OV - exec - Loop 3

	Loop Id: 3	Module: exec	Source: Step10_orig.c:19-31	Coverage: 67.21%

0x401a80 VMOVUPS	(%RSI,%R11,1),%YMM8    [3]

0x401a86 VMOVUPS	(%RDX,%R11,1),%YMM1    [4]

0x401a8c VSUBPS	%YMM25,%YMM8,%YMM8

0x401a92 VSUBPS	%YMM24,%YMM1,%YMM7

0x401a98 VMOVUPS	(%RDI,%R11,1),%YMM9    [2]

0x401a9e VMULPS	%YMM8,%YMM8,%YMM1

0x401aa3 VSUBPS	%YMM26,%YMM9,%YMM9

0x401aa9 VFMADD231PS	%YMM9,%YMM9,%YMM1

0x401aae VFMADD231PS	%YMM7,%YMM7,%YMM1

0x401ab3 VCMPPS	$0x1,%YMM15,%YMM1,%K1

0x401aba VADDPS	%YMM23,%YMM1,%YMM5

0x401ac0 VCMPPS	$0x1,%YMM15,%YMM1,%YMM0

0x401ac6 VMOVUPS	(%R8,%R11,1),%YMM13{%K1}    [1]

0x401acd VCVTPS2PD	%XMM5,%YMM6

0x401ad1 VANDPS	%YMM13,%YMM0,%YMM28

0x401ad7 VEXTRACTF32X4	$0x1,%YMM5,%XMM0

0x401ade VCVTPS2PD	%XMM0,%YMM5

0x401ae2 VMOVAPS	%YMM1,%YMM0

0x401ae6 VFMADD132PS	%YMM22,%YMM21,%YMM0

0x401aec VSQRTPD	%YMM6,%YMM30

0x401af2 VCMPPS	$0xe,%YMM16,%YMM1,%K2

0x401af9 VSQRTPD	%YMM5,%YMM29

0x401aff ADD	$0x20,%R11

0x401b03 VFMADD132PS	%YMM1,%YMM20,%YMM0

0x401b09 VMULPD	%YMM29,%YMM5,%YMM5

0x401b0f VFMADD132PS	%YMM1,%YMM19,%YMM0

0x401b15 VFMADD132PS	%YMM1,%YMM18,%YMM0

0x401b1b VFMADD132PS	%YMM1,%YMM17,%YMM0

0x401b21 VMULPD	%YMM30,%YMM6,%YMM1

0x401b27 VCVTPS2PD	%XMM0,%YMM30

0x401b2d VDIVPD	%YMM1,%YMM14,%YMM6

0x401b31 VEXTRACTF32X4	$0x1,%YMM0,%XMM0

0x401b38 VADDPD	%YMM30,%YMM6,%YMM1

0x401b3e VDIVPD	%YMM5,%YMM14,%YMM6

0x401b42 VCVTPS2PD	%XMM0,%YMM5

0x401b46 VCVTPD2PS	%YMM1,%XMM1

0x401b4a VADDPD	%YMM5,%YMM6,%YMM0

0x401b4e VCVTPD2PS	%YMM0,%XMM6

0x401b52 VINSERTF32X4	$0x1,%XMM6,%YMM1,%YMM5

0x401b59 VMULPS	%YMM28,%YMM5,%YMM0

0x401b5f VFMADD231PS	%YMM0,%YMM9,%YMM10{%K2}

0x401b65 VFMADD231PS	%YMM0,%YMM8,%YMM11{%K2}

0x401b6b VFMADD231PS	%YMM0,%YMM7,%YMM12{%K2}

0x401b71 CMP	%R11,%RBX

0x401b74 JNE	401a80

/home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c: 19 - 31

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

19:     for ( j = 0; j < count1; j++ )

20:     {

21:         dxc = xx1[j] - xxi;

22:         dyc = yy1[j] - yyi;

23:         dzc = zz1[j] - zzi;

24:

25:         r2 = dxc * dxc + dyc * dyc + dzc * dzc;

26:

27:         m = ( r2 < fsrrmax2 ) ? mass1[j] : 0.0f;

28:

29:         f =  pow( r2 + mp_rsm2, -1.5 ) - ( ma0 + r2*(ma1 + r2*(ma2 + r2*(ma3 + r2*(ma4 + r2*ma5)))));

30:

31:         f = ( r2 > 0.0f ) ? m * f : 0.0f;

