OV - exec - Loop 2979

0x54a17e VMOVSD	(%R8,%RDX,8),%XMM8    [1]

0x54a184 VMOVSD	0x8(%R8,%RDX,8),%XMM9    [1]

0x54a18b VFMADD231SD	(%R9,%RDX,8),%XMM8,%XMM0    [2]

0x54a191 VMOVSD	0x10(%R8,%RDX,8),%XMM10    [1]

0x54a198 VMOVSD	0x18(%R8,%RDX,8),%XMM11    [1]

0x54a19f VMOVSD	0x20(%R8,%RDX,8),%XMM12    [1]

0x54a1a6 VMOVSD	0x28(%R8,%RDX,8),%XMM13    [1]

0x54a1ad VFMADD231SD	0x8(%R9,%RDX,8),%XMM9,%XMM0    [2]

0x54a1b4 VMOVSD	0x30(%R8,%RDX,8),%XMM14    [1]

0x54a1bb VMOVSD	0x38(%R8,%RDX,8),%XMM15    [1]

0x54a1c2 VFMADD231SD	0x10(%R9,%RDX,8),%XMM10,%XMM0    [2]

0x54a1c9 VFMADD231SD	0x18(%R9,%RDX,8),%XMM11,%XMM0    [2]

0x54a1d0 VFMADD231SD	0x20(%R9,%RDX,8),%XMM12,%XMM0    [2]

0x54a1d7 VFMADD231SD	0x28(%R9,%RDX,8),%XMM13,%XMM0    [2]

0x54a1de VFMADD231SD	0x30(%R9,%RDX,8),%XMM14,%XMM0    [2]

0x54a1e5 VFMADD231SD	0x38(%R9,%RDX,8),%XMM15,%XMM0    [2]

0x54a1ec ADD	$0x8,%RDX

0x54a1f0 CMP	%RDX,%RAX

0x54a1f3 JNE	54a17e

/home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/seq_mv/vector.c: 486 - 486

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

486:       result += hypre_conj(y_data[i]) * x_data[i];

Coverage (%)	Name	Source Location	Module
○97.65	gomp_thread_start	team.c:130	libgomp.so.1.0.0
○2.35	GOMP_parallel	libgomp.h:982	libgomp.so.1.0.0

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.48
CQA speedup if fully vectorized	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	6.00
Bottlenecks
Function	hypre_SeqVectorInnerProd._omp_fn.0
Source	vector.c:486-486
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	32.00
CQA cycles if no scalar integer	32.00
CQA cycles if FP arith vectorized	21.68
CQA cycles if fully vectorized	4.00
Front-end cycles	4.33
P0 cycles	4.00
P1 cycles	4.00
P2 cycles	5.33
P3 cycles	5.33
P4 cycles	0.00
P5 cycles	0.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	5.33
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	32
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	19.00
Nb uops	18.00
Nb loads	16.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.50
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	8.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	4.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
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	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	12.50
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	NA

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.48
CQA speedup if fully vectorized	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	6.00
Bottlenecks
Function	hypre_SeqVectorInnerProd._omp_fn.0
Source	vector.c:486-486
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	32.00
CQA cycles if no scalar integer	32.00
CQA cycles if FP arith vectorized	21.68
CQA cycles if fully vectorized	4.00
Front-end cycles	4.33
P0 cycles	4.00
P1 cycles	4.00
P2 cycles	5.33
P3 cycles	5.33
P4 cycles	0.00
P5 cycles	0.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	5.33
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	32
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	19.00
Nb uops	18.00
Nb loads	16.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.50
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	8.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	4.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
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	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	12.50
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	12.50
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	hypre_SeqVectorInnerProd._omp_fn.0
Source file and lines	vector.c:486-486
Module	exec

The loop is defined in /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/seq_mv/vector.c:486.

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 8 data elements could be processed at once in vector registers.

By vectorizing your loop, you can lower the cost of an iteration from 32.00 to 4.00 cycles (8.00x speedup).

Details

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

Workaround

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

Execution units bottlenecks

Found no such bottlenecks but see expert reports for more complex bottlenecks.

FMA

Detected 8 FMA (fused multiply-add) operations.

Type of elements and instruction set

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

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

The binary loop is composed of 16 FP arithmetical operations:

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

The binary loop is loading 128 bytes (16 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.12 FP operations per loaded or stored byte.

Unroll opportunity

Loop is data access bound.

Workaround

Unroll your loop if trip count is significantly higher than target unroll factor and if some data references are common to consecutive iterations. This can be done manually. Or by recompiling with -funroll-loops and/or -floop-unroll-and-jam. Or with the unroll (resp. unroll_and_jam) directive on top of the inner (resp. surrounding) loop. You can enforce an unroll factor: #pragma GCC unroll N

General properties

nb instructions	19
nb uops	18
loop length	119
used x86 registers	4
used mmx registers	0
used xmm registers	9
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	4.33 cycles
front end	4.33 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	4.00	4.00	5.33	5.33	0.00	0.00	1.00	0.00	0.00	0.00	0.00	5.33
cycles	4.00	4.00	5.33	5.33	0.00	0.00	1.00	0.00	0.00	0.00	0.00	5.33

