OV - exec - Loop 2039

0x4b4c64 LDR	X0, [X26, X5,LSL #3]    [7]

0x4b4c68 ADD	X5, X5, #1

0x4b4c6c UBFM	X6, X0, #61, #60

0x4b4c70 LDR	X22, [X23, X6]    [6]

0x4b4c74 CMP	X22, #0

0x4b4c78 B.LE	4b4cc0

0x4b4c7c LDR	X6, [X7, X6]    [1]

0x4b4c80 UBFM	X22, X6, #61, #60

0x4b4c84 CMP	X20, X6

0x4b4c88 B.EQ	4b4cc0

0x4b4c8c LDR	X0, [X21, X22]    [5]

0x4b4c90 CMP	X0, X3

0x4b4c94 B.GE	4b4cc0

0x4b4c98 LDR	X14, [SP, #336]    [2]

0x4b4c9c STR	X14, [X21, X22]    [5]

0x4b4ca0 STR	X6, [X10, X14,LSL #3]    [3]

0x4b4ca4 LDR	X6, [SP, #336]    [2]

0x4b4ca8 LDR	X14, [X30]    [4]

0x4b4cac ADD	X22, X6, #1

0x4b4cb0 STR	X22, [SP, #336]    [2]

0x4b4cb4 HINT	#0

0x4b4cb8 HINT	#0

0x4b4cbc HINT	#0

0x4b4cc0 CMP	X5, X14

0x4b4cc4 B.LT	4b4c64

/home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c: 2024 - 2034

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

2024:                  for (jj2 = S_diag_i[i2]; jj2 < S_diag_i[i2+1]; jj2++)

2025:                  {

2026:                     i3 = S_diag_j[jj2];

2027:                     if (CF_marker[i3] > 0)

2028:                     {

2029:                        index = fine_to_coarse[i3];

2030:                        if (index != ic && S_marker[index] < jj_row_begin_diag)

2031:                        {

2032:                           S_marker[index] = num_nonzeros_diag;

2033:                           C_diag_j[num_nonzeros_diag] = index;

2034:                           num_nonzeros_diag++;

Coverage (%)	Name	Source Location	Module
►100.00+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►50.39+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►49.61+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►75.20+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►24.80+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►87.56+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►12.44+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►93.80+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►6.20+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►95.80+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►4.20+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►97.10+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►2.90+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►97.52+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►2.48+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►97.98+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►2.02+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►98.08+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►1.92+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

Coverage (%)	Name	Source Location	Module
►98.47+	omp_fulfill_event		libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►1.53+	GOMP_parallel		libgomp.so.1.0.0
○	hypre_BoomerAMGCreate2ndS	par_strength.c:1668	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:622	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	amg.c:253	exec

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.16
Bottlenecks
Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source	par_strength.c:2024-2034
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.00
CQA cycles if no scalar integer	2.00
CQA cycles if FP arith vectorized	2.00
CQA cycles if fully vectorized	0.50
Front-end cycles	1.78
P0 cycles	1.63
P1 cycles	1.63
P2 cycles	1.56
P3 cycles	1.56
P4 cycles	1.56
P5 cycles	1.56
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	1.63
P11 cycles	1.54
P12 cycles	1.58
P13 cycles	0.38
P14 cycles	0.38
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	15.00
Nb uops	14.25
Nb loads	NA
Nb stores	0.75
Nb stack references	0.75
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	0.50
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	2.50
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.50
Bottlenecks	micro-operation queue, P0, P1, P2, P3, P4, P5,
Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source	par_strength.c:2024-2034
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.00
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.00
CQA cycles if fully vectorized	0.25
Front-end cycles	1.00
P0 cycles	1.00
P1 cycles	1.00
P2 cycles	1.00
P3 cycles	1.00
P4 cycles	1.00
P5 cycles	1.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.67
P11 cycles	0.67
P12 cycles	0.67
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	8.00
Nb uops	8.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	1.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.50
Bottlenecks	micro-operation queue, P0, P1, P2, P3, P4, P5,
Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source	par_strength.c:2024-2034
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.50
CQA cycles if no scalar integer	1.50
CQA cycles if FP arith vectorized	1.50
CQA cycles if fully vectorized	0.38
Front-end cycles	1.50
P0 cycles	1.50
P1 cycles	1.50
P2 cycles	1.50
P3 cycles	1.50
P4 cycles	1.50
P5 cycles	1.50
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	1.00
P11 cycles	1.00
P12 cycles	1.00
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	12.00
Nb uops	12.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	2.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.05
Bottlenecks	P10,
Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source	par_strength.c:2024-2034
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	3.50
CQA cycles if no scalar integer	3.50
CQA cycles if FP arith vectorized	3.50
CQA cycles if fully vectorized	0.88
Front-end cycles	2.75
P0 cycles	2.00
P1 cycles	2.00
P2 cycles	2.00
P3 cycles	2.00
P4 cycles	2.00
P5 cycles	2.00
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	3.50
P11 cycles	3.17
P12 cycles	3.33
P13 cycles	1.50
P14 cycles	1.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	25.00
Nb uops	22.00
Nb loads	NA
Nb stores	3.00
Nb stack references	3.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	2.00
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	4.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	25.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.07
Bottlenecks	P0, P1,
Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source	par_strength.c:2024-2034
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.00
CQA cycles if no scalar integer	2.00
CQA cycles if FP arith vectorized	2.00
CQA cycles if fully vectorized	0.50
Front-end cycles	1.88
P0 cycles	2.00
P1 cycles	2.00
P2 cycles	1.75
P3 cycles	1.75
P4 cycles	1.75
P5 cycles	1.75
P6 cycles	0.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	1.33
P11 cycles	1.33
P12 cycles	1.33
P13 cycles	0.00
P14 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	15.00
Nb uops	15.00
Nb loads	NA
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	0.00
Bytes prefetched	0.00
Bytes loaded	0.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	0.00
Stride indirect	3.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	25.00
Vector-efficiency ratio load	25.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Path /