Coverage (%)	Name	Source Location	Module
►38.47+	main._omp_fn.1	main.c:144	exec
○	gomp_thread_start	team.c:130	libgomp.so.1.0.0
►38.07+	main._omp_fn.1	main.c:144	exec
○	gomp_thread_start	team.c:130	libgomp.so.1.0.0
►17.51+	main._omp_fn.1	main.c:144	exec
○	gomp_thread_start	team.c:130	libgomp.so.1.0.0
►5.53+	main._omp_fn.1	main.c:144	exec
○	gomp_thread_start	team.c:130	libgomp.so.1.0.0

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.27
Bottlenecks	P0,
Function	Step10_orig
Source	Step10_orig.c:19-31
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	34.00
CQA cycles if no scalar integer	34.00
CQA cycles if FP arith vectorized	34.00
CQA cycles if fully vectorized	34.00
Front-end cycles	8.33
P0 cycles	15.00
P1 cycles	15.00
P2 cycles	1.33
P3 cycles	1.33
P4 cycles	0.00
P5 cycles	13.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	1.33
DIV/SQRT cycles	34.00
Inter-iter dependencies cycles	4
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	45.00
Nb uops	50.00
Nb loads	4.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	7.06
Nb FLOP add-sub	40.00
Nb FLOP mul	24.00
Nb FLOP fma	80.00
Nb FLOP div	8.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	8.00
Nb FLOP rsqrt	0.00
Bytes/cycle	3.76
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	4.00
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	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	100.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	100.00
Vector-efficiency ratio all	45.83
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	50.00
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	50.00
Vector-efficiency ratio div_sqrt	50.00
Vector-efficiency ratio other	37.50

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	1.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.27
Bottlenecks	P0,
Function	Step10_orig
Source	Step10_orig.c:19-31
Source loop unroll info	not unrolled or unrolled with no peel/tail loop
Source loop unroll confidence level	max
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	34.00
CQA cycles if no scalar integer	34.00
CQA cycles if FP arith vectorized	34.00
CQA cycles if fully vectorized	34.00
Front-end cycles	8.33
P0 cycles	15.00
P1 cycles	15.00
P2 cycles	1.33
P3 cycles	1.33
P4 cycles	0.00
P5 cycles	13.00
P6 cycles	1.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	0.00
P11 cycles	1.33
DIV/SQRT cycles	34.00
Inter-iter dependencies cycles	4
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	45.00
Nb uops	50.00
Nb loads	4.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	7.06
Nb FLOP add-sub	40.00
Nb FLOP mul	24.00
Nb FLOP fma	80.00
Nb FLOP div	8.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	8.00
Nb FLOP rsqrt	0.00
Bytes/cycle	3.76
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	4.00
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	NA
Vectorization ratio mul	100.00
Vectorization ratio add_sub	100.00
Vectorization ratio fma	100.00
Vectorization ratio div_sqrt	100.00
Vectorization ratio other	100.00
Vector-efficiency ratio all	45.83
Vector-efficiency ratio load	50.00
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	50.00
Vector-efficiency ratio add_sub	50.00
Vector-efficiency ratio fma	50.00
Vector-efficiency ratio div_sqrt	50.00
Vector-efficiency ratio other	37.50

Path /

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

Function	Step10_orig
Source file and lines	Step10_orig.c:19-31
Module	exec

The loop is defined in /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:19-31.

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

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using 45% register length (average across all SSE/AVX instructions).

Details

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

Execution units bottlenecks

Performance is limited by execution of divide and square root operations (the divide/square root unit is a bottleneck). By removing all these bottlenecks, you can lower the cost of an iteration from 34.00 to 15.00 cycles (2.27x speedup).

Workaround

Reduce the number of division or square root instructions:
- If denominator is constant over iterations, use reciprocal (replace x/y with x*(1/y)). Check precision impact. This will be done by your compiler with ffast-math or Ofast
If you accept to loose numerical precision up to 2 ulp, you can speedup your code by passing the following options to your compiler: (ffast-math or Ofast) and mrecip

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

Expensive FP math instructions/calls

Detected performance impact from expensive FP math instructions/calls. By removing/reexpressing them, you can lower the cost of an iteration from 34.00 to 15.00 cycles (2.27x speedup).

Masked instructions

Detected masked instructions.

Details

Vector registers are partially exploited, which is expected if your loop is irregular or mixes elements of different sizes.

