(I,1);
> }
> ctx.RET();
> - //SET_GENIR_DBGINFO(I, 1);
> }
>
> void GenWriter::regAllocateBinaryOperator(Instruction &I) {
> @@ -2865,7 +2859,7 @@ namespace gbe
> const ir::Register dst = this->getRegister(&I);
> const ir::Register src0 = this->getRegister(I.getOperand(0));
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> - bool isSupported = true;
> + SET_GENIR_DBGINFO(I,1);
>
> switch (I.getOpcode()) {
> case Instruction::Add:
> @@ -2886,10 +2880,8 @@ namespace gbe
> case Instruction::Shl: ctx.SHL(type, dst, src0, src1); break;
> case Instruction::LShr: ctx.SHR(getUnsignedType(ctx, I.getType()), dst, src0, src1); break;
> case Instruction::AShr: ctx.ASR(type, dst, src0, src1); break;
> - default: isSupported = false; NOT_SUPPORTED;
> + default: NOT_SUPPORTED;
> }
> - if(isSupported)
> - SET_GENIR_DBGINFO(I,1);
> }
>
> void GenWriter::regAllocateICmpInst(ICmpInst &I) {
> @@ -2925,7 +2917,7 @@ namespace gbe
> const ir::Register src1 = this->getRegister(I.getOperand(1));
>
> // We must invert the condition to simplify the branch code
> - bool isSupported = true;
> + SET_GENIR_DBGINFO(I,1);
> if (conditionSet.find(&I) != conditionSet.end()) {
> switch (I.getPredicate()) {
> case ICmpInst::ICMP_EQ: ctx.NE(type, dst, src0, src1); break;
> @@ -2938,7 +2930,7 @@ namespace gbe
> case ICmpInst::ICMP_SLT: ctx.GE(signedType, dst, src0, src1); break;
> case ICmpInst::ICMP_UGT: ctx.LE(unsignedType, dst, src0, src1); break;
> case ICmpInst::ICMP_SGT: ctx.LE(signedType, dst, src0, src1); break;
> - default: isSupported = false; NOT_SUPPORTED;
> + default: NOT_SUPPORTED;
> }
> }
> // Nothing special to do
> @@ -2954,11 +2946,9 @@ namespace gbe
> case ICmpInst::ICMP_SLT: ctx.LT(signedType, dst, src0, src1); break;
> case ICmpInst::ICMP_UGT: ctx.GT(unsignedType, dst, src0, src1); break;
> case ICmpInst::ICMP_SGT: ctx.GT(signedType, dst, src0, src1); break;
> - default: isSupported = false;NOT_SUPPORTED;
> + default: NOT_SUPPORTED;
> }
> }
> - if(isSupported)
> - SET_GENIR_DBGINFO(I,1);
> }
>
> void GenWriter::regAllocateFCmpInst(FCmpInst &I) {
> @@ -2978,8 +2968,8 @@ namespace gbe
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> const ir::Register tmp = ctx.reg(getFamily(ctx, I.getType()));
> Value *cv = ConstantInt::get(I.getType(), 1);
> - bool isSupported = true;
>
> + SET_GENIR_DBGINFO(I,1);
> switch (I.getPredicate()) {
> case ICmpInst::FCMP_OEQ: ctx.EQ(type, dst, src0, src1); break;
> case ICmpInst::FCMP_ONE: ctx.NE(type, dst, src0, src1); break;
> @@ -3005,47 +2995,38 @@ namespace gbe
> ctx.NE(type, dst, src0, src0);
> else {
> ctx.ORD(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv)); //TODO: Use NOT directly
> }
> break;
> case ICmpInst::FCMP_UEQ:
> ctx.NE(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_UGT:
> ctx.LE(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_UGE:
> ctx.LT(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_ULT:
> ctx.GE(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_ULE:
> ctx.GT(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_UNE:
> ctx.EQ(type, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.XOR(insnType, dst, tmp, getRegister(cv));
> break;
> case ICmpInst::FCMP_TRUE:
> ctx.MOV(insnType, dst, getRegister(cv));
