OV - exec - Loop 3633

	Loop Id: 3633	Module: exec	Source: csr_matvec.c:256-263	Coverage: 0.57%

0x2f8170 VMOVSD	%XMM0,(%R8,%RBX,8)

0x2f8176 INC	%RBX

0x2f8179 CMP	%R14,%RBX

0x2f817c JE	2f990d

0x2f8182 MOV	%R9,%R11

0x2f8185 MOV	0x8(%RDX,%RBX,8),%R9

0x2f818a VXORPD	%XMM0,%XMM0,%XMM0

0x2f818e MOV	%R9,%R15

0x2f8191 SUB	%R11,%R15

0x2f8194 JLE	2f8170

0x2f8196 CMP	$0x10,%R15

0x2f819a JB	2f82e0

0x2f81a0 MOV	-0x30(%RBP),%R8

0x2f81a4 MOV	-0x38(%RBP),%RAX

0x2f81a8 MOV	%R15,%R12

0x2f81ab AND	$-0x10,%R12

0x2f81af VXORPD	%XMM0,%XMM0,%XMM0

0x2f81b3 VXORPD	%XMM1,%XMM1,%XMM1

0x2f81b7 VXORPD	%XMM2,%XMM2,%XMM2

0x2f81bb VXORPD	%XMM3,%XMM3,%XMM3

0x2f81bf XOR	%R10D,%R10D

0x2f81c2 LEA	(%RAX,%R11,8),%R13

0x2f81c6 LEA	(%R8,%R11,8),%R8

0x2f81ca ADD	%R12,%R11

0x2f81cd NOPL	(%RAX)

