OV - exec - Function hypre_ParCSRRelaxThreads._omp

	Function: hypre_ParCSRRelaxThreads._omp_fn.0	Module: exec	Source: ams.c:3655-3659	Coverage (incl. loops): 1.57% \| (excl. loops): 0.00%

/home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c: 3655 - 3659

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

3655: #pragma omp parallel for private(i) HYPRE_SMP_SCHEDULE

3656: #endif

3657:       for (i = 0; i < n; i++)

3658:       {

3659:          Vtemp_data[i] = u_data[i];

0x4ef7b0 PUSH	%RBP

0x4ef7b1 MOV	%RSP,%RBP

0x4ef7b4 PUSH	%R12

0x4ef7b6 MOV	%RDI,%R12

0x4ef7b9 PUSH	%RBX

0x4ef7ba CALL	4110b0 <omp_get_num_threads@plt>

0x4ef7bf MOV	%EAX,%EBX

0x4ef7c1 CALL	4111f0 <omp_get_thread_num@plt>

0x4ef7c6 MOVSXD	%EAX,%RCX

0x4ef7c9 MOV	(%R12),%RAX

0x4ef7cd MOVSXD	%EBX,%RSI

0x4ef7d0 CQTO

0x4ef7d2 IDIV	%RSI

0x4ef7d5 CMP	%RDX,%RCX

0x4ef7d8 JL	4ef930

0x4ef7de IMUL	%RAX,%RCX

0x4ef7e2 ADD	%RCX,%RDX

0x4ef7e5 ADD	%RDX,%RAX

0x4ef7e8 CMP	%RAX,%RDX

0x4ef7eb JGE	4ef929

0x4ef7f1 MOV	%RAX,%RDI

0x4ef7f4 SUB	%RDX,%RDI

0x4ef7f7 MOV	0x10(%R12),%R8

0x4ef7fc MOV	0x8(%R12),%R9

0x4ef801 AND	$0x7,%EDI

0x4ef804 JE	4ef8a0

0x4ef80a CMP	$0x1,%RDI

0x4ef80e JE	4ef888

0x4ef810 CMP	$0x2,%RDI

0x4ef814 JE	4ef879

0x4ef816 CMP	$0x3,%RDI

0x4ef81a JE	4ef86a

0x4ef81c CMP	$0x4,%RDI

0x4ef820 JE	4ef85b

0x4ef822 CMP	$0x5,%RDI

0x4ef826 JE	4ef84c

0x4ef828 CMP	$0x6,%RDI

0x4ef82c JE	4ef83d

0x4ef82e VMOVSD	(%R9,%RDX,8),%XMM0

0x4ef834 VMOVSD	%XMM0,(%R8,%RDX,8)

0x4ef83a INC	%RDX

0x4ef83d VMOVSD	(%R9,%RDX,8),%XMM1

0x4ef843 VMOVSD	%XMM1,(%R8,%RDX,8)

0x4ef849 INC	%RDX

0x4ef84c VMOVSD	(%R9,%RDX,8),%XMM2

0x4ef852 VMOVSD	%XMM2,(%R8,%RDX,8)

0x4ef858 INC	%RDX

0x4ef85b VMOVSD	(%R9,%RDX,8),%XMM3

0x4ef861 VMOVSD	%XMM3,(%R8,%RDX,8)

0x4ef867 INC	%RDX

0x4ef86a VMOVSD	(%R9,%RDX,8),%XMM4

0x4ef870 VMOVSD	%XMM4,(%R8,%RDX,8)

0x4ef876 INC	%RDX

0x4ef879 VMOVSD	(%R9,%RDX,8),%XMM5

0x4ef87f VMOVSD	%XMM5,(%R8,%RDX,8)

0x4ef885 INC	%RDX

0x4ef888 VMOVSD	(%R9,%RDX,8),%XMM6

0x4ef88e VMOVSD	%XMM6,(%R8,%RDX,8)

0x4ef894 INC	%RDX

0x4ef897 CMP	%RDX,%RAX

0x4ef89a JE	4ef929

(2049) 0x4ef8a0 VMOVSD	(%R9,%RDX,8),%XMM7

(2049) 0x4ef8a6 LEA	0x1(%RDX),%R10

(2049) 0x4ef8aa VMOVSD	%XMM7,(%R8,%RDX,8)

(2049) 0x4ef8b0 VMOVSD	(%R9,%R10,8),%XMM8

(2049) 0x4ef8b6 LEA	0x2(%RDX),%R11

(2049) 0x4ef8ba VMOVSD	%XMM8,(%R8,%R10,8)

(2049) 0x4ef8c0 VMOVSD	(%R9,%R11,8),%XMM9

(2049) 0x4ef8c6 LEA	0x3(%RDX),%R12

(2049) 0x4ef8ca VMOVSD	%XMM9,(%R8,%R11,8)