> break;
> - default: isSupported = false; NOT_SUPPORTED;
> + default: NOT_SUPPORTED;
> }
> - if(isSupported)
> - SET_GENIR_DBGINFO(I,1);
> }
>
> void GenWriter::regAllocateCastInst(CastInst &I) {
> @@ -3100,6 +3081,7 @@ namespace gbe
> }
>
> void GenWriter::emitCastInst(CastInst &I) {
> + SET_GENIR_DBGINFO(I,1);
> switch (I.getOpcode())
> {
> case Instruction::PtrToInt:
> @@ -3114,7 +3096,6 @@ namespace gbe
> const ir::Register dst = this->getRegister(&I);
> const ir::Register src = this->getRegister(srcValue);
> ctx.CVT(getType(ctx, dstType), getType(ctx, srcType), dst, src);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
> break;
> @@ -3150,7 +3131,6 @@ namespace gbe
> const ir::Tuple dstTuple = ctx.arrayTuple(&dstTupleData[0], dstElemNum);
>
> ctx.BITCAST(dstType, srcType, dstTuple, srcTuple, dstElemNum, srcElemNum);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
> break; // nothing to emit here
> @@ -3192,9 +3172,7 @@ namespace gbe
> const ir::Register zeroReg = ctx.reg(family);
> const ir::Register oneReg = ctx.reg(family);
> ctx.LOADI(dstType, zeroReg, zero);
> - SET_GENIR_DBGINFO(I,1);
> ctx.LOADI(dstType, oneReg, one);
> - SET_GENIR_DBGINFO(I,1);
> const ir::Register dst = this->getRegister(&I);
> const ir::Register src = this->getRegister(I.getOperand(0));
> ctx.SEL(dstType, dst, src, oneReg, zeroReg);
> @@ -3211,7 +3189,6 @@ namespace gbe
> const ir::Register src = this->getRegister(I.getOperand(0));
> ctx.CVT(dstType, srcType, dst, src);
> }
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> default: NOT_SUPPORTED;
> @@ -3277,8 +3254,8 @@ namespace gbe
> const ir::Register cond = this->getRegister(I.getOperand(0));
> const ir::Register src0 = this->getRegister(I.getOperand(1));
> const ir::Register src1 = this->getRegister(I.getOperand(2));
> - ctx.SEL(type, dst, cond, src0, src1);
> SET_GENIR_DBGINFO(I,1);
> + ctx.SEL(type, dst, cond, src0, src1);
> }
>
> void GenWriter::regAllocatePHINode(PHINode &I) {
> @@ -3295,8 +3272,8 @@ namespace gbe
>
> const ir::Register dst = this->getRegister(&I);
> const ir::Register src = this->getRegister(copy);
> - ctx.MOV(type, dst, src);
> SET_GENIR_DBGINFO(I,1);
> + ctx.MOV(type, dst, src);
> phiMap.insert(std::make_pair(dst, src));
> }
>
> @@ -3311,13 +3288,13 @@ namespace gbe
>
> // Inconditional branch. Just check that we jump to a block which is not our
> // successor
> + SET_GENIR_DBGINFO(I,1);
> if (I.isConditional() == false) {
> BasicBlock *target = I.getSuccessor(0);
> if (std::next(Function::iterator(bb)) != Function::iterator(target)) {
> GBE_ASSERT(labelMap.find(target) != labelMap.end());
> const ir::LabelIndex labelIndex = labelMap[target];
> ctx.BRA(labelIndex);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
> // The LLVM branch has two targets
> @@ -3335,7 +3312,6 @@ namespace gbe
> const ir::LabelIndex index = labelMap[taken];
> const ir::Register reg = this->getRegister(condition);
> ctx.BRA(index, reg);
> - SET_GENIR_DBGINFO(I,1);
>
> // If non-taken target is the next block, there is nothing to do
> BasicBlock *bb = I.getParent();
> @@ -3347,9 +3323,7 @@ namespace gbe
> GBE_ASSERT(labelMap.find(nonTaken) != labelMap.end());
> const ir::LabelIndex untakenIndex = ctx.label();
