OV - exec - Loop 3809

0x5318a0 LDR	X30, [X26, X14,LSL #3]

0x5318a4 UBFM	X0, X30, #61, #60

0x5318a8 ADD	X2, X0, #8

0x5318ac LDR	X1, [X25, X0]

0x5318b0 ADD	X10, X25, X2

0x5318b4 LDR	X3, [X25, X2]

0x5318b8 CMP	X1, X3

0x5318bc B.GE	5318f0

  (3810) 0x5318c0 LDR	X16, [X20, X1,LSL #3]

  (3810) 0x5318c4 UBFM	X5, X16, #61, #60

  (3810) 0x5318c8 LDR	X6, [X19, X5]

  (3810) 0x5318cc CMP	X4, X6

  (3810) 0x5318d0 B.EQ	531d18

  (3810) 0x5318d4 STR	X4, [X19, X5]

  (3810) 0x5318d8 ADD	X1, X1, #1

  (3810) 0x5318dc ADD	X17, X17, #1

  (3810) 0x5318e0 LDR	X3, [X10]

  (3810) 0x5318e4 CMP	X3, X1

  (3810) 0x5318e8 B.GT	5318c0

0x5318ec LDR	X9, [X15, X4,LSL #3]

0x5318f0 ADD	X14, X14, #1

0x5318f4 CMP	X14, X9

0x5318f8 B.LT	5318a0

  (3810) 0x531d18 ADD	X1, X1, #1

  (3810) 0x531d1c CMP	X1, X3

  (3810) 0x531d20 B.LT	5318c0

0x531d24 B	5318ec

/home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c: 222 - 231

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

222: 	for (ia = A_i[ic]; ia < A_i[ic+1]; ia++)

223: 	{

224: 	   ja = A_j[ia];

225: 	   for (ib = B_i[ja]; ib < B_i[ja+1]; ib++)

226: 	   {

227: 	      jb = B_j[ib];

228: 	      if (B_marker[jb] != ic)

229: 	      {

230: 	 	 B_marker[jb] = ic;

231: 		 num_nonzeros++;

Coverage (%)	Name	Source Location	Module
►98.20+	gomp_thread_start	team.c:130	libgomp.so.1.0.0
○	start_thread		libc.so.6
○	thread_start		libc.so.6
►1.80+	GOMP_parallel	libgomp.h:980	libgomp.so.1.0.0
○	hypre_CSRMatrixMultiply	csr_matop.c:185	exec
○	hypre_ParTMatmul	par_csr_matop.c:3287	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:1227	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	2.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.17
Bottlenecks
Function	hypre_CSRMatrixMultiply._omp_fn.0
Source	csr_matop.c:222-222,csr_matop.c:225-225
Source loop unroll info	NA
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	0.75
Front-end cycles	1.50
P0 cycles	1.17
P1 cycles	1.17
P2 cycles	1.00
P3 cycles	1.00
P4 cycles	1.00
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	1.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	1.22
P13 cycles	1.22
P14 cycles	1.22
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
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	19.49
Bytes prefetched	0.00
Bytes loaded	29.33
Bytes stored	0.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	50.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	2.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.38
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMultiply._omp_fn.0
Source	csr_matop.c:222-222,csr_matop.c:225-225
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.38
CQA cycles if no scalar integer	1.38
CQA cycles if FP arith vectorized	1.38
CQA cycles if fully vectorized	0.69
Front-end cycles	1.38
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	1.00
P7 cycles	1.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	1.00
P13 cycles	1.00
P14 cycles	1.00
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	11.00
Nb uops	11.00
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	17.45
Bytes prefetched	0.00
Bytes loaded	24.00
Bytes stored	0.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	50.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	2.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.13
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMultiply._omp_fn.0
Source	csr_matop.c:222-222,csr_matop.c:225-225
Source loop unroll info	NA
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	0.75
Front-end cycles	1.50
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	1.00
P7 cycles	1.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	1.33
P13 cycles	1.33
P14 cycles	1.33
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
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	21.33
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	2.00
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	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	50.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	2.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.08
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMultiply._omp_fn.0
Source	csr_matop.c:222-222,csr_matop.c:225-225
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.63
CQA cycles if no scalar integer	1.63
CQA cycles if FP arith vectorized	1.63
CQA cycles if fully vectorized	0.81
Front-end cycles	1.63
P0 cycles	1.50
P1 cycles	1.50
P2 cycles	1.00
P3 cycles	1.00
P4 cycles	1.00
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	1.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	1.33
P13 cycles	1.33
P14 cycles	1.33
P15 cycles	0.00
P16 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	13.00
Nb uops	13.00
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	19.69
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	1.00
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	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	50.00

Path /

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

Function	hypre_CSRMatrixMultiply._omp_fn.0
Source file and lines	csr_matop.c:222-231
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222-231.

Analyzed code is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222,225.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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. 2 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 0.75 cycles (2.00x 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

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 is loading 29 bytes.

