OV - exec - Loop 20

0x4033a0 LEA	(%R11,%R14,8),%R15

0x4033a4 ADD	$0x4,%R14

0x4033a8 VMOVUPD	(%RDX,%R15,1),%YMM1    [3]

0x4033ae VFMADD213PD	(%RSI,%R15,1),%YMM0,%YMM1    [2]

0x4033b4 VMOVUPD	%YMM1,(%RDI,%R15,1)    [1]

0x4033ba CMP	%RCX,%R14

0x4033bd JB	4033a0

/beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c: 355 - 356

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

355: 	for (j=0; j<STREAM_ARRAY_SIZE; j++)

356: 	    a[j] = b[j]+scalar*c[j];

Coverage (%)	Name	Source Location	Module
►100.00+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so

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	1.20
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.33
Bottlenecks	micro-operation queue, P5, P6, P7,
Function	main.extracted.65
Source	stream.c:355-356
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	1.00
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.00
CQA cycles if fully vectorized	0.83
Front-end cycles	1.00
P0 cycles	0.75
P1 cycles	0.75
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	1.00
P8 cycles	0.50
P9 cycles	0.50
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.50
P13 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	7.00
Nb uops	6.00
Nb loads	2.00
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	8.00
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	96.00
Bytes prefetched	0.00
Bytes loaded	64.00
Bytes stored	32.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	100.00
Vectorization ratio load	100.00
Vectorization ratio store	100.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	50.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	50.00
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	1.00
CQA speedup if fully vectorized	1.20
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.33
Bottlenecks	micro-operation queue, P5, P6, P7,
Function	main.extracted.65
Source	stream.c:355-356
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	1.00
CQA cycles if no scalar integer	1.00
CQA cycles if FP arith vectorized	1.00
CQA cycles if fully vectorized	0.83
Front-end cycles	1.00
P0 cycles	0.75
P1 cycles	0.75
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	1.00
P6 cycles	1.00
P7 cycles	1.00
P8 cycles	0.50
P9 cycles	0.50
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.50
P13 cycles	0.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	1
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	7.00
Nb uops	6.00
Nb loads	2.00
Nb stores	1.00
Nb stack references	0.00
FLOP/cycle	8.00
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	96.00
Bytes prefetched	0.00
Bytes loaded	64.00
Bytes stored	32.00
Stride 0	0.00
Stride 1	0.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	100.00
Vectorization ratio load	100.00
Vectorization ratio store	100.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	NA
Vector-efficiency ratio all	50.00
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	50.00
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	50.00
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	main.extracted.65
Source file and lines	stream.c:355-356
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:355-356.

It is main loop of related source loop which is unrolled by 4 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using only 256 out of 512 bits (AVX/AVX2 instructions on AVX-512 processors).

Details

All SSE/AVX instructions are used in vector version (process two or more data elements in vector registers).

Workaround

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

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 1.00 to 0.75 cycles (1.33x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

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

FMA

Detected 4 FMA (fused multiply-add) operations.

Vector unaligned load/store instructions

Detected 2 optimal vector unaligned load/store instructions.

Details

VMOVUPD: 2 occurrences►
- /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:356
- /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:356

Workaround

Use vector aligned instructions:

align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
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 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 8 FP arithmetical operations:

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

The binary loop is loading 64 bytes (8 double precision FP elements). The binary loop is storing 32 bytes (4 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.08 FP operations per loaded or stored byte.

General properties

nb instructions	7
nb uops	6
loop length	31
used x86 registers	7
used mmx registers	0
used xmm registers	0
used ymm registers	2
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	1.00 cycles
front end	1.00 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	0.75	0.75	0.50	0.50	0.50	1.00	1.00	1.00	0.50	0.50	0.00	0.00	0.50	0.50
cycles	0.75	0.75	0.50	0.50	0.50	1.00	1.00	1.00	0.50	0.50	0.00	0.00	0.50	0.50

Execution ports to units layout:

ALU0/BRU0: ALU, BRU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1: BRU
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

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

Cycles summary

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

Vectorization ratios

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

Vector efficiency ratios

all	50%
load	50%
store	50%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	50%
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 1.00 cycles. At this rate:

66% of peak load performance is reached (64.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
50% of peak store performance is reached (32.00 out of 64.00 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 1.00 to 0.75 cycles (1.33x speedup).