(2049) 0x4ef8d0 VMOVSD	(%R9,%R12,8),%XMM10

(2049) 0x4ef8d6 LEA	0x4(%RDX),%RBX

(2049) 0x4ef8da VMOVSD	%XMM10,(%R8,%R12,8)

(2049) 0x4ef8e0 VMOVSD	(%R9,%RBX,8),%XMM11

(2049) 0x4ef8e6 LEA	0x5(%RDX),%RCX

(2049) 0x4ef8ea VMOVSD	%XMM11,(%R8,%RBX,8)

(2049) 0x4ef8f0 VMOVSD	(%R9,%RCX,8),%XMM12

(2049) 0x4ef8f6 LEA	0x6(%RDX),%RSI

(2049) 0x4ef8fa VMOVSD	%XMM12,(%R8,%RCX,8)

(2049) 0x4ef900 VMOVSD	(%R9,%RSI,8),%XMM13

(2049) 0x4ef906 LEA	0x7(%RDX),%RDI

(2049) 0x4ef90a VMOVSD	%XMM13,(%R8,%RSI,8)

(2049) 0x4ef910 VMOVSD	(%R9,%RDI,8),%XMM14

(2049) 0x4ef916 ADD	$0x8,%RDX

(2049) 0x4ef91a VMOVSD	%XMM14,(%R8,%RDI,8)

(2049) 0x4ef920 CMP	%RDX,%RAX

(2049) 0x4ef923 JNE	4ef8a0

0x4ef929 POP	%RBX

0x4ef92a POP	%R12

0x4ef92c POP	%RBP

0x4ef92d RET

0x4ef92e XCHG	%AX,%AX

0x4ef930 INC	%RAX

0x4ef933 XOR	%EDX,%EDX

0x4ef935 JMP	4ef7de

0x4ef93a NOPW	(%RAX,%RAX,1)

Coverage (%)	Name	Source Location	Module
○97.63	gomp_thread_start	team.c:130	libgomp.so.1.0.0
○2.37	GOMP_parallel	libgomp.h:982	libgomp.so.1.0.0

Path /

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

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run gcc_5

Source file and lines	ams.c:3655-3659
Module	exec

The function is defined in /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655-3659.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 12.67 to 5.00 cycles (2.53x speedup).

Workaround

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

Vectorization

Your function is not vectorized. Only 11% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your function, you can lower the cost of an iteration from 12.67 to 5.00 cycles (2.53x speedup).

Details

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

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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.

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 12.67 to 12.00 cycles (1.06x speedup).

CALL instructions

Detected function call instructions.

Details

Calling (and then returning from) a function prevents many compiler optimizations (like vectorization), breaks control flow (which reduces pipeline performance) and executes extra instructions to save/restore the registers used inside it, which is very expensive (dozens of cycles). Consider to inline small functions.

omp_get_num_threads@plt: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655
omp_get_thread_num@plt: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

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.

IDIV: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

Conversion instructions

Detected expensive conversion instructions.

Details

CQTO: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

Type of elements and instruction set

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

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 80 bytes. The binary function is storing 56 bytes.

General properties

nb instructions	70
nb uops	76
loop length	263
used x86 registers	11
used mmx registers	0
used xmm registers	7
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	12.67 cycles
front end	12.67 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	6.20	6.20	4.67	4.67	6.00	6.20	6.20	6.00	6.00	6.00	6.20	4.67
cycles	6.20	7.27	4.67	4.67	6.00	6.20	6.20	6.00	6.00	6.00	6.20	4.67

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

10.00

Cycles summary

Front-end	12.67
Dispatch	7.27
DIV/SQRT	10.00
Overall L1	12.67

Vectorization ratios

INT

all	0%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

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

INT+FP

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

Vector efficiency ratios

INT

all	11%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
other	11%

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

INT+FP

all	11%
load	12%
store	12%
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	12%
other	10%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 12.67 cycles. At this rate:

3% of peak load performance is reached (6.32 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))
3% of peak store performance is reached (4.42 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck).