> ctx.LABEL(untakenIndex);
> - SET_GENIR_DBGINFO(I,1);
> ctx.BRA(labelMap[nonTaken]);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
>
> @@ -3610,8 +3584,8 @@ namespace gbe
> GBE_ASSERT(AI != AE);
> const ir::Register src = this->getRegister(*AI);
> const ir::Register dst = this->getRegister(&I);
> - ctx.ALU1(opcode, type, dst, src);
> SET_GENIR_DBGINFO(I,1);
> + ctx.ALU1(opcode, type, dst, src);
> }
>
> void GenWriter::emitAtomicInst(CallInst &I, CallSite &CS, ir::AtomicOps opcode) {
> @@ -3629,6 +3603,7 @@ namespace gbe
>
> ir::Register btiReg;
> bool fixedBTI = false;
> + SET_GENIR_DBGINFO(I,1);
> if (isa<ConstantInt>(bti)) {
> fixedBTI = true;
> unsigned index = cast<ConstantInt>(bti)->getZExtValue();
> @@ -3636,7 +3611,6 @@ namespace gbe
> ir::ImmediateIndex immIndex = ctx.newImmediate((uint32_t)index);
> btiReg = ctx.reg(ir::FAMILY_DWORD);
> ctx.LOADI(ir::TYPE_U32, btiReg, immIndex);
> - SET_GENIR_DBGINFO(I,1);
> } else {
> addrSpace = ir::MEM_MIXED;
> btiReg = this->getRegister(bti);
> @@ -3645,7 +3619,6 @@ namespace gbe
> const ir::RegisterFamily pointerFamily = ctx.getPointerFamily();
> const ir::Register ptr = ctx.reg(pointerFamily);
> ctx.SUB(ir::TYPE_U32, ptr, pointer, baseReg);
> - SET_GENIR_DBGINFO(I,1);
>
> const ir::Register dst = this->getRegister(&I);
>
> @@ -3661,7 +3634,6 @@ namespace gbe
> }
> const ir::Tuple srcTuple = ctx.arrayTuple(&src[0], srcNum);
> ctx.ATOMIC(opcode, dst, addrSpace, btiReg, fixedBTI, srcTuple);
> - SET_GENIR_DBGINFO(I,1);
> }
>
> /* append a new sampler. should be called before any reference to
> @@ -3691,6 +3663,7 @@ namespace gbe
>
> void GenWriter::emitCallInst(CallInst &I) {
> if (Function *F = I.getCalledFunction()) {
> + SET_GENIR_DBGINFO(I,1);
> if (F->getIntrinsicID() != 0) {
> const ir::Function &fn = ctx.getFunction();
>
> @@ -3707,7 +3680,6 @@ namespace gbe
> const ir::Register src = ir::ocl::stackptr;
> const ir::RegisterFamily family = fn.getRegisterFamily(dst);
> ctx.MOV(ir::getType(family), dst, src);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> case Intrinsic::stackrestore:
> @@ -3716,7 +3688,6 @@ namespace gbe
> const ir::Register src = this->getRegister(I.getOperand(0));
> const ir::RegisterFamily family = fn.getRegisterFamily(dst);
> ctx.MOV(ir::getType(family), dst, src);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> #if LLVM_VERSION_MINOR >= 2
> @@ -3728,9 +3699,7 @@ namespace gbe
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> const ir::Register src2 = this->getRegister(I.getOperand(2));
> ctx.MUL(ir::TYPE_FLOAT, tmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.ADD(ir::TYPE_FLOAT, dst, tmp, src2);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> case Intrinsic::lifetime_start:
> @@ -3750,12 +3719,10 @@ namespace gbe
> const ir::Register src0 = this->getRegister(I.getOperand(0));
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> ctx.ADD(dst0Type, dst0, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
>
> ir::Register overflow = this->getRegister(&I, 1);
> const ir::Type unsignedType = makeTypeUnsigned(dst0Type);
> ctx.LT(unsignedType, overflow, dst0, src1);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> case Intrinsic::usub_with_overflow:
