OV - attention-clang-gnr256 - Loop 64 - attention-clang-gnr256

	Loop Id: 64	Module: attention-clang-gnr256	Source: attention_v2.cpp:30-31	Coverage: 25.78%

0x5480 LEA	(%RAX,%R8,1),%R13D

0x5484 VMOVSS	(%RSI,%R13,4),%XMM1    [2]

0x548a VCVTSS2SD	%XMM1,%XMM1,%XMM1

0x548e MOV	%R11D,%R13D

0x5491 VMOVSS	(%RDI,%R13,4),%XMM2    [1]

0x5497 VCVTSS2SD	%XMM2,%XMM2,%XMM2

0x549b LEA	0x1(%RAX,%R8,1),%R13D

0x54a0 VMOVSS	(%RSI,%R13,4),%XMM3    [2]

0x54a6 VCVTSS2SD	%XMM3,%XMM3,%XMM3

0x54aa VFMADD213SD	%XMM0,%XMM1,%XMM2

0x54af MOV	%EBX,%R13D

0x54b2 VMOVSS	(%RDI,%R13,4),%XMM0    [1]

0x54b8 VCVTSS2SD	%XMM0,%XMM0,%XMM0

0x54bc VFMADD213SD	%XMM2,%XMM3,%XMM0

0x54c1 LEA	0x2(%RAX,%R8,1),%R13D

0x54c6 VMOVSS	(%RSI,%R13,4),%XMM1    [2]

0x54cc VCVTSS2SD	%XMM1,%XMM1,%XMM1

0x54d0 MOV	%R10D,%R13D

0x54d3 VMOVSS	(%RDI,%R13,4),%XMM2    [1]

0x54d9 VCVTSS2SD	%XMM2,%XMM2,%XMM2

0x54dd LEA	0x3(%RAX,%R8,1),%R13D

0x54e2 VMOVSS	(%RSI,%R13,4),%XMM3    [2]

0x54e8 VCVTSS2SD	%XMM3,%XMM3,%XMM3

0x54ec VFMADD213SD	%XMM0,%XMM1,%XMM2

0x54f1 MOV	%R9D,%R13D

0x54f4 VMOVSS	(%RDI,%R13,4),%XMM0    [1]

0x54fa VCVTSS2SD	%XMM0,%XMM0,%XMM0

0x54fe VFMADD213SD	%XMM2,%XMM3,%XMM0

0x5503 ADD	$0x4,%R8

0x5507 ADD	%RDX,%R9

0x550a ADD	%RDX,%R10

0x550d ADD	%RDX,%R11

0x5510 ADD	%RDX,%RBX

0x5513 CMP	%R8,%RCX

0x5516 JNE	5480

/home/eoseret/llm-attention/attention_v2.cpp: 30 - 31

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

30:          for (unsigned int k = 0; k < K; ++k) { //vectorized

31:             sum += (double)A[i * K + k] * (double)B[k * N + j];

Coverage (%)	Name	Source Location	Module

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	1.60
CQA speedup if fully vectorized	13.02
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.00
Bottlenecks
Function	main
Source	attention_v2.cpp:30-31
Source loop unroll info	multi-versionned
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	16.00
CQA cycles if no scalar integer	16.00
CQA cycles if FP arith vectorized	10.00
CQA cycles if fully vectorized	1.23
Front-end cycles	7.00
P0 cycles	7.50
P1 cycles	7.50
P2 cycles	2.67
P3 cycles	2.67
P4 cycles	0.00
P5 cycles	8.00
P6 cycles	3.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	3.00
P11 cycles	2.67
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	16
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	35.00
Nb uops	42.00
Nb loads	8.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.50
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	2.00
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	1.00
Stride indirect	0.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	8.33
Vector-efficiency ratio load	6.25
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	6.25