  (3635) 0x2f81d0 MOV	-0x50(%R8,%R10,8),%RAX

  (3635) 0x2f81d5 VMOVSD	(%RDI,%RAX,8),%XMM4

  (3635) 0x2f81da MOV	-0x48(%R8,%R10,8),%RAX

  (3635) 0x2f81df VMOVHPD	(%RDI,%RAX,8),%XMM4,%XMM4

  (3635) 0x2f81e4 MOV	-0x60(%R8,%R10,8),%RAX

  (3635) 0x2f81e9 VMOVSD	(%RDI,%RAX,8),%XMM5

  (3635) 0x2f81ee MOV	-0x58(%R8,%R10,8),%RAX

  (3635) 0x2f81f3 VMOVHPD	(%RDI,%RAX,8),%XMM5,%XMM5

  (3635) 0x2f81f8 MOV	-0x30(%R8,%R10,8),%RAX

  (3635) 0x2f81fd VMOVSD	(%RDI,%RAX,8),%XMM6

  (3635) 0x2f8202 MOV	-0x28(%R8,%R10,8),%RAX

  (3635) 0x2f8207 VINSERTF128	$0x1,%XMM4,%YMM5,%YMM4

  (3635) 0x2f820d VFMADD231PD	-0x60(%R13,%R10,8),%YMM4,%YMM0

  (3635) 0x2f8214 VMOVHPD	(%RDI,%RAX,8),%XMM6,%XMM6

  (3635) 0x2f8219 MOV	-0x40(%R8,%R10,8),%RAX

  (3635) 0x2f821e VMOVSD	(%RDI,%RAX,8),%XMM7

  (3635) 0x2f8223 MOV	-0x38(%R8,%R10,8),%RAX

  (3635) 0x2f8228 VMOVHPD	(%RDI,%RAX,8),%XMM7,%XMM7

  (3635) 0x2f822d MOV	-0x10(%R8,%R10,8),%RAX

  (3635) 0x2f8232 VMOVSD	(%RDI,%RAX,8),%XMM8

  (3635) 0x2f8237 MOV	-0x8(%R8,%R10,8),%RAX

  (3635) 0x2f823c VINSERTF128	$0x1,%XMM6,%YMM7,%YMM5

  (3635) 0x2f8242 VFMADD231PD	-0x40(%R13,%R10,8),%YMM5,%YMM1

  (3635) 0x2f8249 VMOVHPD	(%RDI,%RAX,8),%XMM8,%XMM8

  (3635) 0x2f824e MOV	-0x20(%R8,%R10,8),%RAX

  (3635) 0x2f8253 VMOVSD	(%RDI,%RAX,8),%XMM9

  (3635) 0x2f8258 MOV	-0x18(%R8,%R10,8),%RAX

  (3635) 0x2f825d VMOVHPD	(%RDI,%RAX,8),%XMM9,%XMM9

  (3635) 0x2f8262 MOV	0x10(%R8,%R10,8),%RAX

  (3635) 0x2f8267 VMOVSD	(%RDI,%RAX,8),%XMM10

  (3635) 0x2f826c MOV	0x18(%R8,%R10,8),%RAX

  (3635) 0x2f8271 VINSERTF128	$0x1,%XMM8,%YMM9,%YMM6

  (3635) 0x2f8277 VFMADD231PD	-0x20(%R13,%R10,8),%YMM6,%YMM2

  (3635) 0x2f827e VMOVHPD	(%RDI,%RAX,8),%XMM10,%XMM10

  (3635) 0x2f8283 MOV	(%R8,%R10,8),%RAX

  (3635) 0x2f8287 VMOVSD	(%RDI,%RAX,8),%XMM11

  (3635) 0x2f828c MOV	0x8(%R8,%R10,8),%RAX

  (3635) 0x2f8291 VMOVHPD	(%RDI,%RAX,8),%XMM11,%XMM11

  (3635) 0x2f8296 VINSERTF128	$0x1,%XMM10,%YMM11,%YMM7

  (3635) 0x2f829c VFMADD231PD	(%R13,%R10,8),%YMM7,%YMM3

  (3635) 0x2f82a3 ADD	$0x10,%R10

  (3635) 0x2f82a7 CMP	%R10,%R12

  (3635) 0x2f82aa JNE	2f81d0

0x2f82b0 VADDPD	%YMM0,%YMM1,%YMM0

0x2f82b4 VADDPD	%YMM2,%YMM3,%YMM1

0x2f82b8 MOV	-0x40(%RBP),%R8

0x2f82bc VADDPD	%YMM0,%YMM1,%YMM0

0x2f82c0 VEXTRACTF128	$0x1,%YMM0,%XMM1

0x2f82c6 VADDPD	%XMM1,%XMM0,%XMM0

0x2f82ca VSHUFPD	$0x1,%XMM0,%XMM0,%XMM1

0x2f82cf VADDSD	%XMM1,%XMM0,%XMM0

0x2f82d3 CMP	%R12,%R15

0x2f82d6 JE	2f8170

0x2f82dc NOPL	(%RAX)

  (3634) 0x2f82e0 MOV	(%RCX,%R11,8),%RAX

  (3634) 0x2f82e4 VMOVSD	(%RDI,%RAX,8),%XMM1

  (3634) 0x2f82e9 VFMADD231SD	(%RSI,%R11,8),%XMM1,%XMM0

  (3634) 0x2f82ef INC	%R11

  (3634) 0x2f82f2 CMP	%R11,%R9

  (3634) 0x2f82f5 JNE	2f82e0

0x2f82f7 JMP	2f8170

/home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c: 256 - 263

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

256:             for (i = iBegin; i < iEnd; i++)

257:             {

258:                tempx = 0.0;

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

260:                {

261:                   tempx += A_data[jj] * x_data[A_j[jj]];

262:                }

263:                y_data[i] = tempx;