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	32.00

Cycles summary

Front-end	4.33
Dispatch	5.33
Data deps.	32.00
Overall L1	32.00

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

Vector efficiency ratios

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	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 32.00 cycles. At this rate:

2% of peak load performance is reached (4.00 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 54a17e

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
VMOVSD (%R8,%RDX,8),%XMM8	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x8(%R8,%RDX,8),%XMM9	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD (%R9,%RDX,8),%XMM8,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVSD 0x10(%R8,%RDX,8),%XMM10	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x18(%R8,%RDX,8),%XMM11	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x20(%R8,%RDX,8),%XMM12	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x28(%R8,%RDX,8),%XMM13	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD 0x8(%R9,%RDX,8),%XMM9,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVSD 0x30(%R8,%RDX,8),%XMM14	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x38(%R8,%RDX,8),%XMM15	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD 0x10(%R9,%RDX,8),%XMM10,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x18(%R9,%RDX,8),%XMM11,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x20(%R9,%RDX,8),%XMM12,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x28(%R9,%RDX,8),%XMM13,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x30(%R9,%RDX,8),%XMM14,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x38(%R9,%RDX,8),%XMM15,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
ADD $0x8,%RDX	1	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %RDX,%RAX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 54a17e <hypre_SeqVectorInnerProd._omp_fn.0+0xee>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Function	hypre_SeqVectorInnerProd._omp_fn.0
Source file and lines	vector.c:486-486
Module	exec

The loop is defined in /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/seq_mv/vector.c:486.

The related source loop is not unrolled or unrolled with no peel/tail loop.

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. 8 data elements could be processed at once in vector registers.

By vectorizing your loop, you can lower the cost of an iteration from 32.00 to 4.00 cycles (8.00x speedup).

Details

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

Workaround

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

Execution units bottlenecks

Found no such bottlenecks but see expert reports for more complex bottlenecks.

FMA

Detected 8 FMA (fused multiply-add) operations.

Type of elements and instruction set

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

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

The binary loop is composed of 16 FP arithmetical operations:

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

The binary loop is loading 128 bytes (16 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.12 FP operations per loaded or stored byte.

Unroll opportunity

Loop is data access bound.

Workaround

Unroll your loop if trip count is significantly higher than target unroll factor and if some data references are common to consecutive iterations. This can be done manually. Or by recompiling with -funroll-loops and/or -floop-unroll-and-jam. Or with the unroll (resp. unroll_and_jam) directive on top of the inner (resp. surrounding) loop. You can enforce an unroll factor: #pragma GCC unroll N

General properties

nb instructions	19
nb uops	18
loop length	119
used x86 registers	4
used mmx registers	0
used xmm registers	9
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	4.33 cycles
front end	4.33 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	4.00	4.00	5.33	5.33	0.00	0.00	1.00	0.00	0.00	0.00	0.00	5.33
cycles	4.00	4.00	5.33	5.33	0.00	0.00	1.00	0.00	0.00	0.00	0.00	5.33

Execution ports to units layout:

P0 (256 bits): VPU, BRU, ALU, DIV/SQRT
P1 (256 bits): ALU, VPU
P2 (512 bits): load
P3 (512 bits): load
P4 (512 bits): store data
P5 (512 bits): ALU, VPU
P6: ALU, BRU
P7: store address
P8: store address
P9 (512 bits): store data
P10: ALU
P11 (512 bits): load

Cycles executing div or sqrt instructions	NA
Longest recurrence chain latency (RecMII)	32.00

Cycles summary

Front-end	4.33
Dispatch	5.33
Data deps.	32.00
Overall L1	32.00

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

Vector efficiency ratios

all	12%
load	12%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	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 32.00 cycles. At this rate:

2% of peak load performance is reached (4.00 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 54a17e

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
VMOVSD (%R8,%RDX,8),%XMM8	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x8(%R8,%RDX,8),%XMM9	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD (%R9,%RDX,8),%XMM8,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVSD 0x10(%R8,%RDX,8),%XMM10	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x18(%R8,%RDX,8),%XMM11	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x20(%R8,%RDX,8),%XMM12	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x28(%R8,%RDX,8),%XMM13	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD 0x8(%R9,%RDX,8),%XMM9,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VMOVSD 0x30(%R8,%RDX,8),%XMM14	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD 0x38(%R8,%RDX,8),%XMM15	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (12.5%)
VFMADD231SD 0x10(%R9,%RDX,8),%XMM10,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x18(%R9,%RDX,8),%XMM11,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x20(%R9,%RDX,8),%XMM12,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x28(%R9,%RDX,8),%XMM13,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x30(%R9,%RDX,8),%XMM14,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
VFMADD231SD 0x38(%R9,%RDX,8),%XMM15,%XMM0	1	0.50	0.50	0.33	0.33	0	0	0	0.33	4	0.50	scal (12.5%)
ADD $0x8,%RDX	1	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %RDX,%RAX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 54a17e <hypre_SeqVectorInnerProd._omp_fn.0+0xee>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

FMA

Type of elements and instruction set

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

Arithmetic intensity

Unroll opportunity

Workaround

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

Execution units bottlenecks

FMA

Type of elements and instruction set

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

Arithmetic intensity

Unroll opportunity

Workaround

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code