> @@ -3767,12 +3734,10 @@ namespace gbe
> const ir::Register src0 = this->getRegister(I.getOperand(0));
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> ctx.SUB(dst0Type, dst0, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
>
> ir::Register overflow = this->getRegister(&I, 1);
> const ir::Type unsignedType = makeTypeUnsigned(dst0Type);
> ctx.GT(unsignedType, overflow, dst0, src0);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> case Intrinsic::sadd_with_overflow:
> @@ -3807,24 +3772,18 @@ namespace gbe
> const ir::RegisterFamily family = getFamily(tmpType);
> const ir::Register immReg = ctx.reg(family);
> ctx.LOADI(ir::TYPE_S32, immReg, imm);
> - SET_GENIR_DBGINFO(I,1);
>
> ir::Register tmp0 = ctx.reg(getFamily(tmpType));
> ir::Register tmp1 = ctx.reg(getFamily(tmpType));
> ir::Register tmp2 = ctx.reg(getFamily(tmpType));
> ctx.CVT(tmpType, srcType, tmp0, src);
> - SET_GENIR_DBGINFO(I,1);
> ctx.ALU1(ir::OP_LZD, tmpType, tmp1, tmp0);
> - SET_GENIR_DBGINFO(I,1);
> ctx.SUB(tmpType, tmp2, tmp1, immReg);
> - SET_GENIR_DBGINFO(I,1);
> ctx.CVT(dstType, tmpType, dst, tmp2);
> - SET_GENIR_DBGINFO(I,1);
> }
> else
> {
> ctx.ALU1(ir::OP_LZD, dstType, dst, src);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
> break;
> @@ -3836,7 +3795,6 @@ namespace gbe
> const ir::Register src1 = this->getRegister(I.getOperand(1));
> const ir::Register src2 = this->getRegister(I.getOperand(2));
> ctx.MAD(srcType, dst, src0, src1, src2);
> - SET_GENIR_DBGINFO(I,1);
> }
> break;
> case Intrinsic::sqrt: this->emitUnaryCallInst(I,CS,ir::OP_SQR); break;
> @@ -3857,7 +3815,6 @@ namespace gbe
> const ir::Register src1 = this->getRegister(*AI);
> const ir::Register dst = this->getRegister(&I);
> ctx.POW(ir::TYPE_FLOAT, dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> default: NOT_IMPLEMENTED;
> @@ -3884,7 +3841,6 @@ namespace gbe
> const ir::Register src = this->getRegister(*AI);
> const ir::Register dst = this->getRegister(&I);
> ctx.ALU1(ir::OP_ABS, getType(ctx, (*AI)->getType()), dst, src);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SIMD_ALL:
> @@ -3892,7 +3848,6 @@ namespace gbe
> const ir::Register src = this->getRegister(*AI);
> const ir::Register dst = this->getRegister(&I);
> ctx.ALU1(ir::OP_SIMD_ALL, ir::TYPE_S32, dst, src);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SIMD_ANY:
> @@ -3900,14 +3855,12 @@ namespace gbe
> const ir::Register src = this->getRegister(*AI);
> const ir::Register dst = this->getRegister(&I);
> ctx.ALU1(ir::OP_SIMD_ANY, ir::TYPE_S32, dst, src);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_READ_TM:
> {
> const ir::Register dst = this->getRegister(&I);
> ctx.READ_ARF(ir::TYPE_U32, dst, ir::ARF_TM);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_REGION:
> @@ -3922,16 +3875,15 @@ namespace gbe
> const ir::Register src = this->getRegister(*AI);
>
> ctx.REGION(dst, src, x.getIntegerValue());
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_RSQ: this->emitUnaryCallInst(I,CS,ir::OP_RSQ); break;
> case GEN_OCL_RCP: this->emitUnaryCallInst(I,CS,ir::OP_RCP); break;
> case GEN_OCL_FORCE_SIMD8: ctx.setSimdWidth(8); break;
> case GEN_OCL_FORCE_SIMD16: ctx.setSimdWidth(16); break;
> - case GEN_OCL_LBARRIER: ctx.SYNC(ir::syncLocalBarrier); SET_GENIR_DBGINFO(I,1);break;
