Integer constant folding in the presence of new primops

Jan Stolarek

16 May 2013 16 May '13

7:32 a.m.

I'm finishing my patches for T6135, but I still have some work to do on integer-gmp and integer-simpl libraries. I changed the previously existing functions that compare Integers: {-# NOINLINE #-} eqInteger :: Integer -> Integer -> Bool eqInteger = ... to something like this: {-# NOINLINE eqIntegerPrim #-} eqIntegerPrim :: Integer -> Integer -> Int# eqIntegerPrim ... -- same as before, but uses new primops eqInteger :: Integer -> Integer -> Bool eqInteger a b = tagToEnum# (a `eqIntegerPrim` b) I noticed that in some cases this implementation prevents constant folding for Integers. I'm trying to figure out the proper way to fix this. I am considering two ways of doing it: - add new folding rules for eqIntegerPrim and other functions, leave the existing rules for eqInteger. This should fold: (100012 :: Integer) == 100012 to True (as it does now) - add new folding rules for eqIntegerPrim and other functions, remove the existing rules for eqInteger. This would fold (100012 :: Integer) == 100012 to tagToEnum# 1# Which approach is better? Another question related to this: how do I run validation with integer-simpl enabled? Wiki describes how to enable integer-simpl for the build, but I think that validation is always run against integer-gmp? Or am I wrong here? Janek

Show replies by date

Simon Peyton-Jones

16 May 16 May

noon

All the same issues arise for Int# too, right? | {-# NOINLINE eqIntegerPrim #-} | eqIntegerPrim :: Integer -> Integer -> Int# | eqIntegerPrim ... -- same as before, but uses new primops | | eqInteger :: Integer -> Integer -> Bool | eqInteger a b = tagToEnum# (a `eqIntegerPrim` b) | | I noticed that in some cases this implementation prevents constant | folding for Integers. Why? Because eqInteger is now inlined, so the rule doesn't get a chance to fire? | A add new folding rules for eqIntegerPrim and other functions, leave the | existing rules for | eqInteger. This should fold: (100012 :: Integer) == 100012 to True (as | it does now) | | B add new folding rules for eqIntegerPrim and other functions, remove | the existing rules for | eqInteger. This would fold (100012 :: Integer) == 100012 to tagToEnum# | 1# I'm not sure I understand, but I think that the difference between (A) and (B) is whether the existing rules for eqInteger remain or are removed. I'd remove them if you can; simpler, only one way for things to happen S

Jan Stolarek

17 May 17 May

5:49 a.m.

...

All the same issues arise for Int# too, right? I guess you're right :) I removed the rules for wrappers (eqInteger) and created rules for "primOps" (eqIntegerPrim, which I think I'll rename to eqInteger#)

...

Why? Because eqInteger is now inlined, so the rule doesn't get a chance to fire? There's something very subtle going on with inlining and I can't grasp what it is. At the moment I have this in prelude/PrelNames.lhs:

eqIntegerPrimIdKey = mkPreludeMiscIdUnique 70 eqIntegerPrimName = varQual gHC_INTEGER_TYPE (fsLit "eqIntegerPrim") eqIntegerPrimIdKey and in prelude/PrelRules.lhs I have this: builtinIntegerRules = [ ... rule_binop_Prim "eqIntegerPrim" eqIntegerPrimName (==), ...] where rule_binop_Prim str name op = BuiltinRule { ru_name = fsLit str, ru_fn = name, ru_nargs = 2, ru_try = match_Integer_binop_Prim op } match_Integer_binop_Prim binop dflags _ id_unf [xl, yl] | Just (LitInteger x _) <- exprIsLiteral_maybe id_unf xl , Just (LitInteger y _) <- exprIsLiteral_maybe id_unf yl = Just (if x `binop` y then trueValInt dflags else falseValInt dflags) match_Integer_binop_Prim _ _ _ _ _ = Nothing My understanding is that this rule will fire if it finds a function "eqIntegerPrim" with two Integer literals as parameters. Given my definitions in integer-gmp library: {-# NOINLINE eqIntegerPrim #-} eqIntegerPrim :: Integer -> Integer -> Int# eqIntegerPrim ... -- same as before, but uses new primops {-# INLINE eqInteger #-} eqInteger :: Integer -> Integer -> Bool eqInteger a b = tagToEnum# (a `eqIntegerPrim` b) my understanding is that in expression: eqIntegerE = (100012 :: Integer) == 100012 the (==) gets inlined to eqInteger: eqIntegerE = (100012 :: Integer) `eqInteger` 100012 which in turn gets inlined to eqIntegerPrim: eqIntegerE = tagToEnum# ((100012 :: Integer) `eqIntegerPrim` 100012) At this point inling stops (because of NOINLINE) and my rule fires giving: eqIntegerE = tagToEnum# 1# which in turns allows the rule for tagToEnum# to fire, giving: eqIntegerE = GHC.Prim.True Now here's the tricky part. I'm testing this with test lib/integer/integerConstantFolding in the testsuite and the test fails because rules for quotRemInteger, divModInteger, quotInteger and remInteger don't fire, leaving the constants unfolded. I noticed that if I mark eqInteger with NOINLINE, then these rules fire, but then obviously comparisons like (100012 :: Integer) == 100012 don't get folded and the test fails anyway. I'm analyzing how the function quotInteger and others use eqInteger, but I don't see anything that would be obvious. Janek

