OV - exec - Loop 1054

	Loop Id: 1054	Module: exec	Source: inner_product.hpp:155-155 [...]	Coverage: 0.37%

0x45ae90 MOVUPD	-0x58(%R10),%XMM5    [1]

0x45ae96 MOVUPD	-0x40(%R10),%XMM6    [1]

0x45ae9c MOVSD	-0x48(%R10),%XMM3    [1]

0x45aea2 MOVHPD	-0x30(%R10),%XMM3    [1]

0x45aea8 MOVUPD	-0x28(%R10),%XMM7    [1]

0x45aeae MOVSD	-0x18(%R10),%XMM8    [1]

0x45aeb4 MOVHPD	(%R10),%XMM8    [1]

0x45aeb9 MOVUPD	-0x10(%R10),%XMM9    [1]

0x45aebf MOVUPD	(%RSI,%R8,1),%XMM10    [2]

0x45aec5 MOVUPD	0x10(%RSI,%R8,1),%XMM2    [2]

0x45aecc MULPD	%XMM2,%XMM8

0x45aed1 MULPD	%XMM10,%XMM3

0x45aed6 ADDPD	%XMM8,%XMM3

0x45aedb MOVAPD	%XMM10,%XMM8

0x45aee0 UNPCKLPD	%XMM10,%XMM8

0x45aee5 MULPD	%XMM5,%XMM8

0x45aeea ADDPD	%XMM4,%XMM8

0x45aeef UNPCKHPD	%XMM10,%XMM10

0x45aef4 MULPD	%XMM6,%XMM10

0x45aef9 MOVAPD	%XMM2,%XMM4

0x45aefd UNPCKLPD	%XMM2,%XMM4

0x45af01 MULPD	%XMM7,%XMM4

0x45af05 ADDPD	%XMM10,%XMM4

0x45af0a ADDPD	%XMM8,%XMM4

0x45af0f UNPCKHPD	%XMM2,%XMM2

0x45af13 MULPD	%XMM9,%XMM2

0x45af18 ADDPD	%XMM4,%XMM2

0x45af1c MOVAPD	%XMM3,%XMM4

0x45af20 UNPCKHPD	%XMM3,%XMM4

0x45af24 ADDSD	%XMM3,%XMM4

0x45af28 ADDSD	%XMM4,%XMM0

0x45af2c ADD	$0x20,%R8

0x45af30 ADD	$0x60,%R10

0x45af34 MOVAPD	%XMM2,%XMM4

0x45af38 CMP	%R8,%R15

0x45af3b JNE	45ae90

/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 - 63

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

63:     return (a * b);

/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/DiracDeterminantRef.cpp: 157 - 157

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

157:   if (UpdateMode == ORB_PBYP_RATIO)

Coverage (%)	Name	Source Location	Module
►47.52+	_ZN11qmcplusplus12WaveFunction[...]	WaveFunction.cpp:249	exec
○	main.extracted.107	miniqmc.cpp:467	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►47.51+	_ZN11qmcplusplus12WaveFunction[...]	WaveFunction.cpp:249	exec
○	main.extracted.107	miniqmc.cpp:467	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►1.37+	_ZN11qmcplusplus12WaveFunction[...]	WaveFunction.cpp:249	exec
○	main.extracted.107	miniqmc.cpp:467	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_fork_call		libiomp5.so
○	__kmpc_fork_call		libiomp5.so
○	main	miniqmc.cpp:411	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	stl_uninitialized.h:99	exec
►1.37+	_ZN11qmcplusplus12WaveFunction[...]	WaveFunction.cpp:249	exec
○	main.extracted.107	miniqmc.cpp:467	exec
○	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_fork_call		libiomp5.so
○	__kmpc_fork_call		libiomp5.so
○	main	miniqmc.cpp:411	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start	stl_uninitialized.h:99	exec

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	1.89
CQA speedup if fully vectorized	5.43
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.20
Bottlenecks
Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source	inner_product.hpp:155-155,OperatorTags.h:63-63,DiracDeterminantRef.cpp:157-157
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	6.00
CQA cycles if no scalar integer	6.00
CQA cycles if FP arith vectorized	3.17
CQA cycles if fully vectorized	1.10
Front-end cycles	4.38
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	2.50
P7 cycles	2.50
P8 cycles	2.50
P9 cycles	2.50
P10 cycles	5.00
P11 cycles	5.00
P12 cycles	5.00
P13 cycles	5.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	6
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	36.00
Nb uops	35.00
Nb loads	10.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	4.00
Nb FLOP add-sub	12.00
Nb FLOP mul	12.00
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	21.33
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	65.63
Vectorization ratio load	60.00
Vectorization ratio store	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	71.43
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	44.44
Vector-efficiency ratio all	20.70
Vector-efficiency ratio load	20.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	21.43
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	18.06