> - case GEN_OCL_GBARRIER: ctx.SYNC(ir::syncGlobalBarrier); SET_GENIR_DBGINFO(I,1);break;
> - case GEN_OCL_LGBARRIER: ctx.SYNC(ir::syncLocalBarrier | ir::syncGlobalBarrier); SET_GENIR_DBGINFO(I,1);break;
> + case GEN_OCL_LBARRIER: ctx.SYNC(ir::syncLocalBarrier); break;
> + case GEN_OCL_GBARRIER: ctx.SYNC(ir::syncGlobalBarrier); break;
> + case GEN_OCL_LGBARRIER: ctx.SYNC(ir::syncLocalBarrier | ir::syncGlobalBarrier); break;
> case GEN_OCL_ATOMIC_ADD0:
> case GEN_OCL_ATOMIC_ADD1: this->emitAtomicInst(I,CS,ir::ATOMIC_OP_ADD); break;
> case GEN_OCL_ATOMIC_SUB0:
> @@ -3971,7 +3923,6 @@ namespace gbe
> ir::ImageInfoKey key(imageID, infoType);
> const ir::Register infoReg = ctx.getFunction().getImageSet()->appendInfo(key, &ctx);
> ctx.GET_IMAGE_INFO(infoType, reg, imageID, infoReg);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
>
> @@ -4024,7 +3975,6 @@ namespace gbe
>
> ctx.SAMPLE(imageID, dstTuple, srcTuple, imageDim, dstType == ir::TYPE_FLOAT,
> requiredFloatCoord, sampler, samplerOffset);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
>
> @@ -4056,7 +4006,6 @@ namespace gbe
> }
> const ir::Tuple srcTuple = ctx.arrayTuple(&srcTupleData[0], imageDim + 4);
> ctx.TYPED_WRITE(imageID, srcTuple, imageDim + 4, srcType, ir::TYPE_U32);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_MUL_HI_INT:
> @@ -4065,7 +4014,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.MUL_HI(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_MUL_HI_UINT:
> @@ -4074,7 +4022,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.MUL_HI(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_MUL_HI_I64:
> @@ -4083,7 +4030,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.I64_MUL_HI(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_MUL_HI_UI64:
> @@ -4092,7 +4038,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.I64_MUL_HI(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_UPSAMPLE_SHORT:
> @@ -4101,7 +4046,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.UPSAMPLE_SHORT(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_UPSAMPLE_INT:
> @@ -4110,7 +4054,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.UPSAMPLE_INT(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_UPSAMPLE_LONG:
> @@ -4119,7 +4062,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.UPSAMPLE_LONG(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SADD_SAT_CHAR:
> @@ -4131,7 +4073,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.ADDSAT(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_UADD_SAT_CHAR:
> @@ -4143,7 +4084,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.ADDSAT(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SSUB_SAT_CHAR:
> @@ -4155,7 +4095,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.SUBSAT(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_USUB_SAT_CHAR:
> @@ -4167,7 +4106,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.SUBSAT(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_I64_MAD_SAT:
> @@ -4177,7 +4115,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src2 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.I64MADSAT(getType(ctx, I.getType()), dst, src0, src1, src2);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_I64_MAD_SATU:
> @@ -4187,7 +4124,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src2 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.I64MADSAT(getUnsignedType(ctx, I.getType()), dst, src0, src1, src2);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_FMAX:
> @@ -4203,9 +4139,7 @@ namespace gbe
> ctx.GE(getType(ctx, I.getType()), cmp, src0, src1);
> else
> ctx.LT(getType(ctx, I.getType()), cmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> ctx.SEL(getType(ctx, I.getType()), dst, cmp, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_HADD: {
> @@ -4213,7 +4147,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.HADD(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_I64HADD:
> @@ -4224,7 +4157,6 @@ namespace gbe
> const ir::Register src1 = this->getRegister(*(AI++));
> const ir::Register dst = this->getRegister(&I);
> ctx.I64HADD(ir::TYPE_U64, dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_RHADD: {
> @@ -4232,7 +4164,6 @@ namespace gbe
> GBE_ASSERT(AI != AE); const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.RHADD(getUnsignedType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_I64RHADD:
> @@ -4243,11 +4174,10 @@ namespace gbe
> const ir::Register src1 = this->getRegister(*(AI++));
> const ir::Register dst = this->getRegister(&I);
> ctx.I64RHADD(ir::TYPE_U64, dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> #define DEF(DST_TYPE, SRC_TYPE) \
> - { ctx.SAT_CVT(DST_TYPE, SRC_TYPE, getRegister(&I), getRegister(I.getOperand(0)));SET_GENIR_DBGINFO(I,1); break; }
> + { ctx.SAT_CVT(DST_TYPE, SRC_TYPE, getRegister(&I), getRegister(I.getOperand(0))); break; }
> case GEN_OCL_SAT_CONV_U8_TO_I8:
> DEF(ir::TYPE_S8, ir::TYPE_U8);
> case GEN_OCL_SAT_CONV_I16_TO_I8:
> @@ -4310,11 +4240,9 @@ namespace gbe
> DEF(ir::TYPE_U32, ir::TYPE_HALF);
> case GEN_OCL_CONV_F16_TO_F32:
> ctx.F16TO32(ir::TYPE_FLOAT, ir::TYPE_U16, getRegister(&I), getRegister(I.getOperand(0)));
> - SET_GENIR_DBGINFO(I,1);
> break;
> case GEN_OCL_CONV_F32_TO_F16:
> ctx.F32TO16(ir::TYPE_U16, ir::TYPE_FLOAT, getRegister(&I), getRegister(I.getOperand(0)));
> - SET_GENIR_DBGINFO(I,1);
> break;
> #undef DEF
>
> @@ -4329,14 +4257,12 @@ namespace gbe
> {
> const ir::Register dst = this->getRegister(&I);
> ctx.ALU0(ir::OP_SIMD_SIZE, getType(ctx, I.getType()), dst);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SIMD_ID:
> {
> const ir::Register dst = this->getRegister(&I);
> ctx.ALU0(ir::OP_SIMD_ID, getType(ctx, I.getType()), dst);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> case GEN_OCL_SIMD_SHUFFLE:
> @@ -4345,7 +4271,6 @@ namespace gbe
> const ir::Register src1 = this->getRegister(*AI); ++AI;
> const ir::Register dst = this->getRegister(&I);
> ctx.SIMD_SHUFFLE(getType(ctx, I.getType()), dst, src0, src1);
> - SET_GENIR_DBGINFO(I,1);
> break;
> }
> default: break;
> @@ -4392,6 +4317,7 @@ namespace gbe
> // below code assume align is power of 2
> GBE_ASSERT(align && (align & (align-1)) == 0);
>
> + SET_GENIR_DBGINFO(I,1);
> // align the stack pointer according to data alignment
> if(align > 1) {
> uint32_t prevStackPtr = ctx.getFunction().getStackSize();
> @@ -4400,20 +4326,15 @@ namespace gbe
