Skip to content

Commit 69b62ac

Browse files
PRMerger9PRMerger9
authored
Merge pull request #3407 from corob-msft/docs/corob/cpp-docs-2467-2780
Update to add 1204 reason code
2 parents 85df9f8 + f9c591c commit 69b62ac

1 file changed

Lines changed: 92 additions & 76 deletions

File tree

‎docs/error-messages/tool-errors/vectorizer-and-parallelizer-messages.md‎

Lines changed: 92 additions & 76 deletions
Original file line numberDiff line numberDiff line change
@@ -1,41 +1,40 @@
11
---
22
description: "Learn more about: Vectorizer and parallelizer messages"
33
title: "Vectorizer and parallelizer messages"
4-
ms.date: "04/17/2019"
4+
ms.date: 02/16/2021
55
f1_keywords: ["C5011", "C5002", "C5021", "C5001", "C5012"]
6-
ms.assetid: d8f4844a-f414-42ab-b9a5-925a5da9d365
76
---
87
# Vectorizer and parallelizer messages
98

10-
You can use the Microsoft C++ compiler options [/Qpar-report](../../build/reference/qpar-report-auto-parallelizer-reporting-level.md) and [/Qvec-report](../../build/reference/qvec-report-auto-vectorizer-reporting-level.md) to set the [Auto-parallelization and auto-vectorization](../../parallel/auto-parallelization-and-auto-vectorization.md) to output reason codes and informational messages about its activity. This article explains the reason codes and the messages.
9+
You can use the Microsoft C++ compiler options [`/Qpar-report`](../../build/reference/qpar-report-auto-parallelizer-reporting-level.md) and [`/Qvec-report`](../../build/reference/qvec-report-auto-vectorizer-reporting-level.md) to set the [Auto-parallelization and auto-vectorization](../../parallel/auto-parallelization-and-auto-vectorization.md) to output reason codes and informational messages about its activity. This article explains the reason codes and the messages.
1110

1211
## <a name="BKMK_InformationalMessages"></a> Informational messages
1312

1413
Depending on the reporting level that you specify, one of the following informational messages appears for each loop.
1514

1615
For information about reason codes, refer to the next part of this article.
1716

18-
|Informational Message|Description|
19-
|---------------------------|-----------------|
20-
|5001|`Loop vectorized.`|
21-
|5002|`Loop not vectorized due to reason '*description*'.`|
22-
|5011|`Loop parallelized.`|
23-
|5012|`Loop not parallelized due to reason '*description*'.`|
24-
|5021|`Unable to associate loop with pragma.`|
17+
| Informational message | Description |
18+
|--|--|
19+
| 5001 | Loop vectorized. |
20+
| 5002 | Loop not vectorized due to reason '*description*'. |
21+
| 5011 | Loop parallelized. |
22+
| 5012 | Loop not parallelized due to reason '*description*'. |
23+
| 5021 | Unable to associate loop with pragma. |
2524

2625
The following sections list possible reason codes for the parallelizer and vectorizer.
2726

2827
## <a name="BKMK_ReasonCode50x"></a> 5xx reason codes
2928

3029
The 5*xx* reason codes apply to both the parallelizer and the vectorizer.
3130

32-
|Reason Code|Explanation|
33-
|-----------------|-----------------|
34-
|500|A generic message that covers several cases—for example, the loop includes multiple exits, or the loop header doesn't end by incrementing the induction variable.|
35-
|501|`Induction variable is not local; or upper bound is not loop-invariant.`|
36-
|502|`Induction variable is stepped in some manner other than a simple +1.`|
37-
|503|`Loop includes exception-handling or switch statements.`|
38-
|504|`Loop body may throw an exception that requires destruction of a C++ object.`|
31+
| Reason code | Explanation |
32+
|--|--|
33+
| 500 | A generic message that covers several cases—for example, the loop includes multiple exits, or the loop header doesn't end by incrementing the induction variable. |
34+
| 501 | Induction variable isn't local; or upper bound isn't loop-invariant. |
35+
| 502 | Induction variable is stepped in some manner other than a simple +1. |
36+
| 503 | Loop includes exception-handling or switch statements. |
37+
| 504 | Loop body may throw an exception that requires destruction of a C++ object. |
3938