Workaround

If your loop is irregular, try to remove or hoist conditional structures out of your loop. If it mixes elements of different sizes, try to uniformize them.

FMA

Detected 80 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

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).

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

VCVTPD2PS: 2 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
VCVTPS2PD: 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29

Vector unaligned load/store instructions

Detected 4 optimal vector unaligned load/store instructions.

Details

VMOVUPS: 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:22
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:23
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:21
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:27

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.

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTPD2PS (FP64 to FP32, SIMD): 2 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
VCVTPS2PD (FP32 to FP64, SIMD): 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements. Use double instead of single precision only when/where needed by numerical stability and avoid mixing precision. In C/C++, FP constants are double precision by default and must be suffixed by 'f' to make them single precision.

Type of elements and instruction set

19 AVX instructions are processing arithmetic or math operations on single precision FP elements in vector mode (eight at a time). 8 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 240 FP arithmetical operations:

120: addition or subtraction (80 inside FMA instructions)
104: multiply (80 inside FMA instructions)
8: divide
8: square root

The binary loop is loading 128 bytes (32 single precision FP elements).

Arithmetic intensity

Arithmetic intensity is 1.87 FP operations per loaded or stored byte.

General properties

nb instructions	45
nb uops	50
loop length	250
used x86 registers	6
used mmx registers	0
used xmm registers	4
used ymm registers	27
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.50

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	8.33 cycles
front end	8.33 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	15.00	15.00	1.33	1.33	0.00	13.00	1.00	0.00	0.00	0.00	0.00	1.33
cycles	15.00	15.00	1.33	1.33	0.00	13.00	1.00	0.00	0.00	0.00	0.00	1.33

Execution ports to units layout:

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

Cycles executing div or sqrt instructions	34.00
Longest recurrence chain latency (RecMII)	4.00

Cycles summary

Front-end	8.33
Dispatch	15.00
DIV/SQRT	34.00
Data deps.	4.00
Overall L1	34.00

Vectorization ratios

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

Vector efficiency ratios

all	45%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	50%
add-sub	50%
fma	50%
div/sqrt	50%
other	37%

Cycles and memory resources usage

Detected masked instructions: assuming all mask elements are active. Assuming all data fit into the L1 cache, each iteration of the binary loop takes 34.00 cycles. At this rate:

1% of peak load performance is reached (3.76 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 401a80

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
VMOVUPS (%RSI,%R11,1),%YMM8	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VMOVUPS (%RDX,%R11,1),%YMM1	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VSUBPS %YMM25,%YMM8,%YMM8	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VSUBPS %YMM24,%YMM1,%YMM7	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VMOVUPS (%RDI,%R11,1),%YMM9	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VMULPS %YMM8,%YMM8,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VSUBPS %YMM26,%YMM9,%YMM9	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VFMADD231PS %YMM9,%YMM9,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM7,%YMM7,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VCMPPS $0x1,%YMM15,%YMM1,%K1	1	0	0	0	0	1	0	0	0	3	1	vect (50.0%)
VADDPS %YMM23,%YMM1,%YMM5	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VCMPPS $0x1,%YMM15,%YMM1,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMOVUPS (%R8,%R11,1),%YMM13{%K1}	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VCVTPS2PD %XMM5,%YMM6	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VANDPS %YMM13,%YMM0,%YMM28	1	0.33	0.33	0	0	0.33	0	0	0	1	0.33	vect (50.0%)
VEXTRACTF32X4 $0x1,%YMM5,%XMM0	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VCVTPS2PD %XMM0,%YMM5	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VMOVAPS %YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0-1	0.17	vect (50.0%)
VFMADD132PS %YMM22,%YMM21,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VSQRTPD %YMM6,%YMM30	1	1	0	0	0	0	0	0	0	13-19	9	vect (50.0%)
VCMPPS $0xe,%YMM16,%YMM1,%K2	1	0	0	0	0	1	0	0	0	3	1	vect (50.0%)
VSQRTPD %YMM5,%YMM29	1	1	0	0	0	0	0	0	0	13-19	9	vect (50.0%)
ADD $0x20,%R11	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VFMADD132PS %YMM1,%YMM20,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMULPD %YMM29,%YMM5,%YMM5	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM19,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM18,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM17,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMULPD %YMM30,%YMM6,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VCVTPS2PD %XMM0,%YMM30	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VDIVPD %YMM1,%YMM14,%YMM6	1	1	0	0	0	0	0	0	0	13-15	8	vect (50.0%)
VEXTRACTF32X4 $0x1,%YMM0,%XMM0	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VADDPD %YMM30,%YMM6,%YMM1	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VDIVPD %YMM5,%YMM14,%YMM6	1	1	0	0	0	0	0	0	0	13-15	8	vect (50.0%)
VCVTPS2PD %XMM0,%YMM5	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VCVTPD2PS %YMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (50.0%)
VADDPD %YMM5,%YMM6,%YMM0	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VCVTPD2PS %YMM0,%XMM6	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (50.0%)
VINSERTF32X4 $0x1,%XMM6,%YMM1,%YMM5	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VMULPS %YMM28,%YMM5,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM9,%YMM10{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM8,%YMM11{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM7,%YMM12{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
CMP %R11,%RBX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 401a80 <Step10_orig+0xc0>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Function	Step10_orig
Source file and lines	Step10_orig.c:19-31
Module	exec