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

Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source file and lines	par_strength.c:2024-2034
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:2024-2034.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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 not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 2.00 to 0.50 cycles (4.00x speedup).

Details

All VPU 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
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)

No data for this section

Slow data structures access

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

Details

Constant non-unit stride: 1 occurrence(s)
Irregular (variable stride) or indirect: 2 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)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop does not contain any FP arithmetical operations. The binary loop does not load or store any data.

General properties

nb instructions	15
nb uops	14.25
loop length	60
used w registers	0
used x registers	10
used b registers	0
used h registers	0
used s registers	0.25
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	0.75

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	1.78 cycles
front end	1.78 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.63	1.63	1.56	1.56	1.56	1.56	0.00	0.00	0.00	0.00	1.63	1.54	1.58	0.38	0.38
cycles	1.63	1.63	1.56	1.56	1.56	1.56	0.00	0.00	0.00	0.00	1.63	1.54	1.58	0.38	0.38

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.78
Dispatch	2.00
Data deps.	0.00
Overall L1	2.00

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.00 cycles.

Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source file and lines	par_strength.c:2024-2034
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:2024-2034.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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 not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.00 to 0.25 cycles (4.00x speedup).

Details

All VPU 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
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)

No data for this section

Slow data structures access

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

Details

Constant non-unit stride: 1 occurrence(s)
Irregular (variable stride) or indirect: 1 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)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop does not contain any FP arithmetical operations. The binary loop does not load or store any data.

General properties

nb instructions	8
nb uops	8
loop length	32
used w registers	0
used x registers	7
used b registers	0
used h registers	0
used s registers	0
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.00 cycles
front end	1.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	0.67	0.67	0.67	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	0.67	0.67	0.67	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.00
Dispatch	1.00
Data deps.	0.00
Overall L1	1.00

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.00 cycles.

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 1.00 to 0.67 cycles (1.50x speedup).

ASM code

In the binary file, the address of the loop is: 4b4c64

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDR X0, [X26, X5,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
ADD X5, X5, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
UBFM X6, X0, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
LDR X22, [X23, X6]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
CMP X22, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.LE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
CMP X5, X14	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 4b4c64 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xdfc>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A

Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source file and lines	par_strength.c:2024-2034
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:2024-2034.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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 not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.50 to 0.37 cycles (4.00x speedup).

Details

All VPU 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
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)

No data for this section

Slow data structures access

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

Details

Constant non-unit stride: 1 occurrence(s)
Irregular (variable stride) or indirect: 2 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)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop does not contain any FP arithmetical operations. The binary loop does not load or store any data.

