OV - exec - Loop 893

	Loop Id: 893	Module: exec	Source: inner_product.hpp:155-155 [...]	Coverage: 0.38%

0x267930 VMOVUPD	(%R9),%XMM4    [1]

0x267935 VMOVUPD	0x10(%R9),%XMM5    [1]

0x26793b VMOVUPD	0x30(%R9),%XMM7    [1]

0x267941 VMOVUPD	0x40(%R9),%XMM8    [1]

0x267947 VMOVUPD	0x60(%R9),%XMM9    [1]

0x26794d VMOVUPD	0x50(%R9),%XMM1    [1]

0x267953 VMOVUPD	0x80(%R9),%XMM2    [1]

0x26795c VMOVUPD	0xb0(%R9),%XMM3    [1]

0x267965 VMOVUPD	0x20(%R9),%XMM6    [1]

0x26796b VMOVUPD	0x90(%R9),%XMM11    [1]

0x267974 VMOVUPD	-0x30(%R15,%RDX,1),%XMM13    [3]

0x26797b VMOVUPD	-0x20(%R15,%RDX,1),%XMM14    [3]

0x267982 VMOVUPD	-0x10(%R15,%RDX,1),%XMM15    [3]

0x267989 VMOVUPD	(%R15,%RDX,1),%XMM16    [3]

0x267990 ADD	$0x40,%RDX

0x267994 VBLENDPD	$0x1,%XMM4,%XMM5,%XMM0

0x26799a VSHUFPD	$0x1,%XMM6,%XMM4,%XMM4

0x26799f VBLENDPD	$0x1,%XMM5,%XMM6,%XMM5

0x2679a5 VMOVAPD	%XMM0,-0x70(%RBP)    [2]

0x2679aa VBLENDPD	$0x1,%XMM7,%XMM8,%XMM0

0x2679b0 VSHUFPD	$0x1,%XMM1,%XMM7,%XMM7

0x2679b5 VBLENDPD	$0x1,%XMM8,%XMM1,%XMM1

0x2679bb VFMADD231PD	%XMM4,%XMM13,%XMM21

0x2679c1 VFMADD231PD	%XMM5,%XMM13,%XMM26

0x2679c7 VMOVAPD	%XMM0,-0x60(%RBP)    [2]

0x2679cc VMOVUPD	0x70(%R9),%XMM0    [1]

0x2679d2 VFMADD231PD	-0x70(%RBP),%XMM13,%XMM17    [2]

0x2679d9 VFMADD231PD	%XMM7,%XMM14,%XMM22

0x2679df VFMADD231PD	%XMM1,%XMM14,%XMM27

0x2679e5 VFMADD231PD	-0x60(%RBP),%XMM14,%XMM18    [2]

0x2679ec VBLENDPD	$0x1,%XMM9,%XMM0,%XMM10

0x2679f2 VSHUFPD	$0x1,%XMM2,%XMM9,%XMM9

0x2679f7 VBLENDPD	$0x1,%XMM0,%XMM2,%XMM0

0x2679fd VMOVAPD	%XMM10,%XMM31

0x267a03 VMOVUPD	0xa0(%R9),%XMM10    [1]

0x267a0c VFMADD231PD	%XMM31,%XMM15,%XMM19

0x267a12 VFMADD231PD	%XMM9,%XMM15,%XMM23

0x267a18 VFMADD231PD	%XMM0,%XMM15,%XMM28

0x267a1e ADD	$0xc0,%R9

0x267a25 VBLENDPD	$0x1,%XMM11,%XMM10,%XMM12

0x267a2b VSHUFPD	$0x1,%XMM3,%XMM11,%XMM11

0x267a30 VBLENDPD	$0x1,%XMM10,%XMM3,%XMM2

0x267a36 VFMADD231PD	%XMM12,%XMM16,%XMM20

0x267a3c VFMADD231PD	%XMM11,%XMM16,%XMM24

0x267a42 VFMADD231PD	%XMM2,%XMM16,%XMM30

0x267a48 CMP	%RDX,%R13

0x267a4b JNE	267930

/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Platforms/CPU/SIMD/inner_product.hpp: 155 - 155

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

155:   for (int i = 0; i < n; i++)

/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h: 63 - 94

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

63:     return (a * b);

[...]