The loop is defined in /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:19-31.

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

gain
potential
hint
expert

Vectorization

Your loop is vectorized, but using 45% register length (average across all SSE/AVX instructions).

Details

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

Execution units bottlenecks

Workaround

Reduce the number of division or square root instructions:
- If denominator is constant over iterations, use reciprocal (replace x/y with x*(1/y)). Check precision impact. This will be done by your compiler with ffast-math or Ofast
If you accept to loose numerical precision up to 2 ulp, you can speedup your code by passing the following options to your compiler: (ffast-math or Ofast) and mrecip

512-bits vectorization

Workaround

Recompile with -mprefer-vector-width=512

Expensive FP math instructions/calls

Detected performance impact from expensive FP math instructions/calls. By removing/reexpressing them, you can lower the cost of an iteration from 34.00 to 15.00 cycles (2.27x speedup).

Masked instructions

Detected masked instructions.

Details

Vector registers are partially exploited, which is expected if your loop is irregular or mixes elements of different sizes.

Workaround

If your loop is irregular, try to remove or hoist conditional structures out of your loop. If it mixes elements of different sizes, try to uniformize them.

FMA

Detected 80 FMA (fused multiply-add) operations. Presence of both ADD/SUB and MUL operations.

Workaround

Complex instructions

Detected COMPLEX INSTRUCTIONS.

Details

These instructions generate more than one micro-operation and only one of them can be decoded during a cycle and the extra micro-operations increase pressure on execution units.

VCVTPD2PS: 2 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
VCVTPS2PD: 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29

Vector unaligned load/store instructions

Detected 4 optimal vector unaligned load/store instructions.

Details

VMOVUPS: 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:22
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:23
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:21
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:27

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.

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTPD2PS (FP64 to FP32, SIMD): 2 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
VCVTPS2PD (FP32 to FP64, SIMD): 4 occurrences►
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29
- /home/eoseret/qaas_runs_GNR/173-814-2357/intel/HACCmk/build/HACCmk/src/Step10_orig.c:29

Workaround

Type of elements and instruction set

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

The binary loop is composed of 240 FP arithmetical operations:

120: addition or subtraction (80 inside FMA instructions)
104: multiply (80 inside FMA instructions)
8: divide
8: square root

The binary loop is loading 128 bytes (32 single precision FP elements).

Arithmetic intensity

Arithmetic intensity is 1.87 FP operations per loaded or stored byte.

General properties

nb instructions	45
nb uops	50
loop length	250
used x86 registers	6
used mmx registers	0
used xmm registers	4
used ymm registers	27
used zmm registers	0
nb stack references	0
ADD-SUB / MUL ratio	1.50

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	8.33 cycles
front end	8.33 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	15.00	15.00	1.33	1.33	0.00	13.00	1.00	0.00	0.00	0.00	0.00	1.33
cycles	15.00	15.00	1.33	1.33	0.00	13.00	1.00	0.00	0.00	0.00	0.00	1.33

Execution ports to units layout:

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

Cycles executing div or sqrt instructions	34.00
Longest recurrence chain latency (RecMII)	4.00

Cycles summary

Front-end	8.33
Dispatch	15.00
DIV/SQRT	34.00
Data deps.	4.00
Overall L1	34.00

Vectorization ratios

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

Vector efficiency ratios

all	45%
load	50%
store	NA (no store vectorizable/vectorized instructions)
mul	50%
add-sub	50%
fma	50%
div/sqrt	50%
other	37%