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.60
CQA speedup if fully vectorized	13.02
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	2.00
Bottlenecks
Function	main
Source	attention_v2.cpp:30-31
Source loop unroll info	multi-versionned
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	16.00
CQA cycles if no scalar integer	16.00
CQA cycles if FP arith vectorized	10.00
CQA cycles if fully vectorized	1.23
Front-end cycles	7.00
P0 cycles	7.50
P1 cycles	7.50
P2 cycles	2.67
P3 cycles	2.67
P4 cycles	0.00
P5 cycles	8.00
P6 cycles	3.00
P7 cycles	0.00
P8 cycles	0.00
P9 cycles	0.00
P10 cycles	3.00
P11 cycles	2.67
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	16
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	35.00
Nb uops	42.00
Nb loads	8.00
Nb stores	0.00
Nb stack references	0.00
FLOP/cycle	0.50
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	4.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	2.00
Bytes prefetched	0.00
Bytes loaded	32.00
Bytes stored	0.00
Stride 0	0.00
Stride 1	0.00
Stride n	1.00
Stride unknown	1.00
Stride indirect	0.00
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	NA
Vectorization ratio mul	NA
Vectorization ratio add_sub	0.00
Vectorization ratio fma	0.00
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	8.33
Vector-efficiency ratio load	6.25
Vector-efficiency ratio store	NA
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	12.50
Vector-efficiency ratio fma	12.50
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	6.25

Path /

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

Function	main
Source file and lines	attention_v2.cpp:30-31
Module	attention-clang-gnr256

The loop is defined in /home/eoseret/llm-attention/attention_v2.cpp:30-31.

The related source loop is multi-versionned.

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. Only 8% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 16.00 to 1.23 cycles (13.02x 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 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

Detected 4 FMA (fused multiply-add) operations.

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.

VCVTSS2SD: 8 occurrences►
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 1 occurrence(s)
Constant non-unit stride: 1 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

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

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTSS2SD (FP32 to FP64, scalar): 8 occurrences►
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31

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

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

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

The binary loop is composed of 8 FP arithmetical operations:

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

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

Arithmetic intensity

Arithmetic intensity is 0.25 FP operations per loaded or stored byte.

General properties

nb instructions	35
nb uops	42
loop length	156
used x86 registers	11
used mmx registers	0
used xmm registers	4
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	7.00 cycles
front end	7.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	7.50	7.50	2.67	2.67	0.00	8.00	3.00	0.00	0.00	0.00	3.00	2.67
cycles	7.50	7.50	2.67	2.67	0.00	8.00	3.00	0.00	0.00	0.00	3.00	2.67

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	NA
Longest recurrence chain latency (RecMII)	16.00

Cycles summary

Front-end	7.00
Dispatch	8.00
Data deps.	16.00
Overall L1	16.00

Vectorization ratios

INT

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

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

INT+FP

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

Vector efficiency ratios

INT

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

all	7%
load	6%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

INT+FP

all	8%
load	6%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 16.00 cycles. At this rate:

1% of peak load performance is reached (2.00 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: 5480

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
LEA (%RAX,%R8,1),%R13D	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1-2	0.20	N/A
VMOVSS (%RSI,%R13,4),%XMM1	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM1,%XMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
MOV %R11D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM2	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM2,%XMM2,%XMM2	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
LEA 0x1(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM3	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM3,%XMM3,%XMM3	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM0,%XMM1,%XMM2	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %EBX,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM0	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM0,%XMM0,%XMM0	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM2,%XMM3,%XMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
LEA 0x2(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM1	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM1,%XMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
MOV %R10D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM2	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM2,%XMM2,%XMM2	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
LEA 0x3(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM3	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM3,%XMM3,%XMM3	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM0,%XMM1,%XMM2	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %R9D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM0	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM0,%XMM0,%XMM0	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM2,%XMM3,%XMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
ADD $0x4,%R8	1	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD %RDX,%R9	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%R10	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%R11	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%RBX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
CMP %R8,%RCX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 5480 <main+0x2770>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Function	main
Source file and lines	attention_v2.cpp:30-31
Module	attention-clang-gnr256

The loop is defined in /home/eoseret/llm-attention/attention_v2.cpp:30-31.

The related source loop is multi-versionned.

gain
potential
hint
expert

Vectorization

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 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

Detected 4 FMA (fused multiply-add) operations.