General properties

nb instructions	12
nb uops	12
loop length	48
used w registers	0
used x registers	9
used b registers	0
used h registers	0
used s registers	0
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.50 cycles
front end	1.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	1.50	1.50	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	1.00	1.00	1.00	0.00	0.00
cycles	1.50	1.50	1.50	1.50	1.50	1.50	0.00	0.00	0.00	0.00	1.00	1.00	1.00	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.50
Dispatch	1.50
Data deps.	0.00
Overall L1	1.50

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.50 cycles.

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 1.50 to 1.00 cycles (1.50x speedup).

ASM code

In the binary file, the address of the loop is: 4b4c64

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDR X0, [X26, X5,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
ADD X5, X5, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
UBFM X6, X0, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
LDR X22, [X23, X6]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
CMP X22, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.LE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDR X6, [X7, X6]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	N/A
UBFM X22, X6, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	scal (25.0%)
CMP X20, X6	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.EQ 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
CMP X5, X14	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 4b4c64 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xdfc>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A

Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source file and lines	par_strength.c:2024-2034
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:2024-2034.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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 not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 3.50 to 0.88 cycles (4.00x speedup).

Details

All VPU 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
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)

No data for this section

Slow data structures access

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

Details

Constant non-unit stride: 1 occurrence(s)
Irregular (variable stride) or indirect: 4 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)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop does not contain any FP arithmetical operations. The binary loop does not load or store any data.

General properties

nb instructions	25
nb uops	22
loop length	100
used w registers	0
used x registers	13
used b registers	0
used h registers	0
used s registers	1
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	3

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	2.75 cycles
front end	2.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	2.00	2.00	2.00	2.00	2.00	2.00	0.00	0.00	0.00	0.00	3.50	3.17	3.33	1.50	1.50
cycles	2.00	2.00	2.00	2.00	2.00	2.00	0.00	0.00	0.00	0.00	3.50	3.17	3.33	1.50	1.50

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	2.75
Dispatch	3.50
Data deps.	0.00
Overall L1	3.50

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 3.50 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 4b4c64

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR X0, [X26, X5,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
ADD X5, X5, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25	N/A
UBFM X6, X0, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25	N/A
LDR X22, [X23, X6]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	N/A
CMP X22, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33	scal (25.0%)
B.LE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X6, [X7, X6]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	N/A
UBFM X22, X6, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25	N/A
CMP X20, X6	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33	scal (25.0%)
B.EQ 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X0, [X21, X22]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	scal (25.0%)
CMP X0, X3	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33	scal (25.0%)
B.GE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X14, [SP, #336]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	N/A
STR X14, [X21, X22]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (25.0%)
STR X6, [X10, X14,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (25.0%)
LDR X6, [SP, #336]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	N/A
LDR X14, [X30]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	0	0	4	0.33	N/A
ADD X22, X6, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.25	N/A
STR X22, [SP, #336]	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0.50	0.50	1	0.50	scal (25.0%)
HINT #0																			N/A
HINT #0																			N/A
HINT #0																			N/A
CMP X5, X14	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	0	1	0.33	N/A
B.LT 4b4c64 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xdfc>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	hypre_BoomerAMGCreate2ndS._omp_fn.7
Source file and lines	par_strength.c:2024-2034
Module	exec

The loop is defined in /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/build/AMG/AMG/parcsr_ls/par_strength.c:2024-2034.

The related source loop is not unrolled or unrolled with no peel/tail loop.
This loop has 4 execution paths.

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 not vectorized. 4 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 2.00 to 0.50 cycles (4.00x speedup).

Details

All VPU 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
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)

No data for this section

Slow data structures access

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

Details

Constant non-unit stride: 1 occurrence(s)
Irregular (variable stride) or indirect: 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)
Try to remove indirect accesses. If applicable, precompute elements out of the innermost loop.

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

The binary loop does not contain any FP arithmetical operations. The binary loop does not load or store any data.

General properties

nb instructions	15
nb uops	15
loop length	60
used w registers	0
used x registers	11
used b registers	0
used h registers	0
used s registers	0
used d registers	0
used q registers	0
used v registers	0
used z registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.88 cycles
front end	1.88 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	2.00	2.00	1.75	1.75	1.75	1.75	0.00	0.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	2.00	2.00	1.75	1.75	1.75	1.75	0.00	0.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P7 (128 bits): VPU, ALU, FP store data
P8 (128 bits): VPU, ALU, DIV/SQRT
P9 (128 bits): ALU, VPU
P10 (256 bits): store address, load
P11 (256 bits): store address, load
P12 (256 bits): load
P13 (64 bits): store data
P14 (64 bits): store data

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	0.00

Cycles summary

Front-end	1.88
Dispatch	2.00
Data deps.	0.00
Overall L1	2.00

Vectorization ratios

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

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.00 cycles.