Coverage (%)	Name	Source Location	Module

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	1.61
CQA speedup if FP arith vectorized	2.10
CQA speedup if fully vectorized	5.07
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.61
Bottlenecks
Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source	csr_matvec.c:256-256,csr_matvec.c:259-259,csr_matvec.c:263-263
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	2.81
CQA cycles if no scalar integer	1.75
CQA cycles if FP arith vectorized	1.34
CQA cycles if fully vectorized	0.55
Front-end cycles	2.81
P0 cycles	1.58
P1 cycles	1.58
P2 cycles	1.58
P3 cycles	1.58
P4 cycles	1.58
P5 cycles	1.58
P6 cycles	0.88
P7 cycles	0.88
P8 cycles	0.88
P9 cycles	0.88
P10 cycles	0.50
P11 cycles	0.50
P12 cycles	1.25
P13 cycles	1.25
P14 cycles	0.50
P15 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	23.75
Nb uops	22.50
Nb loads	2.50
Nb stores	1.00
Nb stack references	1.50
FLOP/cycle	2.67
Nb FLOP add-sub	7.50
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	10.73
Bytes prefetched	0.00
Bytes loaded	20.00
Bytes stored	8.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	43.60
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	80.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	55.83
Vector-efficiency ratio all	19.97
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	37.50
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	19.17

Metric	Value
CQA speedup if no scalar integer	1.50
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.69
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source	csr_matvec.c:256-256,csr_matvec.c:259-259,csr_matvec.c:263-263
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	1.13
CQA cycles if no scalar integer	0.75
CQA cycles if FP arith vectorized	1.13
CQA cycles if fully vectorized	0.14
Front-end cycles	1.13
P0 cycles	0.67
P1 cycles	0.67
P2 cycles	0.67
P3 cycles	0.67
P4 cycles	0.67
P5 cycles	0.67
P6 cycles	0.50
P7 cycles	0.50
P8 cycles	0.50
P9 cycles	0.50
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.50
P15 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	10.00
Nb uops	9.00
Nb loads	1.00
Nb stores	1.00
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	14.22
Bytes prefetched	0.00
Bytes loaded	8.00
Bytes stored	8.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	25.00
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	50.00
Vector-efficiency ratio all	15.63
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
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	18.75

Metric	Value
CQA speedup if no scalar integer	1.63
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	8.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.22
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source	csr_matvec.c:256-256,csr_matvec.c:259-259,csr_matvec.c:263-263
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	1.00
CQA cycles if FP arith vectorized	1.63
CQA cycles if fully vectorized	0.20
Front-end cycles	1.63
P0 cycles	1.33
P1 cycles	1.33
P2 cycles	1.33
P3 cycles	1.33
P4 cycles	1.33
P5 cycles	1.33
P6 cycles	0.50
P7 cycles	0.50
P8 cycles	0.50
P9 cycles	0.50
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.50
P15 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	13.00
Nb uops	13.00
Nb loads	1.00
Nb stores	1.00
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	9.85
Bytes prefetched	0.00
Bytes loaded	8.00
Bytes stored	8.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	2.00
Stride indirect	1.00
Vectorization ratio all	20.00
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	33.33
Vector-efficiency ratio all	15.00
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
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	16.67

Metric	Value
CQA speedup if no scalar integer	1.67
CQA speedup if FP arith vectorized	3.33
CQA speedup if fully vectorized	4.67
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.75
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source	csr_matvec.c:256-256,csr_matvec.c:259-259,csr_matvec.c:263-263
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.38
CQA cycles if no scalar integer	2.63
CQA cycles if FP arith vectorized	1.32
CQA cycles if fully vectorized	0.94
Front-end cycles	4.38
P0 cycles	2.33
P1 cycles	2.33
P2 cycles	2.33
P3 cycles	2.33
P4 cycles	2.33
P5 cycles	2.33
P6 cycles	1.25
P7 cycles	1.25
P8 cycles	1.25
P9 cycles	1.25
P10 cycles	1.00
P11 cycles	1.00
P12 cycles	2.50
P13 cycles	2.50
P14 cycles	0.50
P15 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	37.00
Nb uops	35.00
Nb loads	4.00
Nb stores	1.00
Nb stack references	3.00
FLOP/cycle	3.43
Nb FLOP add-sub	15.00
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	9.14
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	8.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	4.00
Stride indirect	2.00
Vectorization ratio all	64.71
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	80.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	70.00
Vector-efficiency ratio all	24.63
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	37.50
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	20.63