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.

VCVTSS2SD: 8 occurrences►
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31

Slow data structures access

Detected data structures (typically arrays) that cannot be efficiently read/written

Details

Constant unknown stride: 1 occurrence(s)
Constant non-unit stride: 1 occurrence(s)

Non-unit stride (uncontiguous) accesses are not efficiently using data caches

Workaround

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

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTSS2SD (FP32 to FP64, scalar): 8 occurrences►
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31
- /home/eoseret/llm-attention/attention_v2.cpp:31

Workaround

Type of elements and instruction set

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

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

The binary loop is composed of 8 FP arithmetical operations:

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

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

Arithmetic intensity

Arithmetic intensity is 0.25 FP operations per loaded or stored byte.

General properties

nb instructions	35
nb uops	42
loop length	156
used x86 registers	11
used mmx registers	0
used xmm registers	4
used ymm registers	0
used zmm registers	0
nb stack references	0

Front-end

ASSUMED MACRO FUSION FIT IN UOP CACHE

micro-operation queue	7.00 cycles
front end	7.00 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11
uops	7.50	7.50	2.67	2.67	0.00	8.00	3.00	0.00	0.00	0.00	3.00	2.67
cycles	7.50	7.50	2.67	2.67	0.00	8.00	3.00	0.00	0.00	0.00	3.00	2.67

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	NA
Longest recurrence chain latency (RecMII)	16.00

Cycles summary

Front-end	7.00
Dispatch	8.00
Data deps.	16.00
Overall L1	16.00

Vectorization ratios

INT

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

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

INT+FP

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

Vector efficiency ratios

INT

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

all	7%
load	6%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

INT+FP

all	8%
load	6%
store	NA (no store vectorizable/vectorized instructions)
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	12%
fma	12%
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	6%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 16.00 cycles. At this rate:

1% of peak load performance is reached (2.00 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: 5480

Instruction	Nb FU	P0	P1	P2	P3	P5	P6	P10	P11	Latency	Recip. throughput	Vectorization
LEA (%RAX,%R8,1),%R13D	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1-2	0.20	N/A
VMOVSS (%RSI,%R13,4),%XMM1	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM1,%XMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
MOV %R11D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM2	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM2,%XMM2,%XMM2	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
LEA 0x1(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM3	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM3,%XMM3,%XMM3	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM0,%XMM1,%XMM2	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %EBX,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM0	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM0,%XMM0,%XMM0	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM2,%XMM3,%XMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
LEA 0x2(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM1	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM1,%XMM1,%XMM1	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
MOV %R10D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM2	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM2,%XMM2,%XMM2	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
LEA 0x3(%RAX,%R8,1),%R13D	1	0	1	0	0	0	0	0	0	3	1	N/A
VMOVSS (%RSI,%R13,4),%XMM3	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM3,%XMM3,%XMM3	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM0,%XMM1,%XMM2	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %R9D,%R13D	1	0	0	0	0	0	0	0	0	1	0.17	N/A
VMOVSS (%RDI,%R13,4),%XMM0	1	0	0	0.33	0.33	0	0	0	0.33	1	0.33	scal (6.3%)
VCVTSS2SD %XMM0,%XMM0,%XMM0	2	0.50	0.50	0	0	1	0	0	0	4-5	1	scal (6.3%)
VFMADD213SD %XMM2,%XMM3,%XMM0	1	0.50	0.50	0	0	0	0	0	0	4	0.50	scal (12.5%)
ADD $0x4,%R8	1	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD %RDX,%R9	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%R10	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%R11	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
ADD %RDX,%RBX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	scal (12.5%)
CMP %R8,%RCX	1	0.20	0.20	0	0	0.20	0.20	0.20	0	1	0.20	N/A
JNE 5480 <main+0x2770>	1	0.50	0	0	0	0	0.50	0	0	0	0.50	N/A

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

FMA

Complex instructions

Details

Slow data structures access

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

Workaround

Execution units bottlenecks

FMA

Complex instructions

Details

Slow data structures access

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