ASM code

In the binary file, the address of the function is: 4ef7b0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
PUSH %RBP	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
MOV %RSP,%RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
PUSH %R12	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
MOV %RDI,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
PUSH %RBX	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
CALL 4110b0 <omp_get_num_threads@plt>	2	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	0	1	N/A
MOV %EAX,%EBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
CALL 4111f0 <omp_get_thread_num@plt>	2	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	0	1	N/A
MOVSXD %EAX,%RCX	1	0	0.33	0	0	0	0.33	0	0	0	0	0.33	0	1	0.33	scal (6.3%)
MOV (%R12),%RAX	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
MOVSXD %EBX,%RSI	1	0	0.33	0	0	0	0.33	0	0	0	0	0.33	0	1	0.33	scal (6.3%)
CQTO	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	1	0.50	scal (12.5%)
IDIV %RSI	5	0	3	0	0	0	0	0	0	0	0	0	0	11-16	10	scal (12.5%)
CMP %RDX,%RCX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JL 4ef930 <hypre_ParCSRRelaxThreads._omp_fn.0+0x180>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
IMUL %RAX,%RCX	1	0	1	0	0	0	0	0	0	0	0	0	0	3	1	scal (12.5%)
ADD %RCX,%RDX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
ADD %RDX,%RAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
CMP %RAX,%RDX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JGE 4ef929 <hypre_ParCSRRelaxThreads._omp_fn.0+0x179>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
SUB %RDX,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
MOV 0x10(%R12),%R8	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
MOV 0x8(%R12),%R9	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
AND $0x7,%EDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	scal (6.3%)
JE 4ef8a0 <hypre_ParCSRRelaxThreads._omp_fn.0+0xf0>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x1,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef888 <hypre_ParCSRRelaxThreads._omp_fn.0+0xd8>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x2,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef879 <hypre_ParCSRRelaxThreads._omp_fn.0+0xc9>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x3,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef86a <hypre_ParCSRRelaxThreads._omp_fn.0+0xba>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x4,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef85b <hypre_ParCSRRelaxThreads._omp_fn.0+0xab>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x5,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef84c <hypre_ParCSRRelaxThreads._omp_fn.0+0x9c>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x6,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef83d <hypre_ParCSRRelaxThreads._omp_fn.0+0x8d>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
VMOVSD (%R9,%RDX,8),%XMM0	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM0,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM1	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM1,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM2	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM2,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM3	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM3,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM4	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM4,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM5	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM5,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM6	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM6,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %RDX,%RAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef929 <hypre_ParCSRRelaxThreads._omp_fn.0+0x179>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
POP %RBX	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
POP %R12	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
POP %RBP	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
RET	1	0.50	0	0.33	0.33	0	0	0.50	0	0	0	0	0.33	0	2.13	N/A
XCHG %AX,%AX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
INC %RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %EDX,%EDX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
JMP 4ef7de <hypre_ParCSRRelaxThreads._omp_fn.0+0x2e>	1	0	0	0	0	0	0	0	0	0	0	0	0	0	2.08	N/A
NOPW (%RAX,%RAX,1)	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A

The code analyzed by CQA in that panel excludes loops and represents 0.00% of application time for run gcc_5

Source file and lines	ams.c:3655-3659
Module	exec

The function is defined in /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655-3659.

Warnings:
Ignoring paths for analysis

gain
potential
hint
expert

Code clean check

Detected a slowdown caused by scalar integer instructions (typically used for address computation). By removing them, you can lower the cost of an iteration from 12.67 to 5.00 cycles (2.53x speedup).

Workaround

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

Vectorization

Details

Workaround

Try another compiler or update/tune your current one
Make array accesses unit-stride:
- If your function 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.

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 12.67 to 12.00 cycles (1.06x speedup).

CALL instructions

Detected function call instructions.

Details

omp_get_num_threads@plt: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655
omp_get_thread_num@plt: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

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.

IDIV: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

Conversion instructions

Detected expensive conversion instructions.

Details

CQTO: 1 occurrences►
- /home/eoseret/qaas_runs_GNR/173-927-0874/intel/AMG/build/AMG/AMG/parcsr_ls/ams.c:3655

Workaround

Avoid mixing data with different types. In particular, check if the type of constants is the same as array elements.

Type of elements and instruction set

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

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

The binary function does not contain any FP arithmetical operations. The binary function is loading 80 bytes. The binary function is storing 56 bytes.

General properties

nb instructions	70
nb uops	76
loop length	263
used x86 registers	11
used mmx registers	0
used xmm registers	7
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	12.67 cycles
front end	12.67 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	6.20	6.20	4.67	4.67	6.00	6.20	6.20	6.00	6.00	6.00	6.20	4.67
cycles	6.20	7.27	4.67	4.67	6.00	6.20	6.20	6.00	6.00	6.00	6.20	4.67

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

10.00

Cycles summary

Front-end	12.67
Dispatch	7.27
DIV/SQRT	10.00
Overall L1	12.67

Vectorization ratios

INT

all	0%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	0%
add-sub	0%
fma	NA (no fma vectorizable/vectorized instructions)
other	0%

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

INT+FP

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

Vector efficiency ratios

INT

all	11%
load	NA (no load vectorizable/vectorized instructions)
store	NA (no store vectorizable/vectorized instructions)
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
other	11%

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

INT+FP

all	11%
load	12%
store	12%
mul	12%
add-sub	12%
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	12%
other	10%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each call to the function takes 12.67 cycles. At this rate:

3% of peak load performance is reached (6.32 out of 192.00 bytes loaded per cycle (GB/s @ 1GHz))
3% of peak store performance is reached (4.42 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Performance is limited by instruction throughput (loading/decoding program instructions to execution core) (front-end is a bottleneck).