ASM code

In the binary file, the address of the loop is: 4033a0

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP4	FP5	Latency	Recip. throughput	Vectorization
LEA (%R11,%R14,8),%R15	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
ADD $0x4,%R14	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
VMOVUPD (%RDX,%R15,1),%YMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.50	vect (50.0%)
VFMADD213PD (%RSI,%R15,1),%YMM0,%YMM1	1	0	0	0	0	0	0.33	0.33	0.33	0.50	0.50	0	0	4	0.50	vect (50.0%)
VMOVUPD %YMM1,(%RDI,%R15,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	4	1	vect (50.0%)
CMP %RCX,%R14	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
JB 4033a0 <main.extracted.65+0xd0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1	N/A

Function	main.extracted.65
Source file and lines	stream.c:355-356
Module	exec

The loop is defined in /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:355-356.

It is main loop of related source loop which is unrolled by 4 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using only 256 out of 512 bits (AVX/AVX2 instructions on AVX-512 processors).

Details

All SSE/AVX instructions are used in vector version (process two or more data elements in vector registers).

Workaround

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

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 1.00 to 0.75 cycles (1.33x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

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

FMA

Detected 4 FMA (fused multiply-add) operations.

Vector unaligned load/store instructions

Detected 2 optimal vector unaligned load/store instructions.

Details

VMOVUPD: 2 occurrences►
- /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:356
- /beegfs/hackathon/users/eoseret/qaas_runs_test/gmz16.benchmarkcenter.megware.com/177-374-0958/stream/build/stream/src/stream.c:356

Workaround

Use vector aligned instructions:

align your arrays on 64 bytes boundaries: replace { void *p = malloc (size); } with { void *p; posix_memalign (&p, 64, size); }.
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 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 8 FP arithmetical operations:

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

The binary loop is loading 64 bytes (8 double precision FP elements). The binary loop is storing 32 bytes (4 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.08 FP operations per loaded or stored byte.

General properties

nb instructions	7
nb uops	6
loop length	31
used x86 registers	7
used mmx registers	0
used xmm registers	0
used ymm registers	2
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	1.00 cycles
front end	1.00 cycles

Back-end

	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP2	FP3	FP4	FP5
uops	0.75	0.75	0.50	0.50	0.50	1.00	1.00	1.00	0.50	0.50	0.00	0.00	0.50	0.50
cycles	0.75	0.75	0.50	0.50	0.50	1.00	1.00	1.00	0.50	0.50	0.00	0.00	0.50	0.50

Execution ports to units layout:

ALU0/BRU0: ALU, BRU
ALU1: ALU
ALU2: ALU
ALU3: ALU
BRU1: BRU
AGU0 (256 bits): store address, load
AGU1 (256 bits): store address, load
AGU2 (256 bits): store address, load
FP0 (256 bits): VPU, DIV/SQRT
FP1 (256 bits): VPU, DIV/SQRT
FP2 (256 bits): VPU
FP3 (256 bits): VPU
FP4 (256 bits): FP store data
FP5 (256 bits): FP store data

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

Cycles summary

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

Vectorization ratios

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

Vector efficiency ratios

all	50%
load	50%
store	50%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	50%
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 1.00 cycles. At this rate:

66% of peak load performance is reached (64.00 out of 96.00 bytes loaded per cycle (GB/s @ 1GHz))
50% of peak store performance is reached (32.00 out of 64.00 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 1.00 to 0.75 cycles (1.33x speedup).

ASM code

In the binary file, the address of the loop is: 4033a0

Instruction	Nb FU	ALU0/BRU0	ALU1	ALU2	ALU3	BRU1	AGU0	AGU1	AGU2	FP0	FP1	FP4	FP5	Latency	Recip. throughput	Vectorization
LEA (%R11,%R14,8),%R15	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
ADD $0x4,%R14	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
VMOVUPD (%RDX,%R15,1),%YMM1	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0	0	3	0.50	vect (50.0%)
VFMADD213PD (%RSI,%R15,1),%YMM0,%YMM1	1	0	0	0	0	0	0.33	0.33	0.33	0.50	0.50	0	0	4	0.50	vect (50.0%)
VMOVUPD %YMM1,(%RDI,%R15,1)	1	0	0	0	0	0	0.33	0.33	0.33	0	0	0.50	0.50	4	1	vect (50.0%)
CMP %RCX,%R14	1	0.25	0.25	0.25	0.25	0	0	0	0	0	0	0	0	1	0.25	N/A
JB 4033a0 <main.extracted.65+0xd0>	1	0.50	0	0	0	0.50	0	0	0	0	0	0	0	1	0.50-1	N/A

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

512-bits vectorization

Workaround

FMA

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

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

512-bits vectorization

Workaround

FMA

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