Ian Lynagh

20 May 20 May

7:18 a.m.

On Fri, May 17, 2013 at 11:49:26AM +0200, Jan Stolarek wrote:

...

Now here's the tricky part. I'm testing this with test lib/integer/integerConstantFolding in the testsuite and the test fails because rules for quotRemInteger, divModInteger, quotInteger and remInteger don't fire, leaving the constants unfolded. I noticed that if I mark eqInteger with NOINLINE, then these rules fire, but then obviously comparisons like (100012 :: Integer) == 100012 don't get folded and the test fails anyway. I'm analyzing how the function quotInteger and others use eqInteger, but I don't see anything that would be obvious.

If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

...

eqIntegerPrimIdKey = mkPreludeMiscIdUnique 70 eqIntegerPrimName = varQual gHC_INTEGER_TYPE (fsLit "eqIntegerPrim") eqIntegerPrimIdKey

Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique? Thanks Ian -- Ian Lynagh, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

Jan Stolarek

7:34 a.m.

...

If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156 I noticed that when I define eqInteger wrapper to be NOINLINE, the call to quot is translated to Core as: Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156) but when I change the wrapper to INLINE I get: Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156) All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...

Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique? No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek

Simon Peyton-Jones

19 Jun 19 Jun

8:06 a.m.

Jan I'm conscious that we have not closed the loop on this thread, below. (I think the attached email is also relevant.) If it's not sorted out, can you open a ticket, put in the relevant info (so we don't need to look at the email trail), and we can tackle it when you get here. Simon -----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org] On Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...

If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

...

Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique? No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

Jan Stolarek

10:27 a.m.

...

If it's not sorted out, can you open a ticket, put in the relevant info (so we don't need to look at the email trail), and we can tackle it when you get here. Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly. Janek

...

Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org] On Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call to quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

Nicolas Frisby

11:43 a.m.

I'm also seeing performance regressions in the shootout benchmarks that I can't identify in the asm. The new asm looks better but performs worse, with a ~15% slowdown. I fired up the performance counters in my CPU and the free Intel code for inspecting them showed that my CPU utilization took about a 10% hit, even while executing fewer total instructions. 1) Jan, perhaps we're seeing the same sort of behavior — the shootout benchmarks have extremely hot loops (hundreds of millions of iterations IIRC). I used ticky profiling too, and saw no suspicious changes in any counters. 2) Dear Low-level Gurus: How feasible is it that a ~15% slowdown in a program with a very hot loop is due to incidentally inhibiting some caching behavior (instr? data?)? Or perhaps effecting alignment? FTR my CPU is a Core i7-2620M, Sandy Bridge. Thanks all. On Wed, Jun 19, 2013 at 9:27 AM, Jan Stolarek wrote:

...

...
If it's not sorted out, can you open a ticket, put in the relevant info (so we don't need to look at the email trail), and we can tackle it when you get here. Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly.

Janek

...
Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org]

On

...
Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call to quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

Austin Seipp

12:39 p.m.

I mean, it certainly *seems* reasonable a 15% hit could come from pipelining changes or cache behavior or something. I don't think alignment would really be a huge issue; post-Nehalem I believe non-aligned writes/reads are extremely cheap. Non-intuitive behavior can totally happen too: I've seen cases of adding instructions to a loop which speeds things up (e.g. by taking the extra step, you may mitigate a dependency stall, which massively helps pipelining across the loop body etc.) Nicolas, can I ask what benchmark you're looking at? And what performance tools are you using, Intels'? If you're on Linux, the 'perf' tool on a modern kernel can be used to quickly get an overview of how many cache misses/hits your process has, how many pipeline stalls occur, etc. You can then use it to drill down a bit into the assembly that's problematic. That might not give you an exact culprit (it could be many changes and accumulative hits,) but it's a start. On Wed, Jun 19, 2013 at 10:43 AM, Nicolas Frisby wrote:

...

I'm also seeing performance regressions in the shootout benchmarks that I can't identify in the asm. The new asm looks better but performs worse, with a ~15% slowdown.