ASM code

In the binary file, the address of the function is: 4ef7b0

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	Latency	Recip. throughput	Vectorization
PUSH %RBP	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
MOV %RSP,%RBP	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
PUSH %R12	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
MOV %RDI,%R12	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
PUSH %RBX	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	5-12	0.50	N/A
CALL 4110b0 <omp_get_num_threads@plt>	2	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	0	1	N/A
MOV %EAX,%EBX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
CALL 4111f0 <omp_get_thread_num@plt>	2	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	0	1	N/A
MOVSXD %EAX,%RCX	1	0	0.33	0	0	0	0.33	0	0	0	0	0.33	0	1	0.33	scal (6.3%)
MOV (%R12),%RAX	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
MOVSXD %EBX,%RSI	1	0	0.33	0	0	0	0.33	0	0	0	0	0.33	0	1	0.33	scal (6.3%)
CQTO	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	1	0.50	scal (12.5%)
IDIV %RSI	5	0	3	0	0	0	0	0	0	0	0	0	0	11-16	10	scal (12.5%)
CMP %RDX,%RCX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JL 4ef930 <hypre_ParCSRRelaxThreads._omp_fn.0+0x180>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
IMUL %RAX,%RCX	1	0	1	0	0	0	0	0	0	0	0	0	0	3	1	scal (12.5%)
ADD %RCX,%RDX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
ADD %RDX,%RAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	N/A
CMP %RAX,%RDX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JGE 4ef929 <hypre_ParCSRRelaxThreads._omp_fn.0+0x179>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
SUB %RDX,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
MOV 0x10(%R12),%R8	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
MOV 0x8(%R12),%R9	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	N/A
AND $0x7,%EDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1-2	0.20	scal (6.3%)
JE 4ef8a0 <hypre_ParCSRRelaxThreads._omp_fn.0+0xf0>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x1,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef888 <hypre_ParCSRRelaxThreads._omp_fn.0+0xd8>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x2,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef879 <hypre_ParCSRRelaxThreads._omp_fn.0+0xc9>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x3,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef86a <hypre_ParCSRRelaxThreads._omp_fn.0+0xba>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x4,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef85b <hypre_ParCSRRelaxThreads._omp_fn.0+0xab>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x5,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef84c <hypre_ParCSRRelaxThreads._omp_fn.0+0x9c>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
CMP $0x6,%RDI	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef83d <hypre_ParCSRRelaxThreads._omp_fn.0+0x8d>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
VMOVSD (%R9,%RDX,8),%XMM0	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM0,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM1	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM1,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM2	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM2,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM3	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM3,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM4	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM4,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM5	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM5,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSD (%R9,%RDX,8),%XMM6	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1	0.33	scal (12.5%)
VMOVSD %XMM6,(%R8,%RDX,8)	1	0	0	0	0	0.50	0	0	0.50	0.50	0.50	0	0	1	0.50	scal (12.5%)
INC %RDX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CMP %RDX,%RAX	1	0.20	0.20	0	0	0	0.20	0.20	0	0	0	0.20	0	1	0.20	scal (12.5%)
JE 4ef929 <hypre_ParCSRRelaxThreads._omp_fn.0+0x179>	1	0.50	0	0	0	0	0	0.50	0	0	0	0	0	0	0.50	N/A
POP %RBX	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
POP %R12	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
POP %RBP	1	0	0	0.33	0.33	0	0	0	0	0	0	0	0.33	1-6	0.33	N/A
RET	1	0.50	0	0.33	0.33	0	0	0.50	0	0	0	0	0.33	0	2.13	N/A
XCHG %AX,%AX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
INC %RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %EDX,%EDX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
JMP 4ef7de <hypre_ParCSRRelaxThreads._omp_fn.0+0x2e>	1	0	0	0	0	0	0	0	0	0	0	0	0	0	2.08	N/A
NOPW (%RAX,%RAX,1)	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A

Name	Coverage (%)	Time (s)
▼hypre_ParCSRRelaxThreads._omp_fn.0–	1.57	0.68
○Loop 2049 - ams.c:3659-3659 - exec	1.57	0.59

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Expensive FP math instructions/calls

CALL instructions

Details

Complex instructions

Details

Conversion instructions

Details

Workaround

Type of elements and instruction set

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

Expensive FP math instructions/calls

CALL instructions

Details

Complex instructions

Details

Conversion instructions

Details

Workaround

Type of elements and instruction set

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

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code