OV - exec - Loop 1215

0x49c4c4 LDUR	X3, [X1, #496]    [5]

0x49c4c8 LDUR	D0, [X2, #496]    [6]

0x49c4cc LDR	D1, [X13, X9,LSL #3]    [2]

0x49c4d0 SUBS	X0, X0, #4

0x49c4d4 LDR	D2, [X14, X3,LSL #3]    [7]

0x49c4d8 FMADD	D0, D0, D1, D2

0x49c4dc STR	D0, [X14, X3,LSL #3]    [7]

0x49c4e0 LDUR	X3, [X1, #504]    [5]

0x49c4e4 LDUR	D0, [X2, #504]    [6]

0x49c4e8 LDR	D1, [X13, X9,LSL #3]    [2]

0x49c4ec LDR	D2, [X14, X3,LSL #3]    [3]

0x49c4f0 FMADD	D0, D0, D1, D2

0x49c4f4 STR	D0, [X14, X3,LSL #3]    [3]

0x49c4f8 LDR	X3, [X1]    [5]

0x49c4fc LDR	D0, [X2]    [6]

0x49c500 LDR	D1, [X13, X9,LSL #3]    [2]

0x49c504 LDR	D2, [X14, X3,LSL #3]    [4]

0x49c508 FMADD	D0, D0, D1, D2

0x49c50c STR	D0, [X14, X3,LSL #3]    [4]

0x49c510 LDR	X3, [X1, #8]    [5]

0x49c514 LDR	D0, [X2, #8]    [6]

0x49c518 LDR	D1, [X13, X9,LSL #3]    [2]

0x49c51c ADD	X1, X1, #32

0x49c520 ADD	X2, X2, #32

0x49c524 LDR	D2, [X14, X3,LSL #3]    [1]

0x49c528 FMADD	D0, D0, D1, D2

0x49c52c STR	D0, [X14, X3,LSL #3]    [1]

0x49c530 B.NE	49c4c4

/home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c: 564 - 567

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

564:                for (jj = A_i[i]; jj < A_i[i+1]; jj++)

565:                {

566:                   j = A_j[jj];

567:                   y_data_expand[offset + j] += A_data[jj] * x_data[i];

Coverage (%)	Name	Source Location	Module

Path /

Metric	Value
CQA speedup if no scalar integer	1.25
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	1.90
Bottlenecks	P10, P11, P12,
Function	.omp_outlined..26#0x49c1a0
Source	csr_matvec.c:564-567
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	4
CQA cycles	6.67
CQA cycles if no scalar integer	5.33
CQA cycles if FP arith vectorized	6.67
CQA cycles if fully vectorized	1.67
Front-end cycles	3.50
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	0.75
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	2.00
P6 cycles	2.00
P7 cycles	2.00
P8 cycles	2.00
P9 cycles	6.67
P10 cycles	6.67
P11 cycles	6.67
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	28.00
Nb uops	28.00
Nb loads	NA
Nb stores	4.00
Nb stack references	0.00
FLOP/cycle	1.20
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	24.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	32.00
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	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
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	25.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Metric	Value
CQA speedup if no scalar integer	1.25
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	1.90
Bottlenecks	P10, P11, P12,
Function	.omp_outlined..26#0x49c1a0
Source	csr_matvec.c:564-567
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	4
CQA cycles	6.67
CQA cycles if no scalar integer	5.33
CQA cycles if FP arith vectorized	6.67
CQA cycles if fully vectorized	1.67
Front-end cycles	3.50
DIV/SQRT cycles	0.50
P0 cycles	0.50
P1 cycles	0.75
P2 cycles	0.75
P3 cycles	0.75
P4 cycles	0.75
P5 cycles	2.00
P6 cycles	2.00
P7 cycles	2.00
P8 cycles	2.00
P9 cycles	6.67
P10 cycles	6.67
P11 cycles	6.67
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	28.00
Nb uops	28.00
Nb loads	NA
Nb stores	4.00
Nb stack references	0.00
FLOP/cycle	1.20
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	24.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	32.00
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	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
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	25.00
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Path /

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

Function	.omp_outlined..26#0x49c1a0
Source file and lines	csr_matvec.c:564-567
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:564-567.

The related source loop is unrolled by 4 (including vectorization).

