OV - - Loop 842 - libkripke.so

	Loop Id: 842	Module: libkripke.so	Source: Scattering.cpp:88-97 [...]	Coverage: 60.12%

0x619c0 VXORPD	%XMM0,%XMM0,%XMM0

0x619c4 MOV	-0x100(%RBP),%RDI

0x619cb LEA	(%R11,%RCX,1),%RDX

0x619cf ADD	%R9,%RCX

0x619d2 INC	%RSI

0x619d5 VMOVSD	(%RDI,%RDX,8),%XMM1

0x619da VFMADD213SD	(%R15,%RCX,8),%XMM0,%XMM1

0x619e0 VMOVSD	%XMM1,(%R15,%RCX,8)

0x619e6 CMP	-0x48(%RBP),%RSI

0x619ea JE	61970

0x619ec MOV	-0x68(%RBP),%RCX

0x619f0 MOV	-0x108(%RBP),%RDX

0x619f7 ADD	%RSI,%RCX

0x619fa MOVSXD	(%RDX,%RCX,4),%R8

0x619fe TEST	%R8,%R8

0x61a01 JLE	619c0

0x61a03 MOV	-0xf0(%RBP),%RDX

0x61a0a VXORPD	%XMM0,%XMM0,%XMM0

0x61a0e MOV	(%RDX,%RCX,8),%R10

0x61a12 MOV	-0xf8(%RBP),%RDX

0x61a19 MOV	(%RDX,%RAX,8),%RDX

0x61a1d ADD	%R10,%R8

0x61a20 IMUL	-0xe8(%RBP),%RDX

0x61a28 ADD	-0x40(%RBP),%RDX

0x61a2c NOPL	(%RAX)

  (843) 0x61a30 MOV	(%R14,%R10,8),%RDI

  (843) 0x61a34 IMUL	%R12,%RDI

  (843) 0x61a38 ADD	%RDX,%RDI

  (843) 0x61a3b VMOVSD	(%R13,%RDI,8),%XMM1

  (843) 0x61a42 VFMADD231SD	(%RBX,%R10,8),%XMM1,%XMM0

  (843) 0x61a48 INC	%R10

  (843) 0x61a4b CMP	%R8,%R10

  (843) 0x61a4e JL	61a30

0x61a50 JMP	619c4

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Operators.hpp: 307 - 307

--------------------------------------------------------------------------------

307:     return Ret{lhs} + rhs;

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/index/IndexValue.hpp: 83 - 217

--------------------------------------------------------------------------------

83:     value++;

[...]

217:     return (value < x.value);

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Layout.hpp: 55 - 55

--------------------------------------------------------------------------------

55:     return a * b;

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/src/Kripke/Kernel/Scattering.cpp: 88 - 97

--------------------------------------------------------------------------------

88:             MixElem mix_stop = mix_start + zone_to_num_mixelem(z);

89:

90:             double sigs_z = 0.0;

91:             for(MixElem mix = mix_start;mix < mix_stop;++ mix){

92:               Material mat = mixelem_to_material(mix);

93:               double fraction = mixelem_to_fraction(mix);

94:

95:               sigs_z += sigs(mat, n, global_g, global_gp) * fraction;

96:             }

97:             phi_out(nm, g, z) += sigs_z * phi(nm, gp, z);

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/policy/loop/forall.hpp: 59 - 59

--------------------------------------------------------------------------------

59:   for (decltype(distance_it) i = 0; i < distance_it; ++i) {

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/internal/Iterators.hpp: 177 - 177

--------------------------------------------------------------------------------

177:     return value_type(val + rhs);