General properties

nb instructions	12
loop length	48
nb stack references	0

Front-end

front end

1.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	1.17	1.17	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.22	1.22	1.22	0.00	0.00
cycles	1.17	1.17	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.22	1.22	1.22	0.00	0.00

Execution ports to units layout:

P0: BRU
P1: BRU
P2: ALU
P3: ALU
P4: ALU
P5: ALU
P6: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

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

Cycles summary

Front-end	1.50
Data deps.	1.00
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	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Cycles and memory resources usage

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

20% of peak load performance is reached (19.49 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

Function	hypre_CSRMatrixMultiply._omp_fn.0
Source file and lines	csr_matop.c:222-231
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222-231.

Analyzed code is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222,225.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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. 2 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.38 to 0.69 cycles (2.00x 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

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 is loading 24 bytes.

General properties

nb instructions	11
loop length	44
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.00	1.00	1.00	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	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: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

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

Cycles summary

Front-end	1.38
Data deps.	1.00
Overall L1	1.38

Vectorization ratios

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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.38 cycles. At this rate:

18% of peak load performance is reached (17.45 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 5318a0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR X30, [X26, X14,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
UBFM X0, X30, #61, #60	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
ADD X2, X0, #8	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
LDR X1, [X25, X0]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
ADD X10, X25, X2	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	scal (50.0%)
LDR X3, [X25, X2]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
CMP X1, X3	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	scal (50.0%)
B.GE 5318f0 <hypre_CSRMatrixMultiply._omp_fn.0+0x170>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
ADD X14, X14, #1	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
CMP X14, X9	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 5318a0 <hypre_CSRMatrixMultiply._omp_fn.0+0x120>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	hypre_CSRMatrixMultiply._omp_fn.0
Source file and lines	csr_matop.c:222-231
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222-231.

Analyzed code is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222,225.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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. 2 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 0.75 cycles (2.00x 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

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 is loading 32 bytes.

General properties

nb instructions	12
loop length	48
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	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: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

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

Cycles summary

Front-end	1.50
Data deps.	1.00
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	0%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Cycles and memory resources usage

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

22% of peak load performance is reached (21.33 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 5318a0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR X30, [X26, X14,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
UBFM X0, X30, #61, #60	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
ADD X2, X0, #8	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
LDR X1, [X25, X0]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
ADD X10, X25, X2	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	scal (50.0%)
LDR X3, [X25, X2]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
CMP X1, X3	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	scal (50.0%)
B.GE 5318f0 <hypre_CSRMatrixMultiply._omp_fn.0+0x170>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X9, [X15, X4,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	N/A
ADD X14, X14, #1	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
CMP X14, X9	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 5318a0 <hypre_CSRMatrixMultiply._omp_fn.0+0x120>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Function	hypre_CSRMatrixMultiply._omp_fn.0
Source file and lines	csr_matop.c:222-231
Module	exec

The loop is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222-231.

Analyzed code is defined in /home/hbollore/qaas/qaas-runs/174-161-6712/intel/AMG/build/AMG/AMG/seq_mv/csr_matop.c:222,225.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
This loop has 3 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. 2 data elements could be processed at once in vector registers. By vectorizing your loop, you can lower the cost of an iteration from 1.63 to 0.81 cycles (2.00x 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

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 is loading 32 bytes.

General properties

nb instructions	13
loop length	52
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

1.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	P16
uops	1.50	1.50	1.00	1.00	1.00	1.00	1.00	1.00	0.00	0.00	0.00	0.00	1.33	1.33	1.33	0.00	0.00
cycles	1.50	1.50	1.00	1.00	1.00	1.00	1.00	1.00	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: ALU
P7: ALU
P8 (128 bits): VPU, FP store data, ALU, DIV/SQRT
P9 (128 bits): VPU, ALU, FP store data
P10 (128 bits): VPU, ALU, DIV/SQRT
P11 (128 bits): ALU, VPU
P12 (256 bits): store address, load
P13 (256 bits): store address, load
P14 (256 bits): load
P15 (64 bits): store data
P16 (64 bits): store data

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

Cycles summary

Front-end	1.63
Data deps.	1.00
Overall L1	1.63

Vectorization ratios

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)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.63 cycles. At this rate:

20% of peak load performance is reached (19.69 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 5318a0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P12	P13	P14	Latency	Recip. throughput	Vectorization
LDR X30, [X26, X14,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
UBFM X0, X30, #61, #60	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
ADD X2, X0, #8	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
LDR X1, [X25, X0]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
ADD X10, X25, X2	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	scal (50.0%)
LDR X3, [X25, X2]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	scal (50.0%)
CMP X1, X3	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	scal (50.0%)
B.GE 5318f0 <hypre_CSRMatrixMultiply._omp_fn.0+0x170>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
LDR X9, [X15, X4,LSL #3]	1	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33	N/A
ADD X14, X14, #1	1	0	0	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	1	0.17	N/A
CMP X14, X9	1	0	0	0.25	0.25	0	0	0.25	0.25	0	0	0	1	0.33	N/A
B.LT 5318a0 <hypre_CSRMatrixMultiply._omp_fn.0+0x120>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A
B 5318ec <hypre_CSRMatrixMultiply._omp_fn.0+0x16c>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	1	0.50	N/A

Report Configuration

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Vectorization

Details

Workaround

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code