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 6.67 to 5.33 cycles (1.25x speedup).

Workaround

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

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 6.67 to 1.67 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)

FMA

Detected 4 FMA (fused multiply-add) operations.

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

The binary loop is composed of 8 FP arithmetical operations:

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

The binary loop is loading 128 bytes. The binary loop is storing 32 bytes.

Arithmetic intensity

Arithmetic intensity is 0.05 FP operations per loaded or stored byte.

General properties

nb instructions	28
loop length	112
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

3.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.75	0.75	0.75	0.75	2.00	2.00	2.00	2.00	6.67	6.67	6.67	0.00	0.00
cycles	0.50	0.50	0.75	0.75	0.75	0.75	2.00	2.00	2.00	2.00	6.67	6.67	6.67	0.00	0.00

Execution ports to units layout:

P0:
P1:
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)	1.00

Cycles summary

Front-end	3.50
Data deps.	1.00
Overall L1	6.67

Vectorization ratios

all	0%
load	0%
store	0%
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	NA (no other vectorizable/vectorized instructions)

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 6.67 cycles. At this rate:

20% of peak load performance is reached (19.20 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
10% of peak store performance is reached (4.80 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 49c4c4

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput
LDUR X3, [X1, #496]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDUR D0, [X2, #496]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
SUBS X0, X0, #4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDUR X3, [X1, #504]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDUR D0, [X2, #504]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDR X3, [X1]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR D0, [X2]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDR X3, [X1, #8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR D0, [X2, #8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
ADD X1, X1, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25
ADD X2, X2, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
B.NE 49c4c4 <.omp_outlined..26+0x324>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Function	.omp_outlined..26#0x49c1a0
Source file and lines	csr_matvec.c:564-567
Module	exec

The loop is defined in /home/hbollore/qaas-runs/171-233-5044/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:564-567.

The related source loop is unrolled by 4 (including vectorization).

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 6.67 to 5.33 cycles (1.25x speedup).

Workaround

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

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 6.67 to 1.67 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)

FMA

Detected 4 FMA (fused multiply-add) operations.

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

The binary loop is composed of 8 FP arithmetical operations:

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

The binary loop is loading 128 bytes. The binary loop is storing 32 bytes.

Arithmetic intensity

Arithmetic intensity is 0.05 FP operations per loaded or stored byte.

General properties

nb instructions	28
loop length	112
nb stack references	0

Front-end

FIT IN UOP CACHE

front end

3.50 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14
uops	0.50	0.50	0.75	0.75	0.75	0.75	2.00	2.00	2.00	2.00	6.67	6.67	6.67	0.00	0.00
cycles	0.50	0.50	0.75	0.75	0.75	0.75	2.00	2.00	2.00	2.00	6.67	6.67	6.67	0.00	0.00

Execution ports to units layout:

P0:
P1:
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)	1.00

Cycles summary

Front-end	3.50
Data deps.	1.00
Overall L1	6.67

Vectorization ratios

all	0%
load	0%
store	0%
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	NA (no other vectorizable/vectorized instructions)

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 6.67 cycles. At this rate:

20% of peak load performance is reached (19.20 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
10% of peak store performance is reached (4.80 out of 48.00 bytes stored per cycle (GB/s @ 1GHz))

ASM code

In the binary file, the address of the loop is: 49c4c4

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	Latency	Recip. throughput
LDUR X3, [X1, #496]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDUR D0, [X2, #496]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
SUBS X0, X0, #4	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDUR X3, [X1, #504]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDUR D0, [X2, #504]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDR X3, [X1]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR D0, [X2]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
LDR X3, [X1, #8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	4	0.33
LDR D0, [X2, #8]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
LDR D1, [X13, X9,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
ADD X1, X1, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25
ADD X2, X2, #32	1	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	1	0.25
LDR D2, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0	0	0	0	0.33	0.33	0.33	6	0.33
FMADD D0, D0, D1, D2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	4	0.25
STR D0, [X14, X3,LSL #3]	1	0	0	0	0	0	0	0.50	0.50	0	0	0.50	0.50	0	2	0.50
B.NE 49c4c4 <.omp_outlined..26+0x324>	1	0.50	0.50	0	0	0	0	0	0	0	0	0	0	0	1	0.50

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

FMA

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

FMA

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Cycles and memory resources usage

ASM code