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.89
CQA speedup if fully vectorized	5.43
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.20
Bottlenecks
Function	miniqmcreference::DiracDeterminantRef >::evaluateGL(qmcplusplus::ParticleSet&, qmcplusplus::ParticleAttrib, std::allocator > >&, qmcplusplus::ParticleAttrib >&, bool)
Source	inner_product.hpp:155-155,OperatorTags.h:63-63,DiracDeterminantRef.cpp:157-157
Source loop unroll info	unrolled by 4
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	4
CQA cycles	6.00
CQA cycles if no scalar integer	6.00
CQA cycles if FP arith vectorized	3.17
CQA cycles if fully vectorized	1.10
Front-end cycles	4.38
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	2.50
P7 cycles	2.50
P8 cycles	2.50
P9 cycles	2.50
P10 cycles	5.00
P11 cycles	5.00
P12 cycles	5.00
P13 cycles	5.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	6
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	36.00
Nb uops	35.00
Nb loads	10.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	4.00
Nb FLOP add-sub	12.00
Nb FLOP mul	12.00
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	21.33
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	2.00
Stride n	0.00
Stride unknown	0.00
Stride indirect	0.00
Vectorization ratio all	65.63
Vectorization ratio load	60.00
Vectorization ratio store	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	71.43
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	44.44
Vector-efficiency ratio all	20.70
Vector-efficiency ratio load	20.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	25.00
Vector-efficiency ratio add_sub	21.43
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	18.06

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
/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/DiracDeterminantRef.cpp:157

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

gain
potential
hint
expert

Vectorization

Your loop is partially vectorized. Only 20% of vector register length is used (average across all SSE/AVX instructions). By fully vectorizing your loop, you can lower the cost of an iteration from 6.00 to 1.10 cycles (5.43x speedup).

Details

65% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

60% of SSE/AVX loads are used in vector version.
71% of SSE/AVX addition or subtraction instructions are used in vector version.
44% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Since your execution units are vector units, only a fully vectorized loop can use their full power.

Workaround

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

Execution units bottlenecks

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

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Try to change order in which elements are evaluated (using parentheses) in arithmetic expressions containing both ADD/SUB and MUL operations to enable your compiler to generate FMA instructions wherever possible. For instance a + b*c is a valid FMA (MUL then ADD). However (a+b)* c cannot be translated into an FMA (ADD then MUL).

Vector unaligned load/store instructions

Detected 8 suboptimal vector unaligned load/store instructions.

Details

MOVHPD: 2 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
MOVUPD: 6 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63

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

2 SSE or AVX instructions are processing arithmetic or math operations on double precision FP elements in scalar mode (one at a time). 11 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 24 FP arithmetical operations:

12: addition or subtraction
12: multiply

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

Arithmetic intensity

Arithmetic intensity is 0.19 FP operations per loaded or stored byte.

General properties

nb instructions	36
nb uops	35
loop length	177
used x86 registers	4
used mmx registers	0
used xmm registers	10
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.17

Front-end

ASSUMED MACRO FUSION 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	0.50	0.50	0.50	0.50	0.50	0.50	2.50	2.50	2.50	2.50	5.00	5.00	5.00	5.00	0.00	0.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	2.50	2.50	2.50	2.50	5.00	5.00	5.00	5.00	0.00	0.00

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)	6.00

Cycles summary

Front-end	4.38
Dispatch	5.00
Data deps.	6.00
Overall L1	6.00

Vectorization ratios

all	65%
load	60%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	71%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	44%

Vector efficiency ratios

all	20%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	21%
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 6.00 cycles. At this rate:

8% of peak load performance is reached (21.33 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 45ae90

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	Latency	Recip. throughput	Vectorization
MOVUPD -0x58(%R10),%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD -0x40(%R10),%XMM6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVSD -0x48(%R10),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
MOVHPD -0x30(%R10),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	1	0.50	scal (12.5%)
MOVUPD -0x28(%R10),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVSD -0x18(%R10),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
MOVHPD (%R10),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	1	0.50	scal (12.5%)
MOVUPD -0x10(%R10),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD (%RSI,%R8,1),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x10(%RSI,%R8,1),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM2,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
MULPD %XMM10,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM8,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
MOVAPD %XMM10,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKLPD %XMM10,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM5,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM4,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
UNPCKHPD %XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM6,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
MOVAPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKLPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM7,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM10,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
ADDPD %XMM8,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
UNPCKHPD %XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM9,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM4,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
MOVAPD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKHPD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
ADDSD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
ADDSD %XMM4,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
ADD $0x20,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x60,%R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVAPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
CMP %R8,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
JNE 45ae90 <_ZN16miniqmcreference19DiracDeterminantRefIN11qmcplusplus13DelayedUpdateIddEEE10evaluateGLERNS1_11ParticleSetERNS1_14ParticleAttribINS1_10TinyVectorIdLj3EEESaIS9_EEERNS7_IdSaIdEEEb+0x300>	1	0	0	0	0.33	0.33	0.33	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
/home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/QMCWaveFunctions/DiracDeterminantRef.cpp:157

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

gain
potential
hint
expert

Vectorization

Details

65% of SSE/AVX instructions are used in vector version (process two or more data elements in vector registers):

60% of SSE/AVX loads are used in vector version.
71% of SSE/AVX addition or subtraction instructions are used in vector version.
44% of SSE/AVX instructions that are not load, store, addition, subtraction nor multiply instructions are used in vector version.

