intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 15:53:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix calculation in encodeGreaterThanPredicate()

Browse Source 
encodeGreaterThanPredicate() computes *firstOperand > secondOperand and
stores the result in MI_PREDICATE_RESULT where firstOperand is an
device memory address.

To calculate the "greater than" operation in the device,
(secondOperand - *firstOperand) needs to be used, and if the carry flag
register is set, then (*firstOperand) is greater than secondOperand.

The order of operands in the subtraction (secondOperand - *firstOperand)
in encodeGreaterThanPredicate() is incorrect. Transpose the
operands.

Change-Id: Ie7fee32af338a1a3948abc78cf8332e0be8595ac
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
This commit is contained in:
Sebastian Sanchez
2020-02-05 13:38:11 -08:00
committed by sys_ocldev
parent e26e34e70d
commit f936b3e1cd
2 changed files with 22 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
core/command_container/command_encoder.inl
View File

@@ -93,10 +93,19 @@ void EncodeMathMMIO<Family>::encodeMulRegVal(CommandContainer &container, uint32
EncodeStoreMMIO<Family>::encode(container, CS_GPR_R1, dstAddress); EncodeStoreMMIO<Family>::encode(container, CS_GPR_R1, dstAddress);
} }
/*
* Compute *firstOperand > secondOperand and store the result in
* MI_PREDICATE_RESULT where firstOperand is an device memory address.
*
* To calculate the "greater than" operation in the device,
* (secondOperand - *firstOperand) is used, and if the carry flag register is
* set, then (*firstOperand) is greater than secondOperand.
*/
template <typename Family> template <typename Family>
void EncodeMathMMIO<Family>::encodeGreaterThanPredicate(CommandContainer &container, uint64_t lhsVal, uint32_t rhsVal) { void EncodeMathMMIO<Family>::encodeGreaterThanPredicate(CommandContainer &container, uint64_t firstOperand, uint32_t secondOperand) {
EncodeSetMMIO<Family>::encodeMEM(container, CS_GPR_R0, lhsVal); /* (*firstOperand) will be subtracted from secondOperand */
EncodeSetMMIO<Family>::encodeIMM(container, CS_GPR_R1, rhsVal); EncodeSetMMIO<Family>::encodeIMM(container, CS_GPR_R0, secondOperand);
EncodeSetMMIO<Family>::encodeMEM(container, CS_GPR_R1, firstOperand);
size_t size = sizeof(MI_MATH) + sizeof(MI_MATH_ALU_INST_INLINE) * NUM_ALU_INST_FOR_READ_MODIFY_WRITE; size_t size = sizeof(MI_MATH) + sizeof(MI_MATH_ALU_INST_INLINE) * NUM_ALU_INST_FOR_READ_MODIFY_WRITE;
@@ -136,6 +145,7 @@ void EncodeMathMMIO<Family>::encodeAlu(MI_MATH_ALU_INST_INLINE *pAluParam, uint3
pAluParam->DW0.BitField.Operand2 = srcB; pAluParam->DW0.BitField.Operand2 = srcB;
pAluParam++; pAluParam++;
/* Order of operation: Operand1 <ALUOpcode> Operand2 */
pAluParam->DW0.BitField.ALUOpcode = op; pAluParam->DW0.BitField.ALUOpcode = op;
pAluParam->DW0.BitField.Operand1 = 0; pAluParam->DW0.BitField.Operand1 = 0;
pAluParam->DW0.BitField.Operand2 = 0; pAluParam->DW0.BitField.Operand2 = 0;
@@ -149,6 +159,7 @@ void EncodeMathMMIO<Family>::encodeAlu(MI_MATH_ALU_INST_INLINE *pAluParam, uint3
template <typename Family> template <typename Family>
void EncodeMathMMIO<Family>::encodeAluSubStoreCarry(MI_MATH_ALU_INST_INLINE *pAluParam, uint32_t regA, uint32_t regB, uint32_t finalResultRegister) { void EncodeMathMMIO<Family>::encodeAluSubStoreCarry(MI_MATH_ALU_INST_INLINE *pAluParam, uint32_t regA, uint32_t regB, uint32_t finalResultRegister) {
/* regB is subtracted from regA */
encodeAlu(pAluParam, regA, regB, ALU_OPCODE_SUB, finalResultRegister, ALU_REGISTER_R_CF); encodeAlu(pAluParam, regA, regB, ALU_OPCODE_SUB, finalResultRegister, ALU_REGISTER_R_CF);
} }

16
core/unit_tests/encoders/test_encode_math.cpp
View File

@@ -96,20 +96,20 @@ HWTEST_F(CommandEncoderMathTest, appendsAGreaterThanPredicate) {
auto itor = commands.begin(); auto itor = commands.begin();
itor = find<MI_LOAD_REGISTER_MEM *>(itor, commands.end());
ASSERT_NE(itor, commands.end());
auto cmdMEM = genCmdCast<MI_LOAD_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmdMEM->getRegisterAddress(), CS_GPR_R0);
EXPECT_EQ(cmdMEM->getMemoryAddress(), 0xDEADBEEFCAF0u);
itor = find<MI_LOAD_REGISTER_IMM *>(itor, commands.end()); itor = find<MI_LOAD_REGISTER_IMM *>(itor, commands.end());
ASSERT_NE(itor, commands.end()); ASSERT_NE(itor, commands.end());
auto cmdIMM = genCmdCast<MI_LOAD_REGISTER_IMM *>(*itor); auto cmdIMM = genCmdCast<MI_LOAD_REGISTER_IMM *>(*itor);
EXPECT_EQ(cmdIMM->getRegisterOffset(), CS_GPR_R1); EXPECT_EQ(cmdIMM->getRegisterOffset(), CS_GPR_R0);
EXPECT_EQ(cmdIMM->getDataDword(), 17u); EXPECT_EQ(cmdIMM->getDataDword(), 17u);
itor = find<MI_LOAD_REGISTER_MEM *>(itor, commands.end());
ASSERT_NE(itor, commands.end());
auto cmdMEM = genCmdCast<MI_LOAD_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmdMEM->getRegisterAddress(), CS_GPR_R1);
EXPECT_EQ(cmdMEM->getMemoryAddress(), 0xDEADBEEFCAF0u);
itor = find<MI_MATH *>(itor, commands.end()); itor = find<MI_MATH *>(itor, commands.end());
ASSERT_NE(itor, commands.end()); ASSERT_NE(itor, commands.end());