94:     (const_cast<T1&>(a) += b);

Coverage (%)	Name	Source Location	Module
►48.69+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:250	exec
○	main.omp_outlined.64	stl_vector.h:1046	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►47.64+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:249	exec
○	main.omp_outlined.64	stl_vector.h:1046	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_launch_thread		libomp.so
○	__kmp_launch_worker(void*)		libomp.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.56+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:250	exec
○	main.omp_outlined.64	stl_vector.h:1046	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_fork_call		libomp.so
○	__kmpc_fork_call		libomp.so
○	main	miniqmc.cpp:411	exec
○	__libc_start_call_main		libc.so.6
►1.47+	qmcplusplus::WaveFunction::eva[...]	WaveFunction.cpp:249	exec
○	main.omp_outlined.64	stl_vector.h:1046	exec
○	__kmp_invoke_microtask		libomp.so
○	__kmp_invoke_task_func		libomp.so
○	__kmp_fork_call		libomp.so
○	__kmpc_fork_call		libomp.so
○	main	miniqmc.cpp:411	exec
○	__libc_start_call_main		libc.so.6

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	1.82
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.74
Bottlenecks	P10, P11,
Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source	inner_product.hpp:155-155,OperatorTags.h:63-63,OperatorTags.h:94-94
Source loop unroll info	unrolled by 8
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	8
CQA cycles	10.00
CQA cycles if no scalar integer	10.00
CQA cycles if FP arith vectorized	5.50
CQA cycles if fully vectorized	2.50
Front-end cycles	5.75
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.50
P6 cycles	5.00
P7 cycles	5.00
P8 cycles	5.00
P9 cycles	5.00
P10 cycles	10.00
P11 cycles	10.00
P12 cycles	2.00
P13 cycles	2.00
P14 cycles	1.00
P15 cycles	1.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	4
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	47.00
Nb uops	46.00
Nb loads	18.00
Nb stores	2.00
Nb stack references	2.00
FLOP/cycle	4.80
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	24.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	32.00
Bytes prefetched	0.00
Bytes loaded	288.00
Bytes stored	32.00
Stride 0	1.00
Stride 1	2.00
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	100.00
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	NA
Vector-efficiency ratio fma	25.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.82
CQA speedup if fully vectorized	4.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.74
Bottlenecks	P10, P11,
Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source	inner_product.hpp:155-155,OperatorTags.h:63-63,OperatorTags.h:94-94
Source loop unroll info	unrolled by 8
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	8
CQA cycles	10.00
CQA cycles if no scalar integer	10.00
CQA cycles if FP arith vectorized	5.50
CQA cycles if fully vectorized	2.50
Front-end cycles	5.75
P0 cycles	0.50
P1 cycles	0.50
P2 cycles	0.50
P3 cycles	0.50
P4 cycles	0.50
P5 cycles	0.50
P6 cycles	5.00
P7 cycles	5.00
P8 cycles	5.00
P9 cycles	5.00
P10 cycles	10.00
P11 cycles	10.00
P12 cycles	2.00
P13 cycles	2.00
P14 cycles	1.00
P15 cycles	1.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	4
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	47.00
Nb uops	46.00
Nb loads	18.00
Nb stores	2.00
Nb stack references	2.00
FLOP/cycle	4.80
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	24.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	32.00
Bytes prefetched	0.00
Bytes loaded	288.00
Bytes stored	32.00
Stride 0	1.00
Stride 1	2.00
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	100.00
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	NA
Vector-efficiency ratio fma	25.00
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	25.00

Path /

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

Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source file and lines	inner_product.hpp:155-155
Module	exec

The loop is defined in:

/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Platforms/CPU/SIMD/inner_product.hpp:155
/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63,94

It is main loop of related source loop which is unrolled by 8 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using only 128 out of 512 bits (SSE/AVX-128 instructions on AVX-512 processors).

By fully vectorizing your loop, you can lower the cost of an iteration from 10.00 to 2.50 cycles (4.00x speedup).

Details

All SSE/AVX instructions are used in vector version (process two or more data elements in vector registers). Since your execution units are vector units, only a fully vectorized loop can use their full power.

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.