Metric	Value
CQA speedup if no scalar integer	1.57
CQA speedup if FP arith vectorized	3.22
CQA speedup if fully vectorized	4.40
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.65
Bottlenecks	micro-operation queue,
Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source	csr_matvec.c:256-256,csr_matvec.c:259-259,csr_matvec.c:263-263
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	4.13
CQA cycles if no scalar integer	2.63
CQA cycles if FP arith vectorized	1.28
CQA cycles if fully vectorized	0.94
Front-end cycles	4.13
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	1.25
P7 cycles	1.25
P8 cycles	1.25
P9 cycles	1.25
P10 cycles	1.00
P11 cycles	1.00
P12 cycles	2.50
P13 cycles	2.50
P14 cycles	0.50
P15 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	0
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	35.00
Nb uops	33.00
Nb loads	4.00
Nb stores	1.00
Nb stack references	3.00
FLOP/cycle	3.64
Nb FLOP add-sub	15.00
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	9.70
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	8.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	2.00
Stride indirect	1.00
Vectorization ratio all	64.71
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	80.00
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	70.00
Vector-efficiency ratio all	24.63
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	12.50
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	37.50
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	20.63

Path /

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

Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source file and lines	csr_matvec.c:256-263
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:256-263.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:256,259,263.

Warnings:
Non-innermost loop: analyzing only self part (ignoring child loops).
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 poorly vectorized. Only 19% of vector register length is used (average across all SSE/AVX instructions).

Details

43% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

0% of SSE/AVX loads are used in vector version.
0% of SSE/AVX stores are used in vector version.
80% of SSE/AVX addition or subtraction instructions are used in vector version.
55% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Workaround

Read the "512-bits vectorization" report at "Potential" confidence level.

512-bits vectorization

On some x86 processors supporting 512-bits vectorization, compilers are often too conservative and limit vectorization to 256 bits. Performance can then be improved by enforcing 512-bits vectorization, especially with many vectorized and high trip count loops. 512-bits vectorization performance overhead (compared to 256-bits) is generally lower on newer processors.

Workaround

Recompile with -mprefer-vector-width=512

Type of elements and instruction set

0 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time). 0 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in vector mode (two at a time). 1 AVX instructions are processing arithmetic or math operations on double precision FP elements in vector mode (four at a time).

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

The binary loop is composed of 7.5 FP arithmetical operations:

7.50: addition or subtraction

The binary loop is loading 20 bytes (2 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.27 FP operations per loaded or stored byte.

General properties

nb instructions	23.75
nb uops	22.50
loop length	95
used x86 registers	9.50
used mmx registers	0
used xmm registers	2.50
used ymm registers	2
used zmm registers	0
nb stack references	1.50

Front-end

MACRO FUSION NOT POSSIBLE

micro-operation queue	2.81 cycles
front end	2.81 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	1.58	1.58	1.58	1.58	1.58	1.58	0.88	0.88	0.88	0.88	0.50	0.50	1.25	1.25	0.50	0.50
cycles	1.58	1.58	1.58	1.58	1.58	1.58	0.88	0.88	0.88	0.88	0.50	0.50	1.25	1.25	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

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

Cycles summary

Front-end	2.81
Dispatch	1.75
Data deps.	0.00
Overall L1	2.81

Vectorization ratios

INT

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)
other	0%

all	67%
load	NA (no load vectorizable/vectorized instructions)
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	43%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	55%

Vector efficiency ratios

INT

all	11%
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	NA (no fma vectorizable/vectorized instructions)
other	11%

all	23%
load	NA (no load vectorizable/vectorized instructions)
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	19%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	19%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 2.81 cycles. At this rate:

2% of peak load performance is reached (6.78 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
3% of peak store performance is reached (3.95 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source file and lines	csr_matvec.c:256-263
Module	exec

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

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

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 1.13 to 0.75 cycles (1.50x speedup).