Path /

Metric	Value
CQA speedup if no scalar integer	1.57
CQA speedup if FP arith vectorized	1.69
CQA speedup if fully vectorized	4.19
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.13
Bottlenecks
Function	.omp_outlined..193
Source	Operators.hpp:307-307,Scattering.cpp:88-88,Scattering.cpp:91-91,Scattering.cpp:97-97,forall.hpp:59-59,Iterators.hpp:177-177
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	3.67
CQA cycles if no scalar integer	2.33
CQA cycles if FP arith vectorized	2.17
CQA cycles if fully vectorized	0.88
Front-end cycles	3.25
DIV/SQRT cycles	2.38
P0 cycles	2.38
P1 cycles	2.25
P2 cycles	2.25
P3 cycles	1.25
P4 cycles	3.67
P5 cycles	3.67
P6 cycles	3.67
P7 cycles	0.50
P8 cycles	0.50
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.50
P12 cycles	0.50
P13 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	20.50
Nb uops	19.50
Nb loads	10.00
Nb stores	1.00
Nb stack references	6.00
FLOP/cycle	0.55
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	1.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	22.82
Bytes prefetched	0.00
Bytes loaded	76.00
Bytes stored	8.00
Stride 0	1.00
Stride 1	1.00
Stride n	0.00
Stride unknown	4.50
Stride indirect	0.00
Vectorization ratio all	13.33
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	41.67
Vector-efficiency ratio all	14.17
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	17.71

Metric	Value
CQA speedup if no scalar integer	1.40
CQA speedup if FP arith vectorized	1.75
CQA speedup if fully vectorized	4.15
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.17
Bottlenecks	P5, P6, P7,
Function	.omp_outlined..193
Source	Operators.hpp:307-307,Scattering.cpp:88-88,Scattering.cpp:91-91,Scattering.cpp:97-97,forall.hpp:59-59,Iterators.hpp:177-177
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.67
CQA cycles if no scalar integer	3.33
CQA cycles if FP arith vectorized	2.67
CQA cycles if fully vectorized	1.13
Front-end cycles	4.00
DIV/SQRT cycles	3.00
P0 cycles	3.00
P1 cycles	2.75
P2 cycles	2.75
P3 cycles	1.50
P4 cycles	4.67
P5 cycles	4.67
P6 cycles	4.67
P7 cycles	0.50
P8 cycles	0.50
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.50
P12 cycles	0.50
P13 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	25.00
Nb uops	24.00
Nb loads	13.00
Nb stores	1.00
Nb stack references	8.00
FLOP/cycle	0.43
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	1.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	23.14
Bytes prefetched	0.00
Bytes loaded	100.00
Bytes stored	8.00
Stride 0	1.00
Stride 1	2.00
Stride n	0.00
Stride unknown	6.00
Stride indirect	0.00
Vectorization ratio all	10.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	33.33
Vector-efficiency ratio all	13.75
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	16.67

Metric	Value
CQA speedup if no scalar integer	2.00
CQA speedup if FP arith vectorized	1.60
CQA speedup if fully vectorized	4.27
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.07
Bottlenecks	P5, P6, P7,
Function	.omp_outlined..193
Source	Operators.hpp:307-307,Scattering.cpp:88-88,Scattering.cpp:91-91,Scattering.cpp:97-97,forall.hpp:59-59,Iterators.hpp:177-177
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.67
CQA cycles if no scalar integer	1.33
CQA cycles if FP arith vectorized	1.67
CQA cycles if fully vectorized	0.63
Front-end cycles	2.50
DIV/SQRT cycles	1.75
P0 cycles	1.75
P1 cycles	1.75
P2 cycles	1.75
P3 cycles	1.00
P4 cycles	2.67
P5 cycles	2.67
P6 cycles	2.67
P7 cycles	0.50
P8 cycles	0.50
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.50
P12 cycles	0.50
P13 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	16.00
Nb uops	15.00
Nb loads	7.00
Nb stores	1.00
Nb stack references	4.00
FLOP/cycle	0.75
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	1.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	22.50
Bytes prefetched	0.00
Bytes loaded	52.00
Bytes stored	8.00
Stride 0	1.00
Stride 1	0.00
Stride n	0.00
Stride unknown	3.00
Stride indirect	0.00
Vectorization ratio all	16.67
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	50.00
Vector-efficiency ratio all	14.58
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	18.75

Path /

Average path: Display a virtual path defined by average values of all real paths

Function	.omp_outlined..193
Source file and lines	Scattering.cpp:88-97
Module	libkripke.so

The loop is defined in:

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Operators.hpp:307
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/index/IndexValue.hpp:83,217
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Layout.hpp:55
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/src/Kripke/Kernel/Scattering.cpp:88-97
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/policy/loop/forall.hpp:59
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/internal/Iterators.hpp:177

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
The structure of this loop is probably <if then [else] end>.

