OV - attention-gcc-znver5-512 - Loop 21 - attention-gcc-znver5-512

	Loop Id: 21	Module: attention-gcc-znver5-512	Source: random.tcc:458-3374 [...]	Coverage: 0.15%

0x401a40 LEA	0x1(%RCX),%RDI

0x401a44 MOV	-0x13c0(%RBP,%RCX,8),%RCX    [1]

0x401a4c MOV	%RDI,-0x40(%RBP)    [1]

0x401a50 MOV	%RCX,%RAX

0x401a53 SHR	$0xb,%RAX

0x401a57 MOV	%EAX,%EAX

0x401a59 XOR	%RCX,%RAX

0x401a5c MOV	%RAX,%RCX

0x401a5f SAL	$0x7,%RCX

0x401a63 AND	$-0x62d3a980,%ECX

0x401a69 XOR	%RAX,%RCX

0x401a6c MOV	%RCX,%RAX

0x401a6f SAL	$0xf,%RAX

0x401a73 AND	$-0x103a0000,%EAX

0x401a78 XOR	%RCX,%RAX

0x401a7b MOV	%RAX,%RCX

0x401a7e SHR	$0x12,%RCX

0x401a82 XOR	%RCX,%RAX

0x401a85 VCVTUSI2SS	%RAX,%XMM1,%XMM0

0x401a8b VMULSS	%XMM2,%XMM0,%XMM0

0x401a8f VCMPSS	$0x2,%XMM0,%XMM3,%XMM5

0x401a94 VBLENDVPS	%XMM5,%XMM4,%XMM0,%XMM0

0x401a9a VMOVSS	%XMM0,(%R12)    [2]

0x401aa0 CMP	$0x26f,%RDI

0x401aa7 JA	401bfe

0x401aad MOV	-0x13c0(%RBP,%RDI,8),%RCX    [1]

0x401ab5 LEA	0x1(%RDI),%R8

0x401ab9 MOV	%R8,-0x40(%RBP)    [1]

0x401abd MOV	%RCX,%RAX

0x401ac0 SHR	$0xb,%RAX

0x401ac4 MOV	%EAX,%EAX

0x401ac6 XOR	%RCX,%RAX

0x401ac9 MOV	%RAX,%RCX

0x401acc SAL	$0x7,%RCX

0x401ad0 AND	$-0x62d3a980,%ECX

0x401ad6 XOR	%RAX,%RCX

0x401ad9 MOV	%RCX,%RAX

0x401adc SAL	$0xf,%RAX

0x401ae0 AND	$-0x103a0000,%EAX

0x401ae5 XOR	%RCX,%RAX

0x401ae8 MOV	%RAX,%RCX

0x401aeb SHR	$0x12,%RCX

0x401aef XOR	%RCX,%RAX

0x401af2 VCVTUSI2SS	%RAX,%XMM1,%XMM0

0x401af8 VMULSS	%XMM2,%XMM0,%XMM0

0x401afc VCMPSS	$0x2,%XMM0,%XMM3,%XMM5

0x401b01 VBLENDVPS	%XMM5,%XMM4,%XMM0,%XMM0

0x401b07 VMOVSS	%XMM0,(%R13)    [3]

0x401b0d CMP	$0x26f,%R8

0x401b14 JA	401bcd

0x401b1a MOV	-0x13c0(%RBP,%R8,8),%RDI    [1]

0x401b22 LEA	0x1(%R8),%RCX

0x401b26 ADD	$0x4,%R14

0x401b2a ADD	$0x4,%R13

0x401b2e MOV	%RCX,-0x40(%RBP)    [1]

