selftests/resctrl: fix clang build warnings related to abs(), labs() calls
Subsystems (2)
PATCH 1/1 selftests/resctrl: fix clang build warnings related to abs(), labs() calls
First of all, in order to build with clang at all, one must first apply Valentin Obst's build fix for LLVM [1]. Furthermore, for this particular resctrl directory, my pending fix [2] must also be applied. Once those fixes are in place, then when building with clang, via:
make LLVM=1 -C tools/testing/selftests
...two types of warnings occur:
warning: absolute value function 'abs' given an argument of type 'long' but has parameter of type 'int' which may cause truncation of value
warning: taking the absolute value of unsigned type 'unsigned long' has no effect
Fix these by:
a) using labs() in place of abs(), when long integers are involved, and
b) don't call labs() unnecessarily.
[1] https://lore.kernel.org/all/20240329-selftests-libmk-llvm-rfc-v1-1-2f9ed7d1c49f@valentinobst.de/
[2] https://lore.kernel.org/all/[email protected]/
--- tools/testing/selftests/resctrl/cmt_test.c | 4 ++-- tools/testing/selftests/resctrl/mba_test.c | 2 +- tools/testing/selftests/resctrl/mbm_test.c | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)
tools/testing/selftests/resctrl/cmt_test.c +2 -2
@@ -40,11 +40,11 @@ static int show_results_info(unsigned long sum_llc_val, int no_of_bits,int ret; avg_llc_val = sum_llc_val / num_of_runs; - avg_diff = (long )abs (cache_span - avg_llc_val); + avg_diff = (long )(cache_span - avg_llc_val); diff_percent = ((float )cache_span - avg_llc_val) / cache_span *100 ; ret = platform &&abs ((int )diff_percent) > max_diff_percent && -abs (avg_diff) > max_diff; +labs (avg_diff) > max_diff;ksft_print_msg ("%s Check cache miss rate within %lu%%\n" , ret ?"Fail:" :"Pass:" , max_diff_percent);
tools/testing/selftests/resctrl/mba_test.c +1 -1
@@ -77,7 +77,7 @@ static bool show_mba_info(unsigned long *bw_imc, unsigned long *bw_resc) avg_bw_imc = sum_bw_imc / (NUM_OF_RUNS -1 ); avg_bw_resc = sum_bw_resc / (NUM_OF_RUNS -1 ); - avg_diff = (float )labs (avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float )(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int )(avg_diff *100 ); ksft_print_msg("%s Check MBA diff within %d%% for schemata %u\n",
tools/testing/selftests/resctrl/mbm_test.c +1 -1
@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc /4 ; avg_bw_resc = sum_bw_resc /4 ; - avg_diff = (float )labs (avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float )(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int )(avg_diff *100 ); ret = avg_diff_per >MAX_DIFF_PERCENT ;
base-commit: f03359bca01bf4372cf2c118cd9a987a5951b1c8
prerequisite-patch-id: b901ece2a5b78503e2fb5480f20e304d36a0ea27
prerequisite-patch-id: 8d96c4b8c3ed6d9ea2588ef7f594ae0f9f83c279
Signature · 1 line
2.45.0
Quoted 40 lines · John Hubbard
First of all, in order to build with clang at all, one must first apply Valentin Obst's build fix for LLVM [1]. Furthermore, for this particular resctrl directory, my pending fix [2] must also be applied. Once those fixes are in place, then when building with clang, via:
make LLVM=1 -C
tools/testing/selftests..two types of warnings occur:
warning: absolute value function 'abs' given an argument of type 'long' but has parameter of type 'int' which may cause truncation of value
warning: taking the absolute value of unsigned type 'unsigned long' has no effect
Fix these by:
a) using
labs()in place ofabs(), when long integers are involved, andb) don't call
labs()unnecessarily.[1] https://lore.kernel.org/all/20240329-selftests-libmk-llvm-rfc-v1-1-2f9ed7d1c49f@valentinobst.de/
[2] https://lore.kernel.org/all/[email protected]/Signed-off-by John Hubbard <[email protected]>--- tools/testing/selftests/resctrl/cmt_test.c | 4 ++-- tools/testing/selftests/resctrl/mba_test.c | 2 +- tools/testing/selftests/resctrl/mbm_test.c | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)tools/testing/selftests/resctrl/cmt_test.c +0
-0@@ -40,11 +40,11 @@ static int show_results_info(unsigned long sum_llc_val, int no_of_bits,int ret; avg_llc_val = sum_llc_val / num_of_runs;
tools/testing/selftests/resctrl/cmt_test.c static int show_results_info(unsigned long sum_llc_val, int no_of_bits, · line 50
Quoted 10 lines
tools/testing/selftests/resctrl/cmt_test.c +2
-2- avg_diff = (long )abs (cache_span - avg_llc_val); + avg_diff = (long )(cache_span - avg_llc_val); diff_percent = ((float )cache_span - avg_llc_val) / cache_span *100 ; ret = platform &&abs ((int )diff_percent) > max_diff_percent && -abs (avg_diff) > max_diff; +labs (avg_diff) > max_diff;ksft_print_msg ("%s Check cache miss rate within %lu%%\n" , ret ?"Fail:" :"Pass:" , max_diff_percent);
This seems fine but...