Workaround

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

Vectorization

Your loop is probably not vectorized. Only 15% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 1.13 to 0.14 cycles (8.00x speedup).

Details

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

Workaround

Try another compiler or update/tune your current one
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.

No data for this section

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

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 8 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	10
nb uops	9
loop length	38
used x86 registers	7
used mmx registers	0
used xmm registers	1
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	1.13 cycles
front end	1.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.67	0.67	0.67	0.67	0.67	0.67	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	0.50	0.50
cycles	0.67	0.67	0.67	0.67	0.67	0.67	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

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

Cycles summary

Front-end	1.13
Dispatch	0.67
Data deps.	0.00
Overall L1	1.13

Vectorization ratios

INT

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)
other	0%

all	50%
load	NA (no load vectorizable/vectorized instructions)
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	100%

INT+FP

all	25%
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	50%

Vector efficiency ratios

INT

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

all	18%
load	NA (no load vectorizable/vectorized instructions)
store	12%
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%

INT+FP

all	15%
load	12%
store	12%
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	18%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 1.13 cycles. At this rate:

2% of peak load performance is reached (7.11 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
5% of peak store performance is reached (7.11 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

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 1.13 to 0.67 cycles (1.69x speedup).

ASM code

In the binary file, the address of the loop is: 2f8182

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P14	P15	Latency	Recip. throughput	Vectorization
VMOVSD %XMM0,(%R8,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	1	0.50	scal (12.5%)
INC %RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 2f990d <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x190d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
MOV %R9,%R11	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
MOV 0x8(%RDX,%RBX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	scal (12.5%)
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOV %R9,%R15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
SUB %R11,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
JLE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A

Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source file and lines	csr_matvec.c:256-263
Module	exec

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

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

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 1.63 to 1.00 cycles (1.62x speedup).

Workaround

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

Vectorization

Details

Workaround

Try another compiler or update/tune your current one
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.

No data for this section

Slow data structures access

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

Details

Constant unknown stride: 2 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.

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

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 8 bytes. The binary loop is storing 8 bytes.

General properties

nb instructions	13
nb uops	13
loop length	53
used x86 registers	7
used mmx registers	0
used xmm registers	1
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	1.63 cycles
front end	1.63 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	1.33	1.33	1.33	1.33	1.33	1.33	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	0.50	0.50
cycles	1.33	1.33	1.33	1.33	1.33	1.33	0.50	0.50	0.50	0.50	0.00	0.00	0.00	0.00	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

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

Cycles summary

Front-end	1.63
Dispatch	1.33
Data deps.	0.00
Overall L1	1.63

Vectorization ratios

INT

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)
other	0%

all	50%
load	NA (no load vectorizable/vectorized instructions)
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	100%

INT+FP

all	20%
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	33%

Vector efficiency ratios

INT

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

all	18%
load	NA (no load vectorizable/vectorized instructions)
store	12%
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%

INT+FP

all	15%
load	12%
store	12%
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	16%

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:

1% of peak load performance is reached (4.92 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
3% of peak store performance is reached (4.92 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 2f8182

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P14	P15	Latency	Recip. throughput	Vectorization
VMOVSD %XMM0,(%R8,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	1	0.50	scal (12.5%)
INC %RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 2f990d <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x190d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
MOV %R9,%R11	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	scal (12.5%)
MOV 0x8(%RDX,%RBX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	3	0.25	scal (12.5%)
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOV %R9,%R15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
SUB %R11,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
JLE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
CMP $0x10,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	1	0.17	N/A
JB 2f82e0 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x2e0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	1	0.50-1	N/A
JMP 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	1	N/A

Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source file and lines	csr_matvec.c:256-263
Module	exec

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

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

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 4.38 to 2.63 cycles (1.67x speedup).