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 4b4c64

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P10	P11	P12	Latency	Recip. throughput	Vectorization
LDR X0, [X26, X5,LSL #3]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
ADD X5, X5, #1	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
UBFM X6, X0, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
LDR X22, [X23, X6]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	N/A
CMP X22, #0	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.LE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDR X6, [X7, X6]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	N/A
UBFM X22, X6, #61, #60	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.25	N/A
CMP X20, X6	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.EQ 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
LDR X0, [X21, X22]	1	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (25.0%)
CMP X0, X3	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	scal (25.0%)
B.GE 4b4cc0 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xe58>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A
CMP X5, X14	1	0	0	0.25	0.25	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 4b4c64 <hypre_BoomerAMGCreate2ndS._omp_fn.7+0xdfc>	1	0.50	0.50	0	0	0	0	0	0	0	1	0.50	N/A

▶Scalability Runs Description

Run 1x1	Number processes: 1Number nodes: NARun Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_NUM_THREADS: 1OMP_PLACES: threads
Run 1x2	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 2OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x4	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 4OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x8	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 8OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x16	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 16OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x24	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 24OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x32	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 32OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x40	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 40OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x48	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 48OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x56	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 56OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads
Run 1x64	Number processes: 1Run Command: <executable> -n 400 400 400MPI Command: mpirun -n <number_processes> --bind-to core --map-by package:PE=64 --rank-by fill --report-bindings Dataset: Run Directory: /home/eoseret/qaas/qaas_runs/178-188-3659/intel/AMG/run/oneview_runs/multicore/gcc_1/oneview_run_1781892409OMP_NUM_THREADS: 64OMP_PROC_BIND: spreadOMP_DISPLAY_AFFINITY: TRUEOMP_AFFINITY_FORMAT: 'OMP: pid %P tid %i thread %n bound to OS proc set {%A}'OMP_DISPLAY_ENV: TRUEOMP_PLACES: threads

(1x1) Efficiency	(1x1) Potential Speed-Up (%)	(1x2) Efficiency	(1x2) Potential Speed-Up (%)	(1x4) Efficiency	(1x4) Potential Speed-Up (%)	(1x8) Efficiency	(1x8) Potential Speed-Up (%)	(1x16) Efficiency	(1x16) Potential Speed-Up (%)	(1x24) Efficiency	(1x24) Potential Speed-Up (%)	(1x32) Efficiency	(1x32) Potential Speed-Up (%)	(1x40) Efficiency	(1x40) Potential Speed-Up (%)	(1x48) Efficiency	(1x48) Potential Speed-Up (%)	(1x56) Efficiency	(1x56) Potential Speed-Up (%)	(1x64) Efficiency	(1x64) Potential Speed-Up (%)
1	0	1.87	0	3.53	0	6.34	0	11.62	0	16.44	0	21.21	0	25.44	0	29.16	0	33.36	0	34.96	0

Speed-Up and Efficiency

▶Detailed Speed-Up and Efficiency

Run	Number of threads	Efficiency (ideal is 1)	Speedup	Ideal Speedup	Time (s)	Coverage (%)
1x1	1	1	1	1	10.545002937317	3.0468137264252
1x2	2	1.87	1.87	2	5.4850010871887	3.0148048400879
1x4	4	3.53	3.53	4	2.7449996471405	2.9164528846741
1x8	8	6.34	6.34	8	1.3949998617172	2.6920609474182
1x16	16	11.62	11.62	16	0.7399999499321	1.7273496389389
1x24	24	16.44	16.44	24	0.49500000476837	1.2848041057587
1x32	32	21.21	21.21	32	0.38499999046326	1.0637435913086
1x40	40	25.44	25.44	40	0.31999999284744	0.93836355209351
1x48	48	29.16	29.16	48	0.28000000119209	0.81600111722946
1x56	56	33.36	33.36	56	0.24500000476837	0.72500133514404
1x64	64	34.96	34.96	64	0.23500001430511	0.6790965795517

Report Configuration

Vectorization

Details

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Vectorization

Details

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Slow data structures access

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Slow data structures access

Details

Workaround

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

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