Cycles and memory resources usage

Detected masked instructions: assuming all mask elements are active. Assuming all data fit into the L1 cache, each iteration of the binary loop takes 34.00 cycles. At this rate:

1% of peak load performance is reached (3.76 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 401a80

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
VMOVUPS (%RSI,%R11,1),%YMM8	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VMOVUPS (%RDX,%R11,1),%YMM1	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VSUBPS %YMM25,%YMM8,%YMM8	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VSUBPS %YMM24,%YMM1,%YMM7	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VMOVUPS (%RDI,%R11,1),%YMM9	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VMULPS %YMM8,%YMM8,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VSUBPS %YMM26,%YMM9,%YMM9	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VFMADD231PS %YMM9,%YMM9,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM7,%YMM7,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VCMPPS $0x1,%YMM15,%YMM1,%K1	1	0	0	0	0	1	0	0	0	3	1	vect (50.0%)
VADDPS %YMM23,%YMM1,%YMM5	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VCMPPS $0x1,%YMM15,%YMM1,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMOVUPS (%R8,%R11,1),%YMM13{%K1}	1	0	0	0.33	0.33	0	0	0	0.33	0-1	0.33	vect (50.0%)
VCVTPS2PD %XMM5,%YMM6	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VANDPS %YMM13,%YMM0,%YMM28	1	0.33	0.33	0	0	0.33	0	0	0	1	0.33	vect (50.0%)
VEXTRACTF32X4 $0x1,%YMM5,%XMM0	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VCVTPS2PD %XMM0,%YMM5	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VMOVAPS %YMM1,%YMM0	1	0	0	0	0	0	0	0	0	0-1	0.17	vect (50.0%)
VFMADD132PS %YMM22,%YMM21,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VSQRTPD %YMM6,%YMM30	1	1	0	0	0	0	0	0	0	13-19	9	vect (50.0%)
VCMPPS $0xe,%YMM16,%YMM1,%K2	1	0	0	0	0	1	0	0	0	3	1	vect (50.0%)
VSQRTPD %YMM5,%YMM29	1	1	0	0	0	0	0	0	0	13-19	9	vect (50.0%)
ADD $0x20,%R11	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VFMADD132PS %YMM1,%YMM20,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMULPD %YMM29,%YMM5,%YMM5	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM19,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM18,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD132PS %YMM1,%YMM17,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VMULPD %YMM30,%YMM6,%YMM1	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VCVTPS2PD %XMM0,%YMM30	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VDIVPD %YMM1,%YMM14,%YMM6	1	1	0	0	0	0	0	0	0	13-15	8	vect (50.0%)
VEXTRACTF32X4 $0x1,%YMM0,%XMM0	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VADDPD %YMM30,%YMM6,%YMM1	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VDIVPD %YMM5,%YMM14,%YMM6	1	1	0	0	0	0	0	0	0	13-15	8	vect (50.0%)
VCVTPS2PD %XMM0,%YMM5	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (25.0%)
VCVTPD2PS %YMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (50.0%)
VADDPD %YMM5,%YMM6,%YMM0	1	0	0.50	0	0	0.50	0	0	0	3	0.50	vect (50.0%)
VCVTPD2PS %YMM0,%XMM6	2	0.50	0.50	0	0	1	0	0	0	7	1	vect (50.0%)
VINSERTF32X4 $0x1,%XMM6,%YMM1,%YMM5	1	0	0	0	0	1	0	0	0	3	1	vect (25.0%)
VMULPS %YMM28,%YMM5,%YMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM9,%YMM10{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM8,%YMM11{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
VFMADD231PS %YMM0,%YMM7,%YMM12{%K2}	1	0.50	0.50	0	0	0	0	0	0	4	0.50	vect (50.0%)
CMP %R11,%RBX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 401a80 <Step10_orig+0xc0>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Report Configuration

Vectorization

Details

Execution units bottlenecks

Workaround

512-bits vectorization

Workaround

Expensive FP math instructions/calls

Masked instructions

Details

Workaround

FMA

Workaround

Complex instructions

Details

Vector unaligned load/store instructions

Details

Workaround

Conversion 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

Execution units bottlenecks

Workaround

512-bits vectorization

Workaround

Expensive FP math instructions/calls

Masked instructions

Details

Workaround

FMA

Workaround

Complex instructions

Details

Vector unaligned load/store instructions

Details

Workaround

Conversion 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