Since your execution units are vector units, only a fully vectorized loop can use their full power.

Workaround

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

Execution units bottlenecks

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

FMA

Presence of both ADD/SUB and MUL operations.

Workaround

Pass to your compiler a micro-architecture specialization option:
- use march=native
Try to change order in which elements are evaluated (using parentheses) in arithmetic expressions containing both ADD/SUB and MUL operations to enable your compiler to generate FMA instructions wherever possible. For instance a + b*c is a valid FMA (MUL then ADD). However (a+b)* c cannot be translated into an FMA (ADD then MUL).

Vector unaligned load/store instructions

Detected 8 suboptimal vector unaligned load/store instructions.

Details

MOVHPD: 2 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
MOVUPD: 6 occurrences►
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63
- /home/eoseret/qaas_runs_ZEN5/173-938-0042/intel/miniqmc/build/miniqmc/src/Numerics/PETE/OperatorTags.h:63

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 24 FP arithmetical operations:

12: addition or subtraction
12: multiply

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

Arithmetic intensity

Arithmetic intensity is 0.19 FP operations per loaded or stored byte.

General properties

nb instructions	36
nb uops	35
loop length	177
used x86 registers	4
used mmx registers	0
used xmm registers	10
used ymm registers	0
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.17

Front-end

ASSUMED MACRO FUSION 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	0.50	0.50	0.50	0.50	0.50	0.50	2.50	2.50	2.50	2.50	5.00	5.00	5.00	5.00	0.00	0.00
cycles	0.50	0.50	0.50	0.50	0.50	0.50	2.50	2.50	2.50	2.50	5.00	5.00	5.00	5.00	0.00	0.00

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)	6.00

Cycles summary

Front-end	4.38
Dispatch	5.00
Data deps.	6.00
Overall L1	6.00

Vectorization ratios

all	65%
load	60%
store	NA (no store vectorizable/vectorized instructions)
mul	100%
add-sub	71%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	44%

Vector efficiency ratios

all	20%
load	20%
store	NA (no store vectorizable/vectorized instructions)
mul	25%
add-sub	21%
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 6.00 cycles. At this rate:

8% of peak load performance is reached (21.33 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 45ae90

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	Latency	Recip. throughput	Vectorization
MOVUPD -0x58(%R10),%XMM5	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD -0x40(%R10),%XMM6	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVSD -0x48(%R10),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
MOVHPD -0x30(%R10),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	1	0.50	scal (12.5%)
MOVUPD -0x28(%R10),%XMM7	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVSD -0x18(%R10),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	1	0.50	scal (12.5%)
MOVHPD (%R10),%XMM8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0.25	0.25	0.25	0.25	1	0.50	scal (12.5%)
MOVUPD -0x10(%R10),%XMM9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD (%RSI,%R8,1),%XMM10	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MOVUPD 0x10(%RSI,%R8,1),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	3	0.50	vect (25.0%)
MULPD %XMM2,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
MULPD %XMM10,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM8,%XMM3	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
MOVAPD %XMM10,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKLPD %XMM10,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM5,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM4,%XMM8	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
UNPCKHPD %XMM10,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM6,%XMM10	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
MOVAPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKLPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM7,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM10,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
ADDPD %XMM8,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
UNPCKHPD %XMM2,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
MULPD %XMM9,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3	0.50	vect (25.0%)
ADDPD %XMM4,%XMM2	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	vect (25.0%)
MOVAPD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
UNPCKHPD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
ADDSD %XMM3,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
ADDSD %XMM4,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	2	0.50	scal (12.5%)
ADD $0x20,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x60,%R10	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
MOVAPD %XMM2,%XMM4	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
CMP %R8,%R15	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	1	0.17	N/A
JNE 45ae90 <_ZN16miniqmcreference19DiracDeterminantRefIN11qmcplusplus13DelayedUpdateIddEEE10evaluateGLERNS1_11ParticleSetERNS1_14ParticleAttribINS1_10TinyVectorIdLj3EEESaIS9_EEERNS7_IdSaIdEEEb+0x300>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	1	0.50-1	N/A

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

FMA

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

Vectorization

Details

Workaround

Execution units bottlenecks

FMA

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