0x401b32 ADD	$0x4,%R12

0x401b36 MOV	%RDI,%RAX

0x401b39 SHR	$0xb,%RAX

0x401b3d MOV	%EAX,%EAX

0x401b3f XOR	%RDI,%RAX

0x401b42 MOV	%RAX,%RDI

0x401b45 SAL	$0x7,%RDI

0x401b49 AND	$-0x62d3a980,%EDI

0x401b4f XOR	%RAX,%RDI

0x401b52 MOV	%RDI,%RAX

0x401b55 SAL	$0xf,%RAX

0x401b59 AND	$-0x103a0000,%EAX

0x401b5e XOR	%RDI,%RAX

0x401b61 MOV	%RAX,%RDI

0x401b64 SHR	$0x12,%RDI

0x401b68 XOR	%RDI,%RAX

0x401b6b VCVTUSI2SS	%RAX,%XMM1,%XMM0

0x401b71 VMULSS	%XMM2,%XMM0,%XMM0

0x401b75 VCMPSS	$0x2,%XMM0,%XMM3,%XMM5

0x401b7a VBLENDVPS	%XMM5,%XMM4,%XMM0,%XMM0

0x401b80 VMOVSS	%XMM0,-0x4(%R14)    [4]

0x401b86 CMP	%R14,%RBX

0x401b89 JE	401c2f

0x401b8f CMP	$0x26f,%RCX

0x401b96 JBE	401a40

0x401b9c LEA	-0x13c0(%RBP),%RDI

0x401ba3 CALL	4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>

0x401ba8 VMOVSS	0x2658(%RIP),%XMM4    [5]

0x401bb0 VMOVSS	0x2658(%RIP),%XMM3    [5]

0x401bb8 VMOVSS	0x264c(%RIP),%XMM2    [5]

0x401bc0 MOV	-0x40(%RBP),%RCX    [1]

0x401bc4 VXORPS	%XMM1,%XMM1,%XMM1

0x401bc8 JMP	401a40

0x401bcd LEA	-0x13c0(%RBP),%RDI

0x401bd4 CALL	4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>

0x401bd9 VMOVSS	0x2627(%RIP),%XMM4    [5]

0x401be1 VMOVSS	0x2627(%RIP),%XMM3    [5]

0x401be9 VMOVSS	0x261b(%RIP),%XMM2    [5]

0x401bf1 MOV	-0x40(%RBP),%R8    [1]

0x401bf5 VXORPS	%XMM1,%XMM1,%XMM1

0x401bf9 JMP	401b1a

0x401bfe LEA	-0x13c0(%RBP),%RDI

0x401c05 CALL	4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>

0x401c0a VMOVSS	0x25f6(%RIP),%XMM4    [5]

0x401c12 VMOVSS	0x25f6(%RIP),%XMM3    [5]

0x401c1a VMOVSS	0x25ea(%RIP),%XMM2    [5]

0x401c22 MOV	-0x40(%RBP),%RDI    [1]

0x401c26 VXORPS	%XMM1,%XMM1,%XMM1

0x401c2a JMP	401aad

/cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc: 458 - 3374

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

458:       if (_M_p >= state_size)

459: 	_M_gen_rand();

460:

461:       // Calculate o(x(i)).

462:       result_type __z = _M_x[_M_p++];

463:       __z ^= (__z >> __u) & __d;

464:       __z ^= (__z << __s) & __b;

465:       __z ^= (__z << __t) & __c;

466:       __z ^= (__z >> __l);

[...]

3367: 	  __sum += _RealType(__urng() - __urng.min()) * __tmp;

3368: 	  __tmp *= __r;

3369: 	}

3370:       __ret = __sum / __tmp;

3371:       if (__builtin_expect(__ret >= _RealType(1), 0))

3372: 	{

3373: #if _GLIBCXX_USE_C99_MATH_FUNCS

3374: 	  __ret = std::nextafter(_RealType(1), _RealType(0));

/home/eoseret/llm-attention/attention_v2.cpp: 164 - 167

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

164:       for (size_t i = 0; i < elemsW; ++i) {

165:          h_WQ[i] = dist(rng);

166:          h_WK[i] = dist(rng);

167:          h_WV[i] = dist(rng);