The presence of multiple execution paths is typically the main/first bottleneck.
Try to simplify control inside loop: ideally, try to remove all conditional expressions, for example by (if applicable):

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is probably not vectorized. Only 14% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 3.67 to 0.87 cycles (4.19x speedup).

Details

Store and arithmetical SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

FMA

Detected 1 FMA (fused multiply-add) operations.

Type of elements and instruction set

1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time).

Matching between your loop (in the source code) and the binary loop

The binary loop is composed of 2 FP arithmetical operations:

1: addition or subtraction (all inside FMA instructions)
1: multiply (all inside FMA instructions)

The binary loop is loading 76 bytes (9 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.02 FP operations per loaded or stored byte.

General properties

nb instructions	20.50
nb uops	19.50
loop length	90
used x86 registers	10
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	6

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	3.25 cycles
front end	3.25 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	2.38	2.38	2.25	2.25	1.25	3.67	3.67	3.67	0.50	0.50	0.00	0.00	0.50	0.50
cycles	2.38	2.38	2.25	2.25	1.25	3.67	3.67	3.67	0.50	0.50	0.00	0.00	0.50	0.50

Execution ports to units layout:

ALU0/BRU0: ALU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1:
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	3.25
Dispatch	3.67
Data deps.	1.00
Overall L1	3.67

Vectorization ratios

INT

all	0%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

all	25%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	13%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	41%

Vector efficiency ratios

INT

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
other	12%

all	15%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	14%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	17%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.67 cycles. At this rate:

21% of peak load performance is reached (20.46 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
3% of peak store performance is reached (2.36 out of 64.00 bytes stored per cycle (GB/s @ 1GHz))

Function	.omp_outlined..193
Source file and lines	Scattering.cpp:88-97
Module	libkripke.so

The loop is defined in:

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Operators.hpp:307
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/index/IndexValue.hpp:83,217
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Layout.hpp:55
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/src/Kripke/Kernel/Scattering.cpp:88-97
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/policy/loop/forall.hpp:59
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/internal/Iterators.hpp:177

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 4.67 to 3.33 cycles (1.40x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Your loop is probably not vectorized. Only 13% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 4.67 to 1.12 cycles (4.15x speedup).

Details

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 4.67 to 4.00 cycles (1.17x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop

FMA

Detected 1 FMA (fused multiply-add) operations.

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 6 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Type of elements and instruction set

1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time).

Matching between your loop (in the source code) and the binary loop

The binary loop is composed of 2 FP arithmetical operations:

1: addition or subtraction (all inside FMA instructions)
1: multiply (all inside FMA instructions)

The binary loop is loading 100 bytes (12 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.02 FP operations per loaded or stored byte.

General properties

nb instructions	25
nb uops	24
loop length	113
used x86 registers	11
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	8

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.00 cycles
front end	4.00 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	3.00	3.00	2.75	2.75	1.50	4.67	4.67	4.67	0.50	0.50	0.00	0.00	0.50	0.50
cycles	3.00	3.00	2.75	2.75	1.50	4.67	4.67	4.67	0.50	0.50	0.00	0.00	0.50	0.50

Execution ports to units layout:

ALU0/BRU0: ALU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1:
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	4.00
Dispatch	4.67
Data deps.	1.00
Overall L1	4.67

Vectorization ratios

INT

all	0%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

all	25%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	10%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	33%

Vector efficiency ratios

INT

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
other	12%

all	15%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	13%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	16%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.67 cycles. At this rate:

22% of peak load performance is reached (21.43 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (1.71 out of 64.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 619ec

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP4	FP5	Latency	Recip. throughput
MOV -0x100(%RBP),%RDI	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
LEA (%R11,%RCX,1),%RDX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
ADD %R9,%RCX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
INC %RSI	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
VMOVSD (%RDI,%RDX,8),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50
VFMADD213SD (%R15,%RCX,8),%XMM0,%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0.50	0.50	0	0	4	0.50
VMOVSD %XMM1,(%R15,%RCX,8)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	1	1
CMP -0x48(%RBP),%RSI	1	0.25	0.25	0.25	0.25	0	0.33	0.33	0.33	0	0	0	0	1	0.33
JE 61970 <.omp_outlined..193+0x270>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1
MOV -0x68(%RBP),%RCX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
MOV -0x108(%RBP),%RDX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
ADD %RSI,%RCX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
MOVSXD (%RDX,%RCX,4),%R8	1	0.25	0.25	0.25	0.25	0	0.33	0.33	0.33	0	0	0	0	1	0.33
TEST %R8,%R8	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
JLE 619c0 <.omp_outlined..193+0x2c0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1
MOV -0xf0(%RBP),%RDX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17
MOV (%RDX,%RCX,8),%R10	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
MOV -0xf8(%RBP),%RDX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
MOV (%RDX,%RAX,8),%RDX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
ADD %R10,%R8	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
IMUL -0xe8(%RBP),%RDX	1	0	1	0	0	0	0.33	0.33	0.33	0	0	0	0	3	1
ADD -0x40(%RBP),%RDX	1	0.25	0.25	0.25	0.25	0	0.33	0.33	0.33	0	0	0	0	1	0.33
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09
JMP 619c4 <.omp_outlined..193+0x2c4>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	0	1

Function	.omp_outlined..193
Source file and lines	Scattering.cpp:88-97
Module	libkripke.so

The loop is defined in:

/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Operators.hpp:307
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/index/IndexValue.hpp:83,217
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/util/Layout.hpp:55
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/src/Kripke/Kernel/Scattering.cpp:88-97
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/policy/loop/forall.hpp:59
/beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/build/Kripke/tpl/raja/include/RAJA/internal/Iterators.hpp:177

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 2.67 to 1.33 cycles (2.00x speedup).

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Vectorization

Details

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 2.67 to 2.50 cycles (1.07x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop

FMA

Detected 1 FMA (fused multiply-add) operations.

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 3 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

Try to reorganize arrays of structures to structures of arrays
Consider to permute loops (see vectorization gain report)

Type of elements and instruction set

1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time).

Matching between your loop (in the source code) and the binary loop

The binary loop is composed of 2 FP arithmetical operations:

1: addition or subtraction (all inside FMA instructions)
1: multiply (all inside FMA instructions)

The binary loop is loading 52 bytes (6 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.03 FP operations per loaded or stored byte.

General properties

nb instructions	16
nb uops	15
loop length	67
used x86 registers	9
used mmx registers	0
used xmm registers	2
used ymm registers	0
used zmm registers	0
nb stack references	4

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	2.50 cycles
front end	2.50 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	1.75	1.75	1.75	1.75	1.00	2.67	2.67	2.67	0.50	0.50	0.00	0.00	0.50	0.50
cycles	1.75	1.75	1.75	1.75	1.00	2.67	2.67	2.67	0.50	0.50	0.00	0.00	0.50	0.50

Execution ports to units layout:

ALU0/BRU0: ALU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1:
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	1.00

Cycles summary

Front-end	2.50
Dispatch	2.67
Data deps.	1.00
Overall L1	2.67

Vectorization ratios

INT

all	0%
load	0%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

all	25%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	16%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	0%
fma	0%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	50%

Vector efficiency ratios

INT

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
other	12%

all	15%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	14%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	18%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.67 cycles. At this rate:

20% of peak load performance is reached (19.50 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
4% of peak store performance is reached (3.00 out of 64.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 619ec