4039
```cpp
4140
void code_500(int *A)
@@ -192,19 +191,19 @@ void code_504(int *A) {
192191
193192
The 10*xx* reason codes apply to the parallelizer.
194193
195-
|Reason Code|Explanation|
196-
|-----------------|-----------------|
197-
|1000|`The compiler detected a data dependency in the loop body.`|
198-
|1001|`The compiler detected a store to a scalar variable in the loop body, and that scalar has a use beyond the loop.`|
199-
|1002|`The compiler tried to parallelize a loop that has an inner loop that was already parallelized.`|
200-
|1003|`The loop body contains an intrinsic call that may read or write to memory.`|
201-
|1004|`There is a scalar reduction in the loop body. Scalar reduction can occur if the loop has been vectorized.`|
202-
|1005|`The no_parallel pragma was specified.`|
203-
|1006|`This function contains openmp. Resolve this by removing any openmp in this function.`|
204-
|1007|`The loop induction variable or the loop bounds are not signed 32-bit numbers (int or long). Resolve this by changing the type of the induction variable.`|
205-
|1008|`The compiler detected that this loop does not perform enough work to warrant auto-parallelization.`|
206-
|1009|`The compiler detected an attempt to parallelize a "do-while" loop. The auto-parallelizer only targets "for" loops.`|
207-
|1010|`The compiler detected that the loop is using "not-equals" (!=) for its condition.`|
194+
| Reason code | Explanation |
195+
|--|--|
196+
| 1000 | The compiler detected a data dependency in the loop body. |
197+
| 1001 | The compiler detected a store to a scalar variable in the loop body, and that scalar has a use beyond the loop. |
198+
| 1002 | The compiler tried to parallelize a loop that has an inner loop that was already parallelized. |
199+
| 1003 | The loop body contains an intrinsic call that may read or write to memory. |
200+
| 1004 | There's a scalar reduction in the loop body. Scalar reduction can occur if the loop has been vectorized. |
201+
| 1005 | The `no_parallel` pragma was specified. |
202+
| 1006 | This function contains OpenMP. Resolve it by removing any OpenMP in this function. |
203+
| 1007 | The loop induction variable or the loop bounds aren't signed 32-bit numbers (`int` or `long`). Resolve it by changing the type of the induction variable. |
204+
| 1008 | The compiler detected that this loop doesn't perform enough work to justify auto-parallelization. |
205+
| 1009 | The compiler detected an attempt to parallelize a "`do`-`while`" loop. The auto-parallelizer only targets "`for`" loops. |
206+
| 1010 | The compiler detected that the loop is using "not-equals" (`!=`) for its condition. |
208207
209208
```cpp
210209
int A[1000];
@@ -401,15 +400,15 @@ void code_1010()
401400

402401
The 11*xx* reason codes apply to the vectorizer.
403402

404-
|Reason Code|Explanation|
405-
|-----------------|-----------------|
406-
|1100|`Loop contains control flow—for example, "if" or "?".`|
407-
|1101|`Loop contains datatype conversion—perhaps implicit—that cannot be vectorized.`|
408-
|1102|`Loop contains non-arithmetic or other non-vectorizable operations.`|
409-
|1103|`Loop body includes shift operations whose size might vary within the loop.`|
410-
|1104|`Loop body includes scalar variables.`|
411-
|1105|`Loop includes a unrecognized reduction operation.`|
412-
|1106|`Outer loop not vectorized.`|
403+
| Reason code | Explanation |
404+
|--|--|
405+
| 1100 | Loop contains control flow—for example, "`if`" or "`?:`". |
406+
| 1101 | Loop contains datatype conversion—perhaps implicit—that can't be vectorized. |
407+
| 1102 | Loop contains non-arithmetic or other non-vectorizable operations. |
408+
| 1103 | Loop body includes shift operations whose size might vary within the loop. |
409+
| 1104 | Loop body includes scalar variables. |
410+
| 1105 | Loop includes an unrecognized reduction operation. |
411+
| 1106 | Outer loop not vectorized. |
413412