Quoted 18 lines
tools/testing/selftests/resctrl/mba_test.c +1
-1@@ -77,7 +77,7 @@ static bool show_mba_info(unsigned long *bw_imc, unsigned long *bw_resc) avg_bw_imc = sum_bw_imc / (NUM_OF_RUNS -1 ); avg_bw_resc = sum_bw_resc / (NUM_OF_RUNS -1 ); - avg_diff = (float )labs (avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float )(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int )(avg_diff *100 ); ksft_print_msg("%s Check MBA diff within %d%% for schemata %u\n",tools/testing/selftests/resctrl/mbm_test.c +0
-0
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) · line 39
Quoted 8 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc /4 ; avg_bw_resc = sum_bw_resc /4 ; - avg_diff = (float )labs (avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float )(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int )(avg_diff *100 ); ret = avg_diff_per >MAX_DIFF_PERCENT ;
But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Signature · 1 line
i.
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 12 lines
Quoted 12 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc /4 ; avg_bw_resc = sum_bw_resc /4 ; - avg_diff = (float )labs (avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float )(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int )(avg_diff *100 );
Quoted 4 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) · line 39
Quoted 1 line
tools/testing/selftests/resctrl/mbm_test.c +0
-0ret = avg_diff_per >MAX_DIFF_PERCENT ;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
thanks,
Signature · 2 lines
John Hubbard NVIDIA
Reinette Chatre
maintainerQuoted 18 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 6 lines
Quoted 6 lines
tools/testing/selftests/resctrl/mbm_test.c +0
-0@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc / 4;Quoted 8 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) · line 39
Quoted 5 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1avg_bw_resc = sum_bw_resc / 4; - avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int)(avg_diff * 100); ret = avg_diff_per > MAX_DIFF_PERCENT;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Quoted 3 lines
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
It does not seem as though clang can see when values have been casted. I tried to do so explicitly with a:
avg_diff = labs((long)avg_bw_resc - avg_bw_imc) / (float)avg_bw_imc;
But that still triggers:
warning: taking the absolute value of unsigned type 'unsigned long' has no effect [-Wabsolute-value]
Looks like we may need to be more explicit types and not rely on casting so much to make the compiler happy.
Reinette
Quoted 19 lines · Reinette Chatre
Quoted 18 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 14 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) · line 37
Quoted 11 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc / 4; avg_bw_resc = sum_bw_resc / 4; - avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int)(avg_diff * 100); ret = avg_diff_per > MAX_DIFF_PERCENT;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Quoted 7 lines
Quoted 3 lines
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
It does not seem as though clang can see when values have been casted. I tried to do so explicitly with a:
avg_diff=labs((long)avg_bw_resc - avg_bw_imc)/ (float)avg_bw_imc;
The subtraction result will get promoted to an unsigned long, before being passed into labs(3).
Quoted 3 lines
But that still triggers:
warning: taking the absolute value of unsigned type 'unsigned long' has no effect [-Wabsolute-value]
As expected, yes.
Quoted 4 lines
Looks like we may need to be more explicit types and not rely on casting so much to make the compiler happy.
I assumed that this code did not expect to handle negative numbers, because it is using unsigned math throughout.