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP4	FP5	Latency	Recip. throughput
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17
MOV -0x100(%RBP),%RDI	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
LEA (%R11,%RCX,1),%RDX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
ADD %R9,%RCX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
INC %RSI	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
VMOVSD (%RDI,%RDX,8),%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50
VFMADD213SD (%R15,%RCX,8),%XMM0,%XMM1	1	0	0	0	0	0	0.33	0.33	0.33	0.50	0.50	0	0	4	0.50
VMOVSD %XMM1,(%R15,%RCX,8)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	1	1
CMP -0x48(%RBP),%RSI	1	0.25	0.25	0.25	0.25	0	0.33	0.33	0.33	0	0	0	0	1	0.33
JE 61970 <.omp_outlined..193+0x270>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1
MOV -0x68(%RBP),%RCX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
MOV -0x108(%RBP),%RDX	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.33
ADD %RSI,%RCX	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
MOVSXD (%RDX,%RCX,4),%R8	1	0.25	0.25	0.25	0.25	0	0.33	0.33	0.33	0	0	0	0	1	0.33
TEST %R8,%R8	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25
JLE 619c0 <.omp_outlined..193+0x2c0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1

▶Scalability Runs Description

Run 2x1	Number processes: 2Number nodes: 1Number processes per node: 2Run Command: <executable> --groups 1024 --procs 2,1,1MPI Command: mpirun -np <number_processes>Dataset: Run Directory: /beegfs/hackathon/users/eoseret/qaas_runs/170-850-6313/intel/Kripke/run/oneview_runs/compilers/aocc_10/oneview_run_1708510900I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spreadOMP_NUM_THREADS: 1
Run 2x2	OMP_NUM_THREADS: 2I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x4	OMP_NUM_THREADS: 4I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x8	OMP_NUM_THREADS: 8I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x16	OMP_NUM_THREADS: 16I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x32	OMP_NUM_THREADS: 32I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x64	OMP_NUM_THREADS: 64I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread
Run 2x96	OMP_NUM_THREADS: 96I_MPI_PIN_DOMAIN: auto:scatterOMP_PLACES: threadsOMP_PROC_BIND: spread

(2x1) Efficiency	(2x1) Potential Speed-Up (%)	(2x2) Efficiency	(2x2) Potential Speed-Up (%)	(2x4) Efficiency	(2x4) Potential Speed-Up (%)	(2x8) Efficiency	(2x8) Potential Speed-Up (%)	(2x16) Efficiency	(2x16) Potential Speed-Up (%)	(2x32) Efficiency	(2x32) Potential Speed-Up (%)	(2x64) Efficiency	(2x64) Potential Speed-Up (%)	(2x96) Efficiency	(2x96) Potential Speed-Up (%)
1	0	0.99	0.48	1	0	1	0.07	1	0.06	0.99	0.35	0.92	2.95	0.78	7.73

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency

Run	Number of threads	Efficiency (ideal is 1)	Speedup	Ideal Speedup	Time (s)	Coverage (%)
2x1	2	1	1	1	666.72	60.12
2x2	4	0.99	1.98	2	335.1	59.55
2x4	8	1	4.01	4	167.65	58.36
2x8	16	1	7.99	8	84.03	58.58
2x16	32	1	15.98	16	42.3	56.61
2x32	64	0.99	31.79	32	21.41	53.02
2x64	128	0.92	59.15	64	11.1	38.87
2x96	192	0.78	74.58	96	8.4	34.65

Report Configuration

Vectorization

Details

Workaround

FMA

Type of elements and instruction set

Matching between your loop (in the source code) and the binary loop

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

FMA

Slow data structures access

Details

Workaround

Type of elements and instruction set

Matching between your loop (in the source code) and the binary loop

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

FMA

Slow data structures access

Details

Workaround

Type of elements and instruction set

Matching between your loop (in the source code) and the binary loop

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

▶Scalability Runs Description

▼Colums Filter

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency