Intel® Architecture Code Analyzer

User's Guide

Copyright © 2009-2012 Intel Corporation

All Rights Reserved

Document Number: 321356-001US

Revision: 2.0.1

World Wide Web: http://www.intel.com

Document Number: 321356-001US

Introduction

Legal Information
INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE, EXPRESS OR IMPLIED, BY
ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN
INTEL'S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL
DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR
WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT,
COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.
A "Mission Critical Application" is any application in which failure of the Intel Product could result, directly or indirectly, in personal
injury or death. SHOULD YOU PURCHASE OR USE INTEL'S PRODUCTS FOR ANY SUCH MISSION CRITICAL APPLICATION, YOU SHALL
INDEMNIFY AND HOLD INTEL AND ITS SUBSIDIARIES, SUBCONTRACTORS AND AFFILIATES, AND THE DIRECTORS, OFFICERS, AND
EMPLOYEES OF EACH, HARMLESS AGAINST ALL CLAIMS COSTS, DAMAGES, AND EXPENSES AND REASONABLE ATTORNEYS' FEES
ARISING OUT OF, DIRECTLY OR INDIRECTLY, ANY CLAIM OF PRODUCT LIABILITY, PERSONAL INJURY, OR DEATH ARISING IN ANY WAY
OUT OF SUCH MISSION CRITICAL APPLICATION, WHETHER OR NOT INTEL OR ITS SUBCONTRACTOR WAS NEGLIGENT IN THE DESIGN,
MANUFACTURE, OR WARNING OF THE INTEL PRODUCT OR ANY OF ITS PARTS.
Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the absence
or characteristics of any features or instructions marked "reserved" or "undefined". Intel reserves these for future definition and shall
have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The information here is subject
to change without notice. Do not finalize a design with this information.
The products described in this document may contain design defects or errors known as errata which may cause the product to deviate
from published specifications. Current characterized errata are available on request.
Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order.
Copies of documents which have an order number and are referenced in this document, or other Intel literature, may be obtained by
calling 1-800-548-4725, or go to: http://www.intel.com/design/literature.htm
This document contains information on products in the design phase of development.
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.
Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software,
operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and
performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when
combined with other products.
BlueMoon, BunnyPeople, Celeron, Celeron Inside, Centrino, Centrino Inside, Cilk, Core Inside, E-GOLD, Flexpipe, i960, Intel, the Intel
logo, Intel AppUp, Intel Atom, Intel Atom Inside, Intel Core, Intel Inside, Intel Insider, the Intel Inside logo, Intel NetBurst, Intel
NetMerge, Intel NetStructure, Intel SingleDriver, Intel SpeedStep, Intel Sponsors of Tomorrow., the Intel Sponsors of Tomorrow. logo,
Intel StrataFlash, Intel vPro, Intel XScale, InTru, the InTru logo, the InTru Inside logo, InTru soundmark, Itanium, Itanium Inside, MCS,
MMX, Moblin, Pentium, Pentium Inside, Puma, skoool, the skoool logo, SMARTi, Sound Mark, Stay With It, The Creators Project, The
Journey Inside, Thunderbolt, Ultrabook, vPro Inside, VTune, Xeon, Xeon Inside, X-GOLD, XMM, X-PMU and XPOSYS are trademarks of
Intel Corporation in the U.S. and/or other countries.
* Other names and brands may be claimed as the property of others.

Copyright (C) 2009-2012, Intel Corporation. All rights reserved.

Intel® Architecture Code Analyzer User Guide 2

Introduction

Contents
1 Introduction ...................................................................................................................... 4
1.1 Intel® Architecture Code Analyzer Accuracy.................................................................. 4
1.2 Processor Support ...................................................................................................... 4
1.3 Platform Support ....................................................................................................... 4
2 Analysis ............................................................................................................................. 5
2.1 Throughput Analysis ................................................................................................... 5
2.2 Latency Analysis ........................................................................................................ 6
2.3 Graphs ................................................................................................................... 7
2.4 Analysis Report Notes................................................................................................. 8
2.4.1 Unbound Instructions .................................................................................... 8
2.4.2 Combining 256-bit Intel® AVX and Legacy Intel® SSE ...................................... 8
2.4.3 Unsupported Instructions ............................................................................... 8

3 Using Intel® Architecture Code Analyzer ........................................................................... 9

3.1 Building Your Binary ................................................................................................... 9
3.2 Command Line Options ..............................................................................................10
3.3 Analysis Errors .........................................................................................................10

4 Examples ......................................................................................................................... 11
4.1 Throughput Analysis – 4x4 Matrix Multiply ...................................................................11
4.1.1 Initial Code Version ......................................................................................11
4.1.2 Optimization ...............................................................................................12
4.2 Latency and Graph Analysis – Add Reduction................................................................13
4.2.1 Initial Code Version ......................................................................................13
4.2.2 Optimization 1 .............................................................................................14
4.2.3 Optimization 2 .............................................................................................15
5 Release Contents ............................................................................................................. 16
5.1 Windows* OS ...........................................................................................................16
5.2 Linux* OS ................................................................................................................17
5.3 Mac OS X* ...............................................................................................................18

Intel® Architecture Code Analyzer User Guide 3

Introduction

1 Introduction
Intel® Architecture Code Analyzer helps you statically analyze the data dependency,
latency, and throughput of instruction sequences (kernels) on Intel® microarchitectures.

For a given binary, Intel Architecture Code Analyzer:

• Identifies the binding of the kernel instructions to the processor ports under ideal front-
end, out-of-order engine and memory hierarchy conditions.
• Performs static analysis of throughput and latency and reports their cycle counts.
• Identifies the critical path(s).

1.1 Intel® Architecture Code Analyzer Accuracy

The Intel Architecture Code Analyzer enables you to do a first order estimate of the
relative performance of sections of code on different microarchitectures. It does not
provide absolute performance numbers.

The performance data reported by the tool may significantly deviate from actual
performance observed on an Intel® processor. You can achieve the most accurate
throughput and latency measurements by executing the analyzed code on the processor
itself. The Intel® Architecture Code Analyzer complements such measured data with
information on port binding, bottlenecks, and critical paths.

1.2 Processor Support

The Intel Architecture Code Analyzer supports analysis for 1st, 2nd and 3rd generation
Intel® Core™ processors, which correspond to Intel® microarchitectures codenamed
Nehalem (1st gen), Westmere (1st gen), Sandy Bridge (2nd gen) and Ivy Bridge (3rd gen).

1.3 Platform Support

Intel Architecture Code Analyzer is a command-line utility that can analyze a binary file
that contains code with special markers that delimit the analyzed code. The tool is capable
of analyzing both IA-32 and Intel® 64 code, including Intel® Advanced Vector Extensions
(Intel® AVX) instructions.

Intel Architecture Code Analyzer is available on Windows*, Linux*, and Mac OS X*

operating systems. Both IA-32 and Intel® 64 operating systems are supported. Intel® 64
code can be analyzed on IA-32 operating systems and vice versa.

NOTE: Intel® Architecture Code Analyzer has been validated on 64-bit Windows* 7, 64-bit
Ubuntu* 10.04, and Mac OS X* 10.6 and 10.7. It should work on other versions of
Windows*, Linux*, and Mac OS X* operating systems.

Intel® Architecture Code Analyzer User Guide 4

Analysis

2 Analysis
Intel® Architecture Code Analyzer performs two different types of analysis: Throughput
and Latency.

2.1 Throughput Analysis

The Throughput Analysis is used to analyze the throughput and bottlenecks of a loop body;
it treats the contents of the analyzed block as an infinite loop, including considering inter-
iteration dependencies between instructions within the analyzed block. The Throughput
Analysis report provides the following information:
• Throughput of the whole analyzed block, counted in cycles. The block throughput
is calculated as the maximum between:
o Throughput of the processor ports
o Maximum front-end throughput (4 micro-ops per cycle)
o Divider unit throughput
• Bottleneck source that limited the throughput: front-end, port number, divider unit,
or inter-iteration.
• Total number of cycles each processor port was bound by micro-ops.

The detailed section of the throughput analysis report contains one line for each instruction
in the analyzed block. Each line contains:
• Number of the instruction micro-ops.
• Average number of cycles per iteration that the instruction was bound to each
processor port. For most instructions this simply means the number of cycles the
instruction was bound to each port. However, if a particular micro-op may execute
on more than one port, the average number of cycles per iteration may be a partial
cycle for each port because that micro-op may bind to a different port on each
iteration.
• An indication whether the instruction is on the critical path of the analyzed code.
The critical path for Throughput Analysis is all instructions that use the throughput
bottleneck.
• Instruction disassembly in Intel® Software Developer’s Manual (MASM) style
Some ports have both a regular pipe and a secondary pipe. These ports are separated by a
hyphen, and look like two separate ports in the detailed report. Specifically:
• Port 0 has the Divider pipe split from it. In the first cycle they are both busy, then
port 0 is available for the next micro-op and the Divider pipe is kept busy for the
duration of the divide operation.
• Load ports 2 and 3 have an Address Generation Unit (AGU) split from them. For
256-bit load operations that keep the port busy for two cycles, the AGU gets freed
after the first cycle and can process a store address generation if such micro-op is
available for execution.
Following is an example Throughput Analysis report:

Intel® Architecture Code Analyzer User Guide 5

 Analysis

Throughput Analysis Report

--------------------------
Block Throughput: 28.00 Cycles Throughput Bottleneck: Divider

Port Binding In Cycles Per Iteration:

-------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------------------------
| Cycles | 4.0 28.0 | 1.0 | 1.5 2.0 | 1.5 2.0 | 2.0 | 1.0 |
-------------------------------------------------------------------------

N - port number or number of cycles resource conflict caused delay, DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3), CP - on a critical path
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion happened
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256 instruction, dozens of cycles penalty is expected
! - instruction not supported, was not accounted in Analysis

| Num Of | Ports pressure in cycles | |

| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | |
------------------------------------------------------------------------------
| 1 | | | 1.0 2.0 | | | | | vmovups ymm0, [rbp-0x70]
| 1 | | | | 1.0 2.0 | | | | vmovups ymm1, [rbp-0x50]
| 1 | 1.0 | | | | | | | vmulps ymm2, ymm0, ymm0
| 1 | 1.0 | | | | | | | vmulps ymm3, ymm1, ymm1
| 1 | | 1.0 | | | | | | vaddps ymm4, ymm2, ymm3
| 3 | 2.0 28.0 | | | | | 1.0 | CP | vsqrtps ymm5, ymm4
| 2^ | | | 0.5 | 0.5 | 2.0 | | | vmovups [rbp-0x30], ymm5

2.2 Latency Analysis

The Latency Analysis is used to analyze the latency and resource conflicts in a section of
code; unlike the throughput analysis, it does not treat the code section as a loop. The
Latency Analysis reports the following information:
• Latency of the analyzed code section.
• Resource delay of instructions. A resource delay occurs when all the instruction
sources are ready but the execution unit (front end / execution port / divider) is
occupied.
• The instructions on a path that has the longest latency (including resource delays)
is marked with CP. There may be several critical paths with the same execution
latency.
• Total resource conflict delay for each execution unit.
• Performance dependency between instructions.

Following is an example Latency Analysis report:

Intel® Architecture Code Analyzer User Guide 6

 Analysis

Latency Analysis Report

-----------------------
Latency: 59 Cycles

N - port number or number of cycles resource conflict caused delay, DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3), CP - on a critical path
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion happened
# - ESP Tracking sync uop was issued
@ - Intel(R) AVX to Intel(R) SSE code switch, dozens of cycles penalty is expected
! - instruction not supported, was not accounted in Analysis

The Resource delay is counted since all the sources of the instructions are ready
and until the needed resource becomes available

| Inst | Resource Delay In Cycles | |

| Num | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | FE | |
---------------------------------------------------------------
| 0 | | | | | | | | | vmovups ymm0, ymmword ptr [rbp-0x70]
| 1 | | | | | | | | CP | vmovups ymm1, ymmword ptr [rbp-0x50]
| 2 | | | | | | | | | vmulps ymm2, ymm0, ymm0
| 3 | 1 | | | | | | | CP | vmulps ymm3, ymm1, ymm1
| 4 | | | | | | | | CP | vaddps ymm4, ymm2, ymm3
| 5 | | | | | | | | CP | vsqrtps ymm5, ymm4
| 6 | | | | | | | 2 | CP | vmovups ymmword ptr [rbp-0x30], ymm5

Resource Conflict on Critical Paths:

-------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------
| Cycles | 1 0 | 0 | 0 0 | 0 0 | 0 | 0 |
-------------------------------------------------------

List Of Delays On Critical Paths

-------------------------------
2 --> 3 1 Cycles Delay On Port0

2.3 Graphs
Use the –graph option to set Intel® Architecture Code Analyzer to output the data
dependency graph.

TIP: Graph files produced by Intel® Architecture Code Analyzer can be opened with graphviz.

The data dependency graph may be different for throughput analysis as the throughput
analysis treats the analyzed code block as an infinite loop block, so there may be inter-
iteration dependencies. Red nodes in the graph indicate instructions that are on the critical
path for that particular analysis.

Intel® Architecture Code Analyzer User Guide 7

Analysis

2.4 Analysis Report Notes

2.4.1 Unbound Instructions
Some instructions do not require a processor functional unit to complete their execution.
For example, a xor eax, eax instruction does not require an execution port because the
register is directly set to 0. As a result, their micro-ops are not bound to any port.
Instructions that are not bound to a port are marked with a ‘*’ character next to their
number of micro-ops.

2.4.2 Combining 256-bit Intel® AVX and Legacy Intel® SSE

Transitioning between 256-bit Intel® AVX instructions and legacy Intel Streaming SIMD
Extensions (Intel® SSE) instructions will cause performance penalties. Intel® Architecture
Code Analyzer detects these transitions between 256-bit Intel® AVX and legacy Intel®
SSE within the analyzed block, and ignores the associated performance penalty in the
total throughput and total latency summary report. Instead, the summary report includes
two additional lines at the top indicating that such sequence(s) exist in the analyzed block,
and marks the first transition instruction with a ‘@’ character in the Num of Uops columns.

For more information on transitions between Intel® AVX and Intel® SSE, see Avoiding
AVX-SSE Transition Penalties.

2.4.3 Unsupported Instructions

Intel® Architecture Code Analyzer does not support a small subset of the Intel®
Architecture Instruction Set. When it reaches an unsupported instruction in the analyzed
block it ignores the instruction. It does not take the instruction into account in the port
binding analysis or in the throughput and latency calculations.

In such cases, the summary report includes two additional lines at the top indicating that
such instruction(s) exist in your code, and marks the instruction with a ‘!’ character in all
columns.

Intel® Architecture Code Analyzer User Guide 8

Using Intel® Architecture Code Analyzer

3 Using Intel® Architecture Code Analyzer

This section explains how to build your binary so that the Intel® Architecture Code
Analyzer can analyze it, and it lists the tool command-line options.

3.1 Building Your Binary

The file iacaMarks.h contains macros to denote the start (IACA_START) and end
(IACA_END) of the code section for the Intel® Architecture Code Analyzer to evaluate. The
Intel Architecture Code Analyzer is a static tool. It treats the analyzed code section as a
single consecutive block of instructions. It does not follow branch instructions, not even
unconditional branches.

When analyzing a loop construct, place the macros at the following locations:

while ( condition )
{
IACA_START
<loop body>
}
IACA_END
This placement skips the loop initialization and includes the loop-end branch instruction.

These macros modify the ebx register in IA-32 code. As a result, the compiler saves this
register just before the macro and restores it immediately after the macro. This adds POP
and PUSH instructions at the beginning and end of the analyzed block. By default, Intel®
Architecture Code Analyzer ignores those instructions, as they are not part of the original
code. See section 3.2 how to force the tool analyze those instructions.

For Microsoft* Visual C++ compiler, 64-bit version, use IACA_VC64_START and
IACA_VC64_END, instead.

Once you insert the macros into your code, build your code into an executable file or an
object file.

NOTE: Input files generated with the Intel compiler option –Qipo are not supported.

Intel® Architecture Code Analyzer User Guide 9

Using Intel® Architecture Code Analyzer

3.2 Command Line Options

The following command runs the Intel® Architecture Code Analyzer:

iaca <options> <input file name>

<input file name> represents the name of the input file.

Available <options>:

-32 32-bit input file (default)

-64 64-bit input file (required for 64-bit object files only)

Architecture type.
-arch <type> These are the available types: NHM, WSM, SNB, IVB

-analysis <type> Analysis type: LATENCY, THROUGHPUT (default)

Specifies an output file. The default is stdout. To ensure

your output appears correctly, specify an output file. The
-o <file> stdout output line widh is limited to 80 characters, but
output files have no line width limit.
Specifies an output file for the analysis graph, which can
–graph <file> be viewed with graphviz.
Ignores added pop ebx / push ebx due to Intel
-ignore <boolean> Architecture Code Analyzer Markers. true ignores,
false does not.

-report Generate error report.

3.3 Analysis Errors

Should the analysis fail, the following error messages may appear:

Error message Possible Cause

The supplied path for the input or output file was

COULD NOT OPEN FILE - <file name> incorrect, the input file is not readable or failed to
create the output file.

Code contains an illegal instruction in the specified

ILLEGAL INSTRUCTION - <offset> byte offset.

Mixed files between multiple Intel® Architecture

INCORRECT XED2 VERSION Code Analyzer releases.

COULD NOT FIND START_MARKER Code did not contain the proper marker(s). See
section 3.1 for more details.
COULD NOT FIND END_MARKER
Intel® Architecture Code Analyzer cannot
CAN'T DETERMINE MODE, PLEASE USE determine the supplied file format (32-bit or 64-
ONE OF -32/-64 COMMAND LINE OPTIONS bit). Use the -32 or -64 option to specify.

Intel® Architecture Code Analyzer User Guide 10

 Examples

4 Examples
This section provides examples of how to analyze and optimize code using Intel®
Architecture Code Analyzer.

4.1 Throughput Analysis – 4x4 Matrix Multiply

This example performs a multiply of two 4x4 matrices using Intel® AVX. The initial code
and throughput analysis report are shown below.

4.1.1 Initial Code Version

Throughput Analysis Report
--------------------------
Block Throughput: 12.00 Cycles Throughput Bottleneck: Port5

Port Binding In Cycles Per Iteration:

-------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------------------------
| Cycles | 8.0 0.0 | 6.0 | 4.0 4.0 | 4.0 4.0 | 4.0 | 12.0 |
-------------------------------------------------------------------------

| Num Of | Ports pressure in cycles | |

| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | |
---------------------------------------------------------------------
| 2^ | | | 1.0 1.0 | | | 1.0 | CP | vbroadcastf128 ymm9, xmmword ptr [rcx]
| 2^ | | | | 1.0 1.0 | | 1.0 | CP | vbroadcastf128 ymm10, xmmword ptr [rcx+0x10]
| 2^ | | | 1.0 1.0 | | | 1.0 | CP | vbroadcastf128 ymm11, xmmword ptr [rcx+0x20]
| 2^ | | | | 1.0 1.0 | | 1.0 | CP | vbroadcastf128 ymm12, xmmword ptr [rcx+0x30]
| 1 | | | 1.0 2.0 | | | | | vmovaps ymm0, ymmword ptr [rax]
| 1 | | | | | | 1.0 | CP | vpermilps ymm1, ymm0, 0x0
| 1 | | | | | | 1.0 | CP | vpermilps ymm2, ymm0, 0x55
| 1 | | | | | | 1.0 | CP | vpermilps ymm3, ymm0, 0xcc
| 1 | | | | | | 1.0 | CP | vpermilps ymm4, ymm0, 0xff
| 1 | | | | 1.0 2.0 | | | | vmovaps ymm0, ymmword ptr [rax+0x20]
| 1 | | | | | | 1.0 | CP | vpermilps ymm5, ymm0, 0x0
| 1 | | | | | | 1.0 | CP | vpermilps ymm6, ymm0, 0x55
| 1 | | | | | | 1.0 | CP | vpermilps ymm7, ymm0, 0xcc
| 1 | | | | | | 1.0 | CP | vpermilps ymm8, ymm0, 0xff
| 1 | 1.0 | | | | | | | vmulps ymm1, ymm1, ymm9
| 1 | 1.0 | | | | | | | vmulps ymm2, ymm2, ymm10
| 1 | 1.0 | | | | | | | vmulps ymm3, ymm3, ymm11
| 1 | 1.0 | | | | | | | vmulps ymm4, ymm4, ymm12
| 1 | | 1.0 | | | | | | vaddps ymm1, ymm1, ymm2
| 1 | | 1.0 | | | | | | vaddps ymm3, ymm3, ymm4
| 1 | | 1.0 | | | | | | vaddps ymm1, ymm1, ymm3
| 1 | 1.0 | | | | | | | vmulps ymm5, ymm5, ymm9
| 1 | 1.0 | | | | | | | vmulps ymm6, ymm6, ymm10
| 1 | 1.0 | | | | | | | vmulps ymm7, ymm7, ymm11
| 1 | 1.0 | | | | | | | vmulps ymm8, ymm8, ymm12
| 1 | | 1.0 | | | | | | vaddps ymm5, ymm5, ymm6
| 1 | | 1.0 | | | | | | vaddps ymm7, ymm7, ymm8
| 1 | | 1.0 | | | | | | vaddps ymm5, ymm5, ymm7
| 2^ | | | 1.0 | | 2.0 | | | vmovaps ymmword ptr [rdx], ymm1
| 2^ | | | | 1.0 | 2.0 | | | vmovaps ymmword ptr [rdx+0x20], ymm5

Intel® Architecture Code Analyzer User Guide 11

 Examples

4.1.2 Optimization
The Throughput Analysis Report shows that the total throughput (Block Throughput) is 12
cycles, and port 5 was most pressured (Throughput Bottleneck), with 12 micro-ops
allocated to it.

Examination of the instructions that bind to port 5 in the instruction analysis report shows
that the instructions were broadcasts and vpermilps. The broadcasts can only execute on
port 5, but replacing them with 128-bit loads followed by vinsertf128 instructions
reduces the pressure on port 5 because vinsertf128 can execute on port 0. These
changes reduced the throughput to 10 cycles.

Throughput Analysis Report

--------------------------
Block Throughput: 10.00 Cycles Throughput Bottleneck: Port0, Port5

Port Binding In Cycles Per Iteration:

-------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------------------------
| Cycles | 10.0 0.0 | 6.0 | 6.0 6.0 | 6.0 6.0 | 4.0 | 10.0 |
-------------------------------------------------------------------------

| Num Of | Ports pressure in cycles | |

| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | |
---------------------------------------------------------------------
| 1 | | | 1.0 1.0 | | | | | vmovaps xmm9, xmmword ptr [rcx]
| 1 | | | | 1.0 1.0 | | | | vmovaps xmm10, xmmword ptr [rcx+0x10]
| 1 | | | 1.0 1.0 | | | | | vmovaps xmm11, xmmword ptr [rcx+0x20]
| 1 | | | | 1.0 1.0 | | | | vmovaps xmm12, xmmword ptr [rcx+0x30]
| 2 | 0.1 | | 1.0 1.0 | | | 0.9 | CP | vinsertf128 ymm9, ymm9, xmmword ptr [rcx], 0x1
| 2 | 0.9 | | | 1.0 1.0 | | 0.1 | CP | vinsertf128 ymm10, ymm10, xmmword ptr [rcx+0x10], 0x1
| 2 | | | 1.0 1.0 | | | 1.0 | CP | vinsertf128 ymm11, ymm11, xmmword ptr [rcx+0x20], 0x1
| 2 | 1.0 | | | 1.0 1.0 | | | CP | vinsertf128 ymm12, ymm12, xmmword ptr [rcx+0x30], 0x1
| 1 | | | 1.0 2.0 | | | | | vmovaps ymm0, ymmword ptr [rax]
| 1 | | | | | | 1.0 | CP | vpermilps ymm1, ymm0, 0x0
| 1 | | | | | | 1.0 | CP | vpermilps ymm2, ymm0, 0x55
| 1 | | | | | | 1.0 | CP | vpermilps ymm3, ymm0, 0xcc
| 1 | | | | | | 1.0 | CP | vpermilps ymm4, ymm0, 0xff
| 1 | | | | 1.0 2.0 | | | | vmovaps ymm0, ymmword ptr [rax+0x20]
| 1 | | | | | | 1.0 | CP | vpermilps ymm5, ymm0, 0x0
| 1 | | | | | | 1.0 | CP | vpermilps ymm6, ymm0, 0x55
| 1 | | | | | | 1.0 | CP | vpermilps ymm7, ymm0, 0xcc
| 1 | | | | | | 1.0 | CP | vpermilps ymm8, ymm0, 0xff
| 1 | 1.0 | | | | | | CP | vmulps ymm1, ymm1, ymm9
| 1 | 1.0 | | | | | | CP | vmulps ymm2, ymm2, ymm10
| 1 | 1.0 | | | | | | CP | vmulps ymm3, ymm3, ymm11
| 1 | 1.0 | | | | | | CP | vmulps ymm4, ymm4, ymm12
| 1 | | 1.0 | | | | | | vaddps ymm1, ymm1, ymm2
| 1 | | 1.0 | | | | | | vaddps ymm3, ymm3, ymm4
| 1 | | 1.0 | | | | | | vaddps ymm1, ymm1, ymm3
| 1 | 1.0 | | | | | | CP | vmulps ymm5, ymm5, ymm9
| 1 | 1.0 | | | | | | CP | vmulps ymm6, ymm6, ymm10
| 1 | 1.0 | | | | | | CP | vmulps ymm7, ymm7, ymm11
| 1 | 1.0 | | | | | | CP | vmulps ymm8, ymm8, ymm12
| 1 | | 1.0 | | | | | | vaddps ymm5, ymm5, ymm6
| 1 | | 1.0 | | | | | | vaddps ymm7, ymm7, ymm8
| 1 | | 1.0 | | | | | | vaddps ymm5, ymm5, ymm7
| 2^ | | | 1.0 | | 2.0 | | | vmovaps ymmword ptr [rdx], ymm1
| 2^ | | | | 1.0 | 2.0 | | | vmovaps ymmword ptr [rdx+0x20], ymm5

Intel® Architecture Code Analyzer User Guide 12

Examples

4.2 Latency and Graph Analysis – Add Reduction

This example performs an add reduction on 8 XMM registers. The initial code, latency
analysis report, and dependency graph (produced with the –graph option) are shown
below.

4.2.1 Initial Code Version

Latency Analysis Report
-----------------------
Latency: 7 Cycles

| Inst | Resource Delay In Cycles | |

| Num | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | FE | |
---------------------------------------------------------------
| 0 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm1
| 1 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm2
| 2 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm3
| 3 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm4
| 4 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm5
| 5 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm6
| 6 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm7

Resource Conflict on Critical Paths:

-------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------
| Cycles | 0 0 | 0 | 0 0 | 0 0 | 0 | 0 |
-------------------------------------------------------

List Of Delays On Critical Paths

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

Intel® Architecture Code Analyzer User Guide 13

Examples

4.2.2 Optimization 1
The analysis report and graph show that all instructions are on the same data dependency
path because they all depend on xmm0. We can optimize this code by constructing an add
tree, which reduces the dependency between instructions. This change reduced the
latency from 7 to 5 cycles.

Latency Analysis Report

-----------------------
Latency: 5 Cycles

| Inst | Resource Delay In Cycles | |

| Num | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | FE | |
---------------------------------------------------------------
| 0 | | | | | | | | | vpaddd xmm0, xmm0, xmm1
| 1 | | | | | | | | | vpaddd xmm2, xmm2, xmm3
| 2 | | | | | | | | | vpaddd xmm0, xmm0, xmm2
| 3 | | | | | | 1 | | | vpaddd xmm4, xmm4, xmm5
| 4 | | 1 | | | | | 1 | CP | vpaddd xmm6, xmm6, xmm7
| 5 | | | | | | | | CP | vpaddd xmm4, xmm4, xmm6
| 6 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm4

Resource Conflict on Critical Paths:

-------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------
| Cycles | 0 0 | 1 | 0 0 | 0 0 | 0 | 0 |
-------------------------------------------------------

List Of Delays On Critical Paths

-------------------------------
2 --> 4 1 Cycles Delay On Port1

Intel® Architecture Code Analyzer User Guide 14

Examples

4.2.3 Optimization 2
The analysis report tells us that instruction 4 (vpaddd xmm6, xmm6, xmm7) was delayed by
instruction 2 due to a resource conflict, and that instruction 4 is on a critical path. Because
instruction 5 depends on instruction 4 and instruction 6 depends on instruction 5, both of
these instructions are also delayed, and these last three add insturctions can only be
executed at a rate of one per cycle. The result from instruction 2 (vpaddd xmm0, xmm0,
xmm2) are not needed until instruction 6 (vpaddd xmm0, xmm0 xmm4), so we can resolve
this issue by moving vpaddd xmm0, xmm0, xmm2 lower in the add tree, which enables the
code to fully utilize the resources, reducing the latency to 4 cycles.

latency Analysis Report

-----------------------
Latency: 4 Cycles

| Inst | Resource Delay In Cycles | |

| Num | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | FE | |
---------------------------------------------------------------
| 0 | | | | | | | | | vpaddd xmm0, xmm0, xmm1
| 1 | | | | | | | | | vpaddd xmm4, xmm4, xmm5
| 2 | | 1 | | | | | | CP | vpaddd xmm2, xmm2, xmm3
| 3 | | | | | | 1 | | CP | vpaddd xmm6, xmm6, xmm7
| 4 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm2
| 5 | | | | | | | | CP | vpaddd xmm4, xmm4, xmm6
| 6 | | | | | | | | CP | vpaddd xmm0, xmm0, xmm4

Resource Conflict on Critical Paths:

-------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 |
-------------------------------------------------------
| Cycles | 0 0 | 1 | 0 0 | 0 0 | 0 | 1 |
-------------------------------------------------------

List Of Delays On Critical Paths

-------------------------------
0 --> 2 1 Cycles Delay On Port1
1 --> 3 1 Cycles Delay On Port5

Intel® Architecture Code Analyzer User Guide 15

Release Contents

5 Release Contents
This section lists the files required for running on Windows*, Linux*, and Mac OS X*
operating systems to analyze IA-32 and Intel® 64 code. Each section also explains which
environmental variables to modify.

5.1 Windows* OS
Add the iaca-mac32 directory to the PATH environment variable.

Include include/iacaMarks.h in your code.

Filename Description
iaca.exe Intel® Architecture Code Analyzer command-line tool.
iacaLoader.dll Intel Architecture Code Analyzer shared library.
iacaLogicNHM.dll
iacaLogicWSM.dll Intel Architecture Code Analyzer shared library for each
iacaLogicSNB.dll of the supported architectures.
iacaLogicIVB.dll
iacaArchDataNHM.dll
iacaArchDataWSM.dll Instruction databases for each of the supported
iacaArchDataSNB.dll architectures.
iacaArchDataIVB.dll
XED2NHM.dll
XED2WSM.dll XED2 shared libraries for each of the supported
XED2SNB.dll architectures.
XED2IVB.dll
Header file for the start/end markers.
iacaMarks.h
Place this file in another directory.
msvcp100.dll Microsoft Visual Studio* 2010 runtime redistributable
msvcr100.dll packages.

Intel® Architecture Code Analyzer User Guide 16

Release Contents

5.2 Linux* OS
Add the bin/ directory to the PATH environment variable.

Add the lib/ directory to the LD_LIBRARY_PATH environment variable.

Include include/iacaMarks.h in your code.

Filename Description
bin/iaca Intel Architecture Code Analyzer command-line tool
bin/iaca.sh Intel Architecture Code Analyzer invocation script
lib/libiacaLoader.so Intel Architecture Code Analyzer shared objects
lib/libiacaLogicNHM.so
lib/libiacaLogicWSM.so Intel Architecture Code Analyzer shared objects for each
lib/libiacaLogicSNB.so of the supported architectures
lib/libiacaLogicIVB.so
lib/libiacaArchDataNHM.so
lib/libiacaArchDataWSM.so Instruction databases for each of the supported
lib/libiacaArchDataSNB.so architectures
lib/libiacaArchDataIVB.so
lib/libXED2NHM.so
lib/libXED2WSM.so XED2 shared objects for each of the supported
lib/libXED2SNB.so architectures
lib/libXED2IVB.so
include/iacaMarks.h Header file for the start/end markers

Intel® Architecture Code Analyzer User Guide 17

Release Contents

5.3 Mac OS X*
Add the bin/ directory to the PATH environment variable.

Add the lib/ directory to the DYLD_LIBRARY_PATH environment variable.

Include include/iacaMarks.h in your code.

Filename Description
bin/iaca Intel Architecture Code Analyzer command-line tool
bin/iaca.sh Intel Architecture Code Analyzer invocation script
lib/libiacaLoader.so Intel Architecture Code Analyzer shared objects
lib/libiacaLogicNHM.so
lib/libiacaLogicWSM.so Intel Architecture Code Analyzer shared objects for each
lib/libiacaLogicSNB.so of the supported architectures
lib/libiacaLogicIVB.so
lib/libiacaArchDataNHM.so
lib/libiacaArchDataWSM.so Instruction databases for each of the supported
lib/libiacaArchDataSNB.so architectures
lib/libiacaArchDataIVB.so
lib/libXED2NHM.so
lib/libXED2WSM.so XED2 shared objects for each of the supported
lib/libXED2SNB.so architectures
lib/libXED2IVB.so
include/iacaMarks.h The header file for the start/end markers

Intel® Architecture Code Analyzer User Guide 18