Execution units bottlenecks

Performance is limited by execution of FP multiply or FMA (fused multiply-add) operations (the FP multiply/FMA unit is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 10.00 to 5.75 cycles (1.74x speedup).

Workaround

Reduce the number of FP multiply/FMA instructions

FMA

Detected 24 FMA (fused multiply-add) operations.

Type of elements and instruction set

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

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

The binary loop is composed of 48 FP arithmetical operations:

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

The binary loop is loading 288 bytes (36 double precision FP elements). The binary loop is storing 32 bytes (4 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.15 FP operations per loaded or stored byte.

General properties

nb instructions	47
nb uops	46
loop length	289
used x86 registers	5
used mmx registers	0
used xmm registers	30
used ymm registers	0
used zmm registers	0
nb stack references	2

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	5.75 cycles
front end	5.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.50	0.50	0.50	0.50	0.50	0.50	5.00	5.00	5.00	5.00	10.00	10.00	2.00	2.00	1.00	1.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	5.00	5.00	5.00	5.00	10.00	10.00	2.00	2.00	1.00	1.00

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)	4.00

Cycles summary

Front-end	5.75
Dispatch	10.00
Data deps.	4.00
Overall L1	10.00

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	100%

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 10.00 cycles. At this rate:

11% of peak load performance is reached (28.80 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (3.20 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 267930

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VMOVUPD (%R9),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x10(%R9),%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x30(%R9),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x40(%R9),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x60(%R9),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x50(%R9),%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x80(%R9),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0xb0(%R9),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x20(%R9),%XMM6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x90(%R9),%XMM11	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x30(%R15,%RDX,1),%XMM13	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x20(%R15,%RDX,1),%XMM14	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x10(%R15,%RDX,1),%XMM15	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD (%R15,%RDX,1),%XMM16	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
ADD $0x40,%RDX	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
VBLENDPD $0x1,%XMM4,%XMM5,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM6,%XMM4,%XMM4	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%)
VBLENDPD $0x1,%XMM5,%XMM6,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VMOVAPD %XMM0,-0x70(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	4	0.50	vect (25.0%)
VBLENDPD $0x1,%XMM7,%XMM8,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM1,%XMM7,%XMM7	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%)
VBLENDPD $0x1,%XMM8,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VFMADD231PD %XMM4,%XMM13,%XMM21	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM5,%XMM13,%XMM26	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VMOVAPD %XMM0,-0x60(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	4	0.50	vect (25.0%)
VMOVUPD 0x70(%R9),%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VFMADD231PD -0x70(%RBP),%XMM13,%XMM17	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM7,%XMM14,%XMM22	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM1,%XMM14,%XMM27	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD -0x60(%RBP),%XMM14,%XMM18	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VBLENDPD $0x1,%XMM9,%XMM0,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM2,%XMM9,%XMM9	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%)
VBLENDPD $0x1,%XMM0,%XMM2,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VMOVAPD %XMM10,%XMM31	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VMOVUPD 0xa0(%R9),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VFMADD231PD %XMM31,%XMM15,%XMM19	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM9,%XMM15,%XMM23	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM0,%XMM15,%XMM28	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
ADD $0xc0,%R9	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
VBLENDPD $0x1,%XMM11,%XMM10,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM3,%XMM11,%XMM11	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%)
VBLENDPD $0x1,%XMM10,%XMM3,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VFMADD231PD %XMM12,%XMM16,%XMM20	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM11,%XMM16,%XMM24	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM2,%XMM16,%XMM30	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
CMP %RDX,%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	0.17	N/A
JNE 267930 <_ZN16miniqmcreference19DiracDeterminantRefIN11qmcplusplus13DelayedUpdateIddEEE10evaluateGLERNS1_11ParticleSetERNS1_14ParticleAttribINS1_10TinyVectorIdLj3EEESaIS9_EEERNS7_IdSaIdEEEb+0x310>	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

Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source file and lines	inner_product.hpp:155-155
Module	exec

The loop is defined in:

/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Platforms/CPU/SIMD/inner_product.hpp:155
/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63,94

It is main loop of related source loop which is unrolled by 8 (including vectorization).

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using only 128 out of 512 bits (SSE/AVX-128 instructions on AVX-512 processors).