If you do expect it to handle cases where, for example, this happens:
avg_bw_imc > avg_bw_resc
...then a proper solution is easy, and looks like this:
tools/testing/selftests/resctrl/mbm_test.c +2 -2
@@ -17,8 +17,8 @@static int show_bw_info (unsigned long *bw_imc,unsigned long *bw_resc,size_t span) { -unsigned long avg_bw_imc =0 , avg_bw_resc =0 ; -unsigned long sum_bw_imc =0 , sum_bw_resc =0 ; +long avg_bw_imc =0 , avg_bw_resc =0 ; +long sum_bw_imc =0 , sum_bw_resc =0 ;int runs, ret, avg_diff_per;float avg_diff =0 ;
Should I resend the patch with that approach?
thanks,
Signature · 2 lines
John Hubbard NVIDIA
Reinette Chatre
maintainerHi John,
Quoted 42 lines · John Hubbard
Quoted 19 lines · Reinette Chatre
Quoted 18 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 14 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) · line 37
Quoted 11 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc / 4; avg_bw_resc = sum_bw_resc / 4; - avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int)(avg_diff * 100); ret = avg_diff_per > MAX_DIFF_PERCENT;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Quoted 7 lines
Quoted 3 lines
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
It does not seem as though clang can see when values have been casted. I tried to do so explicitly with a:
avg_diff=labs((long)avg_bw_resc - avg_bw_imc)/ (float)avg_bw_imc;The subtraction result will get promoted to an unsigned long, before being passed into labs(3).
Quoted 3 lines
But that still triggers:
warning: taking the absolute value of unsigned type 'unsigned long' has no effect [-Wabsolute-value]As expected, yes.
Quoted 4 lines
Looks like we may need to be more explicit types and not rely on casting so much to make the compiler happy.
Quoted 6 lines
I assumed that this code did not expect to handle negative numbers, because it is using unsigned math throughout.
If you do expect it to handle cases where, for example, this happens:
avg_bw_imc>avg_bw_resc
The existing code seems to handle this ok. A sample program with this scenario comparing existing computation with your first proposal is below:
#include <stdio.h>
#include <stdlib.h>
void main(void) {
unsigned long avg_bw_resc = 20000;
unsigned long avg_bw_imc = 40000;
float avg_diff;
/* Existing code */
avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc;
printf("Existing code: avg_diff = %f\n", avg_diff);
/* Original proposed fix */
avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc;
printf("Original proposed fix: avg_diff = %f\n", avg_diff);
}output:
Existing code: avg_diff = 0.500000
Original proposed fix: avg_diff = 461168590192640.000000
Quoted 18 lines
...then a proper solution is easy, and looks like this:
tools/testing/selftests/resctrl/mbm_test.c +0
-0@@ -17,8 +17,8 @@static int
show_bw_info(unsigned long*bw_imc, unsigned long*bw_resc,size_tspan) {
- unsigned longavg_bw_imc= 0,avg_bw_resc= 0;
- unsigned longsum_bw_imc= 0,sum_bw_resc= 0;
+ longavg_bw_imc= 0,avg_bw_resc= 0;
+ longsum_bw_imc= 0,sum_bw_resc= 0;
int runs, ret,avg_diff_per;
floatavg_diff= 0;
Quoted 1 line
Should I resend the patch with that approach?
ok. That indeed makes the computations easier to understand. I assume you intend to fix the snippet in mba_test.c also?
Reinette
Quoted 7 lines · Reinette Chatre
Hi John,
Quoted 5 lines · John Hubbard
Quoted 4 lines · Reinette Chatre
Quoted 3 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
...
Quoted 28 lines
Quoted 6 lines
I assumed that this code did not expect to handle negative numbers, because it is using unsigned math throughout.
If you do expect it to handle cases where, for example, this happens:
avg_bw_imc>avg_bw_rescThe existing code seems to handle this ok. A sample program with this scenario comparing existing computation with your first proposal is below:
#include <
stdio.h>
#include <stdlib.h>void main(void) { unsigned long avg_bw_resc = 20000; unsigned long avg_bw_imc = 40000; float avg_diff; /* Existing code */ avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; printf("Existing code: avg_diff = %f\n", avg_diff); /* Original proposed fix */ avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; printf("Original proposed fix: avg_diff = %f\n", avg_diff); }
Quoted 3 lines
output:
Existing code:avg_diff= 0.500000
Original proposed fix:avg_diff= 461168590192640.000000
That seems "a little bit" wrong. haha :)
Quoted 19 lines
Quoted 19 lines
...then a proper solution is easy, and looks like this:
tools/testing/selftests/resctrl/mbm_test.c +2
-2@@ -17,8 +17,8 @@ static int show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) { - unsigned long avg_bw_imc = 0, avg_bw_resc = 0; - unsigned long sum_bw_imc = 0, sum_bw_resc = 0; + long avg_bw_imc = 0, avg_bw_resc = 0; + long sum_bw_imc = 0, sum_bw_resc = 0; int runs, ret, avg_diff_per; float avg_diff = 0;
Quoted 5 lines
Quoted 1 line
Should I resend the patch with that approach?
ok. That indeed makes the computations easier to understand. I assume you intend to fix the snippet in
mba_test.calso?
Yes, will do that. Thanks for spotting the bug in the original "fix"!
thanks,
Signature · 2 lines
John Hubbard NVIDIA
Quoted 22 lines · John Hubbard
Quoted 21 lines · Reinette Chatre
Quoted 20 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 16 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long · line 38
Quoted 13 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc / 4; avg_bw_resc = sum_bw_resc / 4; - avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int)(avg_diff * 100); ret = avg_diff_per > MAX_DIFF_PERCENT;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
(I see there's a better patch posted already but since there are a few incorrect claims in this discussion, I'll do for the record type of reply.)
This discussion now went to a tangent about the warning. My main point is that logic is not correct after removing labs().
I also disagree with the claim that using labs() on unsigned value is no-op because labs() takes long so unsigned is just forced into signed when calling which is why the warning triggers but it's very misleading warning (see below).
Quoted 4 lines
Quoted 3 lines
It does not seem as though clang can see when values have been casted. I tried to do so explicitly with a:
avg_diff=labs((long)avg_bw_resc - avg_bw_imc)/ (float)avg_bw_imc;
Quoted 8 lines
The subtraction result will get promoted to an unsigned long, before being passed into labs(3).
Quoted 3 lines
But that still triggers:
warning: taking the absolute value of unsigned type 'unsigned long' has no effect [-Wabsolute-value]As expected, yes.
That error message isn't factually correct:
unsigned long a = LONG_MAX;
long b;
a += 2;
b = (long)a;
printf("%llu %lli %lli\n", a, b, labs(a));Prints (at least when built with gcc):
9223372036854775809 -9223372036854775807 9223372036854775807
labs(LONG_MAX + 1) won't work though since it's not positively presentable with long and the value is left untouched.
Signature · 1 line
i.
Quoted 32 lines · Ilpo Järvinen
Quoted 22 lines · John Hubbard
Quoted 21 lines · Reinette Chatre
Quoted 20 lines · John Hubbard
Quoted 1 line · Ilpo Järvinen
On Thu, 2 May 2024, John Hubbard wrote:
...
Quoted 16 lines
tools/testing/selftests/resctrl/mbm_test.c show_bw_info(unsigned long *bw_imc, unsigned long · line 38
Quoted 13 lines
tools/testing/selftests/resctrl/mbm_test.c +1
-1@@ -33,7 +33,7 @@ show_bw_info(unsigned long *bw_imc, unsigned long *bw_resc, size_t span) avg_bw_imc = sum_bw_imc / 4; avg_bw_resc = sum_bw_resc / 4; - avg_diff = (float)labs(avg_bw_resc - avg_bw_imc) / avg_bw_imc; + avg_diff = (float)(avg_bw_resc - avg_bw_imc) / avg_bw_imc; avg_diff_per = (int)(avg_diff * 100); ret = avg_diff_per > MAX_DIFF_PERCENT;But how are these two cases same after your change when you ended up removing taking the absolute value entirely?
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
All of the arguments are unsigned integers, so all arithmetic results are interpreted as unsigned, so taking the absolute value of that is always a no-op.
(I see there's a better patch posted already but since there are a few incorrect claims in this discussion, I'll do for the record type of reply.)
Quoted 8 lines
This discussion now went to a tangent about the warning. My main point is that logic is not correct after removing
labs().I also disagree with the claim that using
labs()on unsigned value is no-op becauselabs()takes long so unsigned is just forced into signed when calling which is why the warning triggers but it's very misleading warning (see below).
Yes you are correct.
Quoted 22 lines
Quoted 4 lines
Quoted 3 lines
It does not seem as though clang can see when values have been casted. I tried to do so explicitly with a:
avg_diff = labs((long)avg_bw_resc - avg_bw_imc) / (float)avg_bw_imc;Quoted 8 lines
The subtraction result will get promoted to an unsigned long, before being passed into labs(3).
Quoted 3 lines
But that still triggers:
warning: taking the absolute value of unsigned type 'unsigned long' has no effect [-Wabsolute-value]As expected, yes.
That error message isn't factually correct:
unsigned long a = LONG_MAX; long b; a += 2; b = (long)a; printf("%llu %lli %lli\n", a, b, labs(a));
Quoted 7 lines
Prints (at least when built with gcc):
9223372036854775809 -9223372036854775807 9223372036854775807
labs(LONG_MAX + 1)won't work though since it's not positively presentable with long and the value is left untouched.
Thanks for setting the detailed record straight! :)
thanks,
Signature · 2 lines
John Hubbard NVIDIA