> ir::ImmediateIndex stepImm = ctx.newIntegerImmediate(step, ir::TYPE_U32);
> ir::Register stepReg = ctx.reg(ctx.getPointerFamily());
> ctx.LOADI(ir::TYPE_U32, stepReg, stepImm);
> - SET_GENIR_DBGINFO(I,1);
> ctx.ADD(ir::TYPE_U32, stack, stack, stepReg);
> - SET_GENIR_DBGINFO(I,1);
> ctx.getFunction().pushStackSize(step);
> }
> }
> // Set the destination register properly
> ctx.MOV(imm.getType(), dst, stack);
> - SET_GENIR_DBGINFO(I,1);
>
> ctx.LOADI(imm.getType(), reg, immIndex);
> - SET_GENIR_DBGINFO(I,1);
> ctx.ADD(imm.getType(), stack, stack, reg);
> - SET_GENIR_DBGINFO(I,1);
> ctx.getFunction().pushStackSize(elementSize);
> }
>
> @@ -4438,6 +4359,7 @@ namespace gbe
> uint32_t msgNum = totalSize > 16 ? totalSize / 16 : 1;
> const uint32_t perMsgNum = elemNum / msgNum;
>
> + SET_GENIR_DBGINFO(llvmValues,3);
> for (uint32_t msg = 0; msg < msgNum; ++msg) {
> // Build the tuple data in the vector
> vector<ir::Register> tupleData; // put registers here
> @@ -4473,9 +4395,7 @@ namespace gbe
>
> addr = ctx.reg(pointerFamily);
> ctx.LOADI(immType, offset, immIndex);
> - SET_GENIR_DBGINFO(llvmValues,3);
> ctx.ADD(immType, addr, ptr, offset);
> - SET_GENIR_DBGINFO(llvmValues,3);
> }
>
> // Emit the instruction
> @@ -4483,7 +4403,6 @@ namespace gbe
> ctx.LOAD(type, tuple, addr, addrSpace, perMsgNum, dwAligned, fixedBTI, bti);
> else
> ctx.STORE(type, tuple, addr, addrSpace, perMsgNum, dwAligned, fixedBTI, bti);
> - SET_GENIR_DBGINFO(llvmValues,3);
> }
> }
>
> @@ -4521,14 +4440,12 @@ namespace gbe
> }
> const ir::Tuple byteTuple = ctx.arrayTuple(&byteTupleData[0], byteSize);
>
> + SET_GENIR_DBGINFO(llvmValues,3);
> if (isLoad) {
> ctx.LOAD(ir::TYPE_U8, byteTuple, ptr, addrSpace, byteSize, dwAligned, fixedBTI, bti);
> - SET_GENIR_DBGINFO(llvmValues,3);
> ctx.BITCAST(type, ir::TYPE_U8, tuple, byteTuple, elemNum, byteSize);
> - SET_GENIR_DBGINFO(llvmValues,3);
> } else {
> ctx.BITCAST(ir::TYPE_U8, type, byteTuple, tuple, byteSize, elemNum);
> - SET_GENIR_DBGINFO(llvmValues,3);
> // FIXME: byte scatter does not handle correctly vector store, after fix that,
> // we can directly use on store instruction like:
> // ctx.STORE(ir::TYPE_U8, byteTuple, ptr, addrSpace, byteSize, dwAligned, fixedBTI, bti);
> @@ -4544,12 +4461,9 @@ namespace gbe
> immIndex = ctx.newImmediate(int32_t(elemID));
> addr = ctx.reg(pointerFamily);
> ctx.LOADI(ir::TYPE_S32, offset, immIndex);
> - SET_GENIR_DBGINFO(llvmValues,3);
> ctx.ADD(ir::TYPE_S32, addr, ptr, offset);
> - SET_GENIR_DBGINFO(llvmValues,3);
> }
> ctx.STORE(type, addr, addrSpace, dwAligned, fixedBTI, bti, reg);
> - SET_GENIR_DBGINFO(llvmValues,3);
> }
> }
> }
> @@ -4576,6 +4490,7 @@ namespace gbe
>
> ir::Register btiReg;
> bool fixedBTI = false;
> + SET_GENIR_DBGINFO(I,1);
> if (isa<ConstantInt>(bti)) {
> fixedBTI = true;
> unsigned index = cast<ConstantInt>(bti)->getZExtValue();
> @@ -4583,7 +4498,6 @@ namespace gbe
> ir::ImmediateIndex immIndex = ctx.newImmediate((uint32_t)index);
> btiReg = ctx.reg(ir::FAMILY_DWORD);
> ctx.LOADI(ir::TYPE_U32, btiReg, immIndex);
> - SET_GENIR_DBGINFO(I,1);
> } else {
> addrSpace = ir::MEM_MIXED;
> btiReg = this->getRegister(bti);
> @@ -4601,7 +4515,6 @@ namespace gbe
> // so, I avoid subtracting zero base to satisfy ArgumentLower pass.