By fully vectorizing your loop, you can lower the cost of an iteration from 10.00 to 2.50 cycles (4.00x speedup).

Details

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.

Execution units bottlenecks

Workaround

Reduce the number of FP multiply/FMA instructions

FMA

Detected 24 FMA (fused multiply-add) operations.

Type of elements and instruction set

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

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

The binary loop is composed of 48 FP arithmetical operations:

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

The binary loop is loading 288 bytes (36 double precision FP elements). The binary loop is storing 32 bytes (4 double precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.15 FP operations per loaded or stored byte.

General properties

nb instructions	47
nb uops	46
loop length	289
used x86 registers	5
used mmx registers	0
used xmm registers	30
used ymm registers	0
used zmm registers	0
nb stack references	2

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	5.75 cycles
front end	5.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	0.50	0.50	0.50	0.50	0.50	0.50	5.00	5.00	5.00	5.00	10.00	10.00	2.00	2.00	1.00	1.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	5.00	5.00	5.00	5.00	10.00	10.00	2.00	2.00	1.00	1.00

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)	4.00

Cycles summary

Front-end	5.75
Dispatch	10.00
Data deps.	4.00
Overall L1	10.00

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	100%

Vector efficiency ratios

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

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 10.00 cycles. At this rate:

11% of peak load performance is reached (28.80 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (3.20 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 267930

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
VMOVUPD (%R9),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x10(%R9),%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x30(%R9),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x40(%R9),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x60(%R9),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x50(%R9),%XMM1	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x80(%R9),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0xb0(%R9),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x20(%R9),%XMM6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD 0x90(%R9),%XMM11	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x30(%R15,%RDX,1),%XMM13	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x20(%R15,%RDX,1),%XMM14	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD -0x10(%R15,%RDX,1),%XMM15	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VMOVUPD (%R15,%RDX,1),%XMM16	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
ADD $0x40,%RDX	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
VBLENDPD $0x1,%XMM4,%XMM5,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM6,%XMM4,%XMM4	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%)
VBLENDPD $0x1,%XMM5,%XMM6,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VMOVAPD %XMM0,-0x70(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	4	0.50	vect (25.0%)
VBLENDPD $0x1,%XMM7,%XMM8,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM1,%XMM7,%XMM7	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%)
VBLENDPD $0x1,%XMM8,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VFMADD231PD %XMM4,%XMM13,%XMM21	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM5,%XMM13,%XMM26	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VMOVAPD %XMM0,-0x60(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	4	0.50	vect (25.0%)
VMOVUPD 0x70(%R9),%XMM0	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VFMADD231PD -0x70(%RBP),%XMM13,%XMM17	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM7,%XMM14,%XMM22	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM1,%XMM14,%XMM27	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD -0x60(%RBP),%XMM14,%XMM18	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VBLENDPD $0x1,%XMM9,%XMM0,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM2,%XMM9,%XMM9	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%)
VBLENDPD $0x1,%XMM0,%XMM2,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VMOVAPD %XMM10,%XMM31	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
VMOVUPD 0xa0(%R9),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.50	vect (25.0%)
VFMADD231PD %XMM31,%XMM15,%XMM19	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM9,%XMM15,%XMM23	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM0,%XMM15,%XMM28	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
ADD $0xc0,%R9	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
VBLENDPD $0x1,%XMM11,%XMM10,%XMM12	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VSHUFPD $0x1,%XMM3,%XMM11,%XMM11	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%)
VBLENDPD $0x1,%XMM10,%XMM3,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	1	0.50	vect (25.0%)
VFMADD231PD %XMM12,%XMM16,%XMM20	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM11,%XMM16,%XMM24	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
VFMADD231PD %XMM2,%XMM16,%XMM30	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	4	0.50	vect (25.0%)
CMP %RDX,%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	0.17	N/A
JNE 267930 <_ZN16miniqmcreference19DiracDeterminantRefIN11qmcplusplus13DelayedUpdateIddEEE10evaluateGLERNS1_11ParticleSetERNS1_14ParticleAttribINS1_10TinyVectorIdLj3EEESaIS9_EEERNS7_IdSaIdEEEb+0x310>	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

Execution units bottlenecks

Workaround

FMA

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

FMA

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