Workaround

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

Vectorization

Your loop is partially vectorized. Only 24% of vector register length is used (average across all SSE/AVX instructions).

Details

64% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

0% of SSE/AVX loads are used in vector version.
0% of SSE/AVX stores are used in vector version.
80% of SSE/AVX addition or subtraction instructions are used in vector version.
70% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Workaround

Read the "512-bits vectorization" report at "Potential" confidence level.

Execution units bottlenecks

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

512-bits vectorization

Workaround

Recompile with -mprefer-vector-width=512

Slow data structures access

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

Details

Constant unknown stride: 4 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.

Vector unaligned load/store instructions

Detected 1 suboptimal vector unaligned load/store instructions.

Details

VEXTRACTF128: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:259

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Type of elements and instruction set

1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time). 1 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in vector mode (two at a time). 3 AVX instructions are processing arithmetic or math operations on double precision FP elements in vector mode (four at a time).

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

The binary loop is composed of 15 FP arithmetical operations:

15: addition or subtraction

The binary loop is loading 32 bytes (4 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.37 FP operations per loaded or stored byte.

General properties

nb instructions	37
nb uops	35
loop length	149
used x86 registers	12
used mmx registers	0
used xmm registers	4
used ymm registers	4
used zmm registers	0
nb stack references	3

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.38 cycles
front end	4.38 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.33	2.33	2.33	2.33	2.33	2.33	1.25	1.25	1.25	1.25	1.00	1.00	2.50	2.50	0.50	0.50
cycles	2.33	2.33	2.33	2.33	2.33	2.33	1.25	1.25	1.25	1.25	1.00	1.00	2.50	2.50	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

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

Cycles summary

Front-end	4.38
Dispatch	2.50
Data deps.	0.00
Overall L1	4.38

Vectorization ratios

INT

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)
other	0%

all	84%
load	NA (no load vectorizable/vectorized instructions)
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	64%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	70%

Vector efficiency ratios

INT

all	10%
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	NA (no fma vectorizable/vectorized instructions)
other	10%

all	28%
load	NA (no load vectorizable/vectorized instructions)
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	24%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	20%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.38 cycles. At this rate:

2% of peak load performance is reached (7.31 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
1% of peak store performance is reached (1.83 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 2f8182

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VMOVSD %XMM0,(%R8,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
INC %RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 2f990d <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x190d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
MOV %R9,%R11	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
MOV 0x8(%RDX,%RBX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	scal (12.5%)
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOV %R9,%R15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
SUB %R11,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JLE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
CMP $0x10,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JB 2f82e0 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x2e0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
MOV -0x30(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV -0x38(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %R15,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
AND $-0x10,%R12	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM2,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM3,%XMM3,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
XOR %R10D,%R10D	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	scal (6.3%)
LEA (%RAX,%R11,8),%R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.25	N/A
LEA (%R8,%R11,8),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.25	N/A
ADD %R12,%R11	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
VADDPD %YMM0,%YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
VADDPD %YMM2,%YMM3,%YMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
MOV -0x40(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VADDPD %YMM0,%YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
VEXTRACTF128 $0x1,%YMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	3	0.50	vect (25.0%)
VADDPD %XMM1,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM0,%XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
VADDSD %XMM1,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (12.5%)
CMP %R12,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
JMP 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	1	N/A

Function	hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18
Source file and lines	csr_matvec.c:256-263
Module	exec

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

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

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 4.13 to 2.63 cycles (1.57x speedup).

Workaround

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

Vectorization

Your loop is partially vectorized. Only 24% of vector register length is used (average across all SSE/AVX instructions).

Details

64% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

0% of SSE/AVX loads are used in vector version.
0% of SSE/AVX stores are used in vector version.
80% of SSE/AVX addition or subtraction instructions are used in vector version.
70% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Workaround

Read the "512-bits vectorization" report at "Potential" confidence level.