> if (!zeroBase) {
> ctx.SUB(ir::TYPE_U32, ptr, pointer, baseReg);
> - SET_GENIR_DBGINFO(I,1);
> }
> else
> ptr = pointer;
> @@ -4621,7 +4534,6 @@ namespace gbe
> ctx.LOAD(type, ptr, addrSpace, dwAligned, fixedBTI, btiReg, values);
> else
> ctx.STORE(type, ptr, addrSpace, dwAligned, fixedBTI, btiReg, values);
> - SET_GENIR_DBGINFO(I,1);
> }
> // A vector type requires to build a tuple
> else {
> @@ -4667,7 +4579,6 @@ namespace gbe
> ctx.LOAD(type, tuple, ptr, addrSpace, elemNum, dwAligned, fixedBTI, btiReg);
> else
> ctx.STORE(type, tuple, ptr, addrSpace, elemNum, dwAligned, fixedBTI, btiReg);
> - SET_GENIR_DBGINFO(I,1);
> }
> // Not supported by the hardware. So, we split the message and we use
> // strided loads and stores
> @@ -4694,15 +4605,12 @@ namespace gbe
> immIndex = ctx.newImmediate(int32_t(elemID * elemSize));
> addr = ctx.reg(pointerFamily);
> ctx.LOADI(ir::TYPE_S32, offset, immIndex);
> - SET_GENIR_DBGINFO(I,1);
> ctx.ADD(ir::TYPE_S32, addr, ptr, offset);
> - SET_GENIR_DBGINFO(I,1);
> }
> if (isLoad)
> ctx.LOAD(type, addr, addrSpace, dwAligned, fixedBTI, btiReg, reg);
> else
> ctx.STORE(type, addr, addrSpace, dwAligned, fixedBTI, btiReg, reg);
> - SET_GENIR_DBGINFO(I,1);
> }
> }
> }
> diff --git a/backend/src/llvm/llvm_to_gen.cpp b/backend/src/llvm/llvm_to_gen.cpp
> index 012b754..19e681f 100644
> --- a/backend/src/llvm/llvm_to_gen.cpp
> +++ b/backend/src/llvm/llvm_to_gen.cpp
> @@ -303,10 +303,10 @@ namespace gbe
> passes.add(createScalarizePass()); // Expand all vector ops
>
> //setenv("OCL_PROFILING","0",0);
> - if(OCL_PROFILING) {
> + /*if(OCL_PROFILING) {
> passes.add(createTimestampParserPass()); // by ynbai
> passes.add(createExpandConstantExprPass());
> - }
> + }*/
>
> if(OCL_OUTPUT_CFG)
> passes.add(createCFGPrinterPass());
> --
> 1.9.1
>
> _______________________________________________
> Beignet mailing list
> Beignet at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet
More information about the Beignet
mailing list