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	2.24
CQA speedup if FP arith vectorized	2.58
CQA speedup if fully vectorized	13.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.65
Bottlenecks	micro-operation queue,
Function	main
Source	random.tcc:458-466,random.tcc:3367-3374,attention_v2.cpp:164-167
Source loop unroll info	unrolled by 3
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	3
CQA cycles	13.75
CQA cycles if no scalar integer	6.13
CQA cycles if FP arith vectorized	5.32
CQA cycles if fully vectorized	1.06
Front-end cycles	13.75
P0 cycles	8.33
P1 cycles	8.33
P2 cycles	8.33
P3 cycles	8.33
P4 cycles	8.33
P5 cycles	8.33
P6 cycles	5.25
P7 cycles	5.25
P8 cycles	5.25
P9 cycles	5.25
P10 cycles	3.00
P11 cycles	3.00
P12 cycles	3.00
P13 cycles	3.00
P14 cycles	1.50
P15 cycles	1.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	104.00
Nb uops	110.00
Nb loads	15.00
Nb stores	6.00
Nb stack references	2.00
FLOP/cycle	0.22
Nb FLOP add-sub	0.00
Nb FLOP mul	3.00
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	8.73
Bytes prefetched	0.00
Bytes loaded	84.00
Bytes stored	36.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	10.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	0.00
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	14.29
Vector-efficiency ratio all	11.25
Vector-efficiency ratio load	6.25
Vector-efficiency ratio store	9.38
Vector-efficiency ratio mul	6.25
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.95

Metric	Value
CQA speedup if no scalar integer	2.24
CQA speedup if FP arith vectorized	2.58
CQA speedup if fully vectorized	13.00
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.65
Bottlenecks	micro-operation queue,
Function	main
Source	random.tcc:458-466,random.tcc:3367-3374,attention_v2.cpp:164-167
Source loop unroll info	unrolled by 3
Source loop unroll confidence level	max
Unroll/vectorization loop type	main
Unroll factor	3
CQA cycles	13.75
CQA cycles if no scalar integer	6.13
CQA cycles if FP arith vectorized	5.32
CQA cycles if fully vectorized	1.06
Front-end cycles	13.75
P0 cycles	8.33
P1 cycles	8.33
P2 cycles	8.33
P3 cycles	8.33
P4 cycles	8.33
P5 cycles	8.33
P6 cycles	5.25
P7 cycles	5.25
P8 cycles	5.25
P9 cycles	5.25
P10 cycles	3.00
P11 cycles	3.00
P12 cycles	3.00
P13 cycles	3.00
P14 cycles	1.50
P15 cycles	1.50
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	104.00
Nb uops	110.00
Nb loads	15.00
Nb stores	6.00
Nb stack references	2.00
FLOP/cycle	0.22
Nb FLOP add-sub	0.00
Nb FLOP mul	3.00
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	8.73
Bytes prefetched	0.00
Bytes loaded	84.00
Bytes stored	36.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	10.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	0.00
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	14.29
Vector-efficiency ratio all	11.25
Vector-efficiency ratio load	6.25
Vector-efficiency ratio store	9.38
Vector-efficiency ratio mul	6.25
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.95

Path /

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

Function	main
Source file and lines	random.tcc:458-3374
Module	attention-gcc-znver5-512

The loop is defined in:

/cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:458-466,3367-3374
/home/eoseret/llm-attention/attention_v2.cpp:164-167

It is main loop of related source loop which is unrolled by 3 (including vectorization).
Warnings:

Ignoring paths for analysis
Too many paths. Rerun with max-paths=8
RecMII not computed since number of paths is unknown or > max_paths
Streams not analyzed since number of paths is unknown or > max_paths

Try to simplify control and/or increase the maximum number of paths per function/loop through the 'max-paths-nb' option.

This loop has 8 execution paths.

The presence of multiple execution paths is typically the main/first bottleneck.
Try to simplify control inside loop: ideally, try to remove all conditional expressions, for example by (if applicable):

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

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 13.75 to 6.13 cycles (2.24x speedup).

Workaround

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

Vectorization

Your loop is probably not vectorized. Only 11% 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 13.75 to 1.06 cycles (13.00x speedup).

Details

Store and arithmetical 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
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.

No data for this section

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.

std::mersenne_twister_engine
::_M_gen_rand(): 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:459

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.

VCVTUSI2SS: 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTUSI2SS (INT32/64 to FP32, scalar): 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367

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

6 SSE or AVX instructions are processing arithmetic or math operations on single 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 3 FP arithmetical operations:

3: multiply

The binary loop is loading 84 bytes (21 single precision FP elements). The binary loop is storing 36 bytes (9 single precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.02 FP operations per loaded or stored byte.

General properties

nb instructions	104
nb uops	110
loop length	495
used x86 registers	9
used mmx registers	0
used xmm registers	6
used ymm registers	0
used zmm registers	0
nb stack references	2

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	13.75 cycles
front end	13.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	8.33	8.33	8.33	8.33	8.33	8.33	5.25	5.25	5.25	5.25	3.00	3.00	3.00	3.00	1.50	1.50
cycles	8.33	8.33	8.33	8.33	8.33	8.33	5.25	5.25	5.25	5.25	3.00	3.00	3.00	3.00	1.50	1.50

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

Cycles summary

Front-end	13.75
Dispatch	8.33
Overall L1	13.75

Vectorization ratios

INT

all	0%
load	NA (no load vectorizable/vectorized instructions)
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)
other	0%

all	25%
load	0%
store	0%
mul	0%
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	66%

INT+FP

all	10%
load	0%
store	0%
mul	0%
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	14%

Vector efficiency ratios

INT

all	11%
load	NA (no load vectorizable/vectorized instructions)
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)
other	11%

all	10%
load	6%
store	6%
mul	6%
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	18%

INT+FP

all	11%
load	6%
store	9%
mul	6%
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	12%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 13.75 cycles. At this rate:

2% of peak load performance is reached (6.11 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (2.62 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). By removing all these bottlenecks, you can lower the cost of an iteration from 13.75 to 8.33 cycles (1.65x speedup).

ASM code

In the binary file, the address of the loop is: 401b8f

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
LEA 0x1(%RCX),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV -0x13c0(%RBP,%RCX,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %RDI,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,(%R12)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP $0x26f,%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JA 401bfe <main+0x8fe>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
MOV -0x13c0(%RBP,%RDI,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
LEA 0x1(%RDI),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %R8,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,(%R13)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP $0x26f,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JA 401bcd <main+0x8cd>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
MOV -0x13c0(%RBP,%R8,8),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
LEA 0x1(%R8),%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x4,%R14	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x4,%R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
ADD $0x4,%R12	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RDI,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RDI	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%EDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RDI,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RDI	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,-0x4(%R14)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 401c2f <main+0x92f>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
CMP $0x26f,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JBE 401a40 <main+0x740>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x2658(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x2658(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x264c(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401a40 <main+0x740>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x2627(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x2627(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x261b(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401b1a <main+0x81a>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x25f6(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x25f6(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x25ea(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401aad <main+0x7ad>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A

Function	main
Source file and lines	random.tcc:458-3374
Module	attention-gcc-znver5-512

The loop is defined in:

/cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:458-466,3367-3374
/home/eoseret/llm-attention/attention_v2.cpp:164-167

It is main loop of related source loop which is unrolled by 3 (including vectorization).
Warnings:

Ignoring paths for analysis
Too many paths. Rerun with max-paths=8
RecMII not computed since number of paths is unknown or > max_paths
Streams not analyzed since number of paths is unknown or > max_paths

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

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 13.75 to 6.13 cycles (2.24x 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
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.

No data for this section

CALL instructions

Detected function call instructions.

Details

std::mersenne_twister_engine
::_M_gen_rand(): 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:459

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.

VCVTUSI2SS: 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367

Conversion instructions

Detected expensive conversion instructions.

Details

VCVTUSI2SS (INT32/64 to FP32, scalar): 3 occurrences►
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367
- /cluster/comp/gcc/15.1.0/include/c++/15.1.0/bits/random.tcc:3367

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

6 SSE or AVX instructions are processing arithmetic or math operations on single 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 3 FP arithmetical operations:

3: multiply

The binary loop is loading 84 bytes (21 single precision FP elements). The binary loop is storing 36 bytes (9 single precision FP elements).

Arithmetic intensity

Arithmetic intensity is 0.02 FP operations per loaded or stored byte.

General properties

nb instructions	104
nb uops	110
loop length	495
used x86 registers	9
used mmx registers	0
used xmm registers	6
used ymm registers	0
used zmm registers	0
nb stack references	2

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	13.75 cycles
front end	13.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	8.33	8.33	8.33	8.33	8.33	8.33	5.25	5.25	5.25	5.25	3.00	3.00	3.00	3.00	1.50	1.50
cycles	8.33	8.33	8.33	8.33	8.33	8.33	5.25	5.25	5.25	5.25	3.00	3.00	3.00	3.00	1.50	1.50

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

Cycles summary

Front-end	13.75
Dispatch	8.33
Overall L1	13.75

Vectorization ratios

INT

all	0%
load	NA (no load vectorizable/vectorized instructions)
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)
other	0%

all	25%
load	0%
store	0%
mul	0%
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	66%

INT+FP

all	10%
load	0%
store	0%
mul	0%
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	14%

Vector efficiency ratios

INT

all	11%
load	NA (no load vectorizable/vectorized instructions)
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)
other	11%

all	10%
load	6%
store	6%
mul	6%
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	18%

INT+FP

all	11%
load	6%
store	9%
mul	6%
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	12%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 13.75 cycles. At this rate:

2% of peak load performance is reached (6.11 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
2% of peak store performance is reached (2.62 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

ASM code

In the binary file, the address of the loop is: 401b8f

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15	Latency	Recip. throughput	Vectorization
LEA 0x1(%RCX),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV -0x13c0(%RBP,%RCX,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
MOV %RDI,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,(%R12)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP $0x26f,%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JA 401bfe <main+0x8fe>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
MOV -0x13c0(%RBP,%RDI,8),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
LEA 0x1(%RDI),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %R8,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%ECX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RCX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RCX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RCX,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,(%R13)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP $0x26f,%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JA 401bcd <main+0x8cd>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
MOV -0x13c0(%RBP,%R8,8),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
LEA 0x1(%R8),%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x4,%R14	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
ADD $0x4,%R13	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RCX,-0x40(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	4	0.50	scal (12.5%)
ADD $0x4,%R12	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RDI,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SHR $0xb,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
MOV %EAX,%EAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (6.3%)
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SAL $0x7,%RDI	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
AND $-0x62d3a980,%EDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
XOR %RAX,%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
MOV %RDI,%RAX	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	scal (12.5%)
SAL $0xf,%RAX	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	scal (12.5%)
AND $-0x103a0000,%EAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (6.3%)
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
MOV %RAX,%RDI	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.13	N/A
SHR $0x12,%RDI	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	1	0.33	N/A
XOR %RDI,%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
VCVTUSI2SS %RAX,%XMM1,%XMM0	2	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	3-9	1	scal (12.5%)
VMULSS %XMM2,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	3	0.50	scal (6.3%)
VCMPSS $0x2,%XMM0,%XMM3,%XMM5	1	0	0	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	2	0.50	scal (6.3%)
VBLENDVPS %XMM5,%XMM4,%XMM0,%XMM0	1	0	0	0	0	0	0	0	0	0	0	0.50	0.50	0	0	0	0	2	0.50	vect (25.0%)
VMOVSS %XMM0,-0x4(%R14)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0.50	0.50	1	0.50	scal (6.3%)
CMP %R14,%RBX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	scal (12.5%)
JE 401c2f <main+0x92f>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
CMP $0x26f,%RCX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
JBE 401a40 <main+0x740>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.33-0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x2658(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x2658(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x264c(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401a40 <main+0x740>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x2627(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x2627(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x261b(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401b1a <main+0x81a>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A
LEA -0x13c0(%RBP),%RDI	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	0	0	0	0	0	0	1	0.17	N/A
CALL 4033b0 <_ZNSt23mersenne_twister_engineImLm32ELm624ELm397ELm31ELm2567483615ELm11ELm4294967295ELm7ELm2636928640ELm15ELm4022730752ELm18ELm1812433253EE11_M_gen_randEv>	2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	1	N/A
VMOVSS 0x25f6(%RIP),%XMM4	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x25f6(%RIP),%XMM3	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
VMOVSS 0x25ea(%RIP),%XMM2	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	1	0.50	scal (6.3%)
MOV -0x40(%RBP),%RDI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	0	0	0	0	0	0	3	0.25	N/A
VXORPS %XMM1,%XMM1,%XMM1	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0.17	vect (25.0%)
JMP 401aad <main+0x7ad>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	0	0	0	0	0	0	0.50	N/A

Report Configuration

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

CALL instructions

Details

Complex instructions

Details

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

Code clean check

Workaround

Vectorization

Details

Workaround

Execution units bottlenecks

CALL instructions

Details

Complex instructions

Details

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