Execution units bottlenecks

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

512-bits vectorization

Workaround

Recompile with -mprefer-vector-width=512

Slow data structures access

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

Details

Constant unknown stride: 2 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.

Vector unaligned load/store instructions

Detected 1 suboptimal vector unaligned load/store instructions.

Details

VEXTRACTF128: 1 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/seq_mv/csr_matvec.c:259

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Use vector aligned instructions:
1. align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
2. inform your compiler that your arrays are vector aligned: if array 'foo' is 64 bytes-aligned, define a pointer 'p_foo' as __builtin_assume_aligned (foo, 64) and use it instead of 'foo' in the loop.

Type of elements and instruction set

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

The binary loop is composed of 15 FP arithmetical operations:

15: addition or subtraction

The binary loop is loading 32 bytes (4 double precision FP elements). The binary loop is storing 8 bytes (1 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.37 FP operations per loaded or stored byte.

General properties

nb instructions	35
nb uops	33
loop length	140
used x86 registers	12
used mmx registers	0
used xmm registers	4
used ymm registers	4
used zmm registers	0
nb stack references	3

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	4.13 cycles
front end	4.13 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	2.00	2.00	2.00	2.00	2.00	2.00	1.25	1.25	1.25	1.25	1.00	1.00	2.50	2.50	0.50	0.50
cycles	2.00	2.00	2.00	2.00	2.00	2.00	1.25	1.25	1.25	1.25	1.00	1.00	2.50	2.50	0.50	0.50

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

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

Cycles summary

Front-end	4.13
Dispatch	2.50
Data deps.	0.00
Overall L1	4.13

Vectorization ratios

INT

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)
other	0%

all	84%
load	NA (no load vectorizable/vectorized instructions)
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	100%

INT+FP

all	64%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	80%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	70%

Vector efficiency ratios

INT

all	10%
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	NA (no fma vectorizable/vectorized instructions)
other	10%

all	28%
load	NA (no load vectorizable/vectorized instructions)
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	25%

INT+FP

all	24%
load	12%
store	12%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	37%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	20%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 4.13 cycles. At this rate:

3% of peak load performance is reached (7.76 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
1% of peak store performance is reached (1.94 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 2f8182

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VMOVSD %XMM0,(%R8,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (12.5%)
INC %RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 2f990d <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x190d>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
MOV %R9,%R11	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
MOV 0x8(%RDX,%RBX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	scal (12.5%)
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
MOV %R9,%R15	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
SUB %R11,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JLE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
CMP $0x10,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JB 2f82e0 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x2e0>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A
MOV -0x30(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV -0x38(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %R15,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
AND $-0x10,%R12	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VXORPD %XMM0,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM2,%XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VXORPD %XMM3,%XMM3,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
XOR %R10D,%R10D	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.25	scal (6.3%)
LEA (%RAX,%R11,8),%R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.25	N/A
LEA (%R8,%R11,8),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.25	N/A
ADD %R12,%R11	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
NOPL (%RAX)	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.09	N/A
VADDPD %YMM0,%YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
VADDPD %YMM2,%YMM3,%YMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
MOV -0x40(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VADDPD %YMM0,%YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (50.0%)
VEXTRACTF128 $0x1,%YMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	3	0.50	vect (25.0%)
VADDPD %XMM1,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM0,%XMM0,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	1	0.25	vect (25.0%)
VADDSD %XMM1,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (12.5%)
CMP %R12,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JE 2f8170 <hypre_CSRMatrixMatvecOutOfPlace.omp_outlined.18+0x170>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.50-1	N/A

Report Configuration

Vectorization

Details

Workaround

512-bits vectorization

Workaround

Type of elements and instruction set

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Type of elements and instruction set

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

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Slow data structures access

Details

Workaround

Type of elements and instruction set

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

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

512-bits vectorization

Workaround

Slow data structures access

Details

Workaround

Vector unaligned load/store instructions

Details

Workaround

Type of elements and instruction set

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

Arithmetic intensity

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Code clean check