414413
```cpp
415414
void code_1100(int *A, int x)
@@ -549,12 +548,13 @@ void code_1106(int *A)
549548
550549
The 12*xx* reason codes apply to the vectorizer.
551550
552-
|Reason Code|Explanation|
553-
|-----------------|-----------------|
554-
|1200|`Loop contains loop-carried data dependences that prevent vectorization. Different iterations of the loop interfere with each other such that vectorizing the loop would produce wrong answers, and the auto-vectorizer cannot prove to itself that there are no such data dependences.`|
555-
|1201|`Array base changes during the loop.`|
556-
|1202|`Field in a struct is not 32 or 64 bits wide.`|
557-
|1203|`Loop body includes non-contiguous accesses into an array.`|
551+
| Reason code | Explanation |
552+
|--|--|
553+
| 1200 | Loop contains loop-carried data dependencies that prevent vectorization. Different iterations of the loop interfere with each other such that vectorizing the loop would produce wrong answers, and the auto-vectorizer can't prove to itself that there are no such data dependences. |
554+
| 1201 | Array base changes during the loop. |
555+
| 1202 | Field in a struct isn't 32 or 64 bits wide. |
556+
| 1203 | Loop body includes non-contiguous accesses into an array. |
557+
| 1204 | Compiler internal data structure limit hit: too many data dependence edges. |
558558
559559
```cpp
560560
void fn();
@@ -618,20 +618,36 @@ void code_1203(int *A)
618618
A[i] += A[i*2+2] + 2; // non-contiguous memory access not vectorized
619619
}
620620
}
621+
622+
void code_1204(int *A)
623+
{
624+
// Code 1204 is emitted when internal compiler data structures
625+
// hit a limit on the number of data dependence edges recorded.
626+
// Resolve this by moving the innermost loop to another function.
627+
628+
for (int i=0; i<1000; i++)
629+
for (int j=0; j<1000; j++)
630+
for (int k=0; k<1000; k++)
631+
for (int l=0; l<1000; l++)
632+
{
633+
for (int m=0; m<1000; m++)
634+
A[m] = A[m+i] + A[m+j] + A[m+k] + A[m+l];
635+
}
636+
}
621637
```
622638

623639
## <a name="BKMK_ReasonCode130x"></a> 13xx reason codes
624640

625641
The 13*xx* reason codes apply to the vectorizer.
626642

627-
|Reason Code|Explanation|
628-
|-----------------|-----------------|
629-
|1300|`Loop body contains no—or very little—computation.`|
630-
|1301|`Loop stride is not +1.`|
631-
|1302|`Loop is a "do-while".`|
632-
|1303|`Too few loop iterations for vectorization to provide value.`|
633-
|1304|`Loop includes assignments that are of different sizes.`|
634-
|1305|`Not enough type information.`|
643+
| Reason code | Explanation |
644+
|--|--|
645+
| 1300 | Loop body contains no—or very little—computation. |
646+
| 1301 | Loop stride isn't +1. |
647+
| 1302 | Loop is a "`do`-`while`". |
648+
| 1303 | Too few loop iterations for vectorization to provide value. |
649+
| 1304 | Loop includes assignments that are of different sizes. |
650+
| 1305 | Not enough type information. |
635651

636652
```cpp
637653
void code_1300(int *A, int *B)
@@ -756,14 +772,14 @@ void code_1305( S_1305 *s, S_1305 x)
756772
757773
The 14*xx* reason codes occur when some option that is incompatible with vectorization is specified.
758774
759-
|Reason Code|Explanation|
760-
|-----------------|-----------------|
761-
|1400|`#pragma loop(no_vector) is specified.`|
762-
|1401|`/kernel switch is specified when targeting x86 or ARM.`|
763-
|1402|`/arch:SSE2 or higher switch is not specified when targeting x86.`|
764-
|1403|`/arch:ATOM switch is specified and the loop includes operations on doubles.`|
765-
|1404|`/O1 or /Os switch is specified.`|
766-
|1405|`Vectorization is disabled to aid in dynamic-initializer-to-static-initializer optimization.`|
775+
| Reason code | Explanation |
776+
|--|--|
777+
| 1400 | `#pragma loop(no_vector)` is specified. |
778+
| 1401 | `/kernel` switch is specified when targeting x86 or ARM. |
779+
| 1402 | `/arch:SSE2` or higher switch isn't specified when targeting x86. |
780+
| 1403 | `/arch:ATOM` switch is specified and the loop includes operations on doubles. |
781+
| 1404 | `/O1` or `/Os` switch is specified. |
782+
| 1405 | Vectorization is disabled to aid in dynamic-initializer-to-static-initializer optimization. |
767783
768784
```cpp
769785
void code_1400(int *A)
@@ -828,14 +844,14 @@ void code_1404(int *A)
828844

829845
The 15*xx* reason codes apply to aliasing. Aliasing occurs when a location in memory can be accessed by two different names.
830846

831-
|Reason Code|Explanation|
832-
|-----------------|-----------------|
833-
|1500|`Possible aliasing on multi-dimensional arrays.`|
834-
|1501|`Possible aliasing on arrays-of-structs.`|
835-
|1502|`Possible aliasing and array index is other than n + K.`|
836-
|1503|`Possible aliasing and array index has multiple offsets.`|
837-
|1504|`Possible aliasing; would require too many runtime checks.`|
838-
|1505|`Possible aliasing, but runtime checks are too complex.`|
847+
| Reason code | Explanation |
848+
|--|--|
849+
| 1500 | Possible aliasing on multi-dimensional arrays. |
850+
| 1501 | Possible aliasing on arrays-of-structs. |
851+
| 1502 | Possible aliasing and array index is other than n + K. |
852+
| 1503 | Possible aliasing and array index has multiple offsets. |
853+
| 1504 | Possible aliasing; would require too many runtime checks. |
854+
| 1505 | Possible aliasing, but runtime checks are too complex. |
839855

840856
```cpp
841857
void code_1500(int A[100][100], int B[100][100])
@@ -954,10 +970,10 @@ void code_1505(int *A, int *B)
954970
955971
## See also
956972
957-
[C/C++ Compiler and build tools errors and warnings](../compiler-errors-1/c-cpp-build-errors.md)
958-
[Auto-parallelization and auto-vectorization](../../parallel/auto-parallelization-and-auto-vectorization.md) \
959-
[Auto-Vectorizer in Visual Studio 2012 – Overview](/archive/blogs/nativeconcurrency/auto-vectorizer-in-visual-studio-2012-overview) \
960-
[#pragma loop()](../../preprocessor/loop.md) \
961-
[/Q Options (Low-Level Operations)](../../build/reference/q-options-low-level-operations.md) \
962-
[/Qpar-report (Auto-Parallelizer Reporting Level)](../../build/reference/qpar-report-auto-parallelizer-reporting-level.md) \
963-
[/Qvec-report (Auto-Vectorizer Reporting Level)](../../build/reference/qvec-report-auto-vectorizer-reporting-level.md)
973+
[C/C++ Compiler and build tools errors and warnings](../compiler-errors-1/c-cpp-build-errors.md)\
974+
[Auto-parallelization and auto-vectorization](../../parallel/auto-parallelization-and-auto-vectorization.md)\
975+
[Auto-Vectorizer in Visual Studio 2012 – Overview](/archive/blogs/nativeconcurrency/auto-vectorizer-in-visual-studio-2012-overview)\
976+
[`#pragma loop()`](../../preprocessor/loop.md)\
977+
[`/Q` Options (Low-Level Operations)](../../build/reference/q-options-low-level-operations.md)\
978+
[`/Qpar-report` (Auto-Parallelizer Reporting Level)](../../build/reference/qpar-report-auto-parallelizer-reporting-level.md)\
979+
[`/Qvec-report` (Auto-Vectorizer Reporting Level)](../../build/reference/qvec-report-auto-vectorizer-reporting-level.md)

0 commit comments

Comments
 (0)