I fired up the performance counters in my CPU and the free Intel code for inspecting them showed that my CPU utilization took about a 10% hit, even while executing fewer total instructions.

1) Jan, perhaps we're seeing the same sort of behavior — the shootout benchmarks have extremely hot loops (hundreds of millions of iterations IIRC). I used ticky profiling too, and saw no suspicious changes in any counters.

2) Dear Low-level Gurus: How feasible is it that a ~15% slowdown in a program with a very hot loop is due to incidentally inhibiting some caching behavior (instr? data?)? Or perhaps effecting alignment? FTR my CPU is a Core i7-2620M, Sandy Bridge.

Thanks all.

On Wed, Jun 19, 2013 at 9:27 AM, Jan Stolarek wrote:

...
...
If it's not sorted out, can you open a ticket, put in the relevant info (so we don't need to look at the email trail), and we can tackle it when you get here. Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly.

Janek

...
Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org] On Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call to quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

-- Regards, Austin - PGP: 4096R/0x91384671

Nicolas Frisby

12:46 p.m.

Thanks Austin. The program exhibiting these behaviors is shootout/reverse-complement. The performance monitoring I used was Intel's pcm from http://software.intel.com/en-us/articles/intel-performance-counter-monitor-a... I've been working only on my MBP, so no perfmon yet. I plan to investigate this with different architectures/machines when this issue percolates back up my todo list. On Wed, Jun 19, 2013 at 11:39 AM, Austin Seipp wrote:

...

I mean, it certainly *seems* reasonable a 15% hit could come from pipelining changes or cache behavior or something. I don't think alignment would really be a huge issue; post-Nehalem I believe non-aligned writes/reads are extremely cheap. Non-intuitive behavior can totally happen too: I've seen cases of adding instructions to a loop which speeds things up (e.g. by taking the extra step, you may mitigate a dependency stall, which massively helps pipelining across the loop body etc.)

Nicolas, can I ask what benchmark you're looking at? And what performance tools are you using, Intels'? If you're on Linux, the 'perf' tool on a modern kernel can be used to quickly get an overview of how many cache misses/hits your process has, how many pipeline stalls occur, etc. You can then use it to drill down a bit into the assembly that's problematic.

That might not give you an exact culprit (it could be many changes and accumulative hits,) but it's a start.

On Wed, Jun 19, 2013 at 10:43 AM, Nicolas Frisby wrote:

...
I'm also seeing performance regressions in the shootout benchmarks that I can't identify in the asm. The new asm looks better but performs worse, with a ~15% slowdown.

I fired up the performance counters in my CPU and the free Intel code for inspecting them showed that my CPU utilization took about a 10% hit, even while executing fewer total instructions.

1) Jan, perhaps we're seeing the same sort of behavior — the shootout benchmarks have extremely hot loops (hundreds of millions of iterations IIRC). I used ticky profiling too, and saw no suspicious changes in any counters.

2) Dear Low-level Gurus: How feasible is it that a ~15% slowdown in a program with a very hot loop is due to incidentally inhibiting some caching behavior (instr? data?)? Or perhaps effecting alignment? FTR my CPU is a Core i7-2620M, Sandy Bridge.

Thanks all.

On Wed, Jun 19, 2013 at 9:27 AM, Jan Stolarek wrote:

...
...
If it's not sorted out, can you open a ticket, put in the relevant

info

...
...
(so we don't need to look at the email trail), and we can tackle it when you get here. Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly.

Janek

...
Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:

ghc-devs-bounces@haskell.org]

...
On Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call to quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to work on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for later.

Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

-- Regards, Austin - PGP: 4096R/0x91384671

Jan Stolarek

4:20 p.m.

Nicolas, I kinda like that explanation, because it relieves me of any responsibility for this problem :) Still, I have reasons to suspect that this might actually be my fault. Generated Core is slightly different - the generated worker function accepts parameters in different order - and I don't know why that happens. I also don't see why this would impact performance. Looks like I will need to become familiar with the profiling tools that you mentioned. Janek Dnia środa, 19 czerwca 2013, Nicolas Frisby napisał:

...

I'm also seeing performance regressions in the shootout benchmarks that I can't identify in the asm. The new asm looks better but performs worse, with a ~15% slowdown.

I fired up the performance counters in my CPU and the free Intel code for inspecting them showed that my CPU utilization took about a 10% hit, even while executing fewer total instructions.

1) Jan, perhaps we're seeing the same sort of behavior — the shootout benchmarks have extremely hot loops (hundreds of millions of iterations IIRC). I used ticky profiling too, and saw no suspicious changes in any counters.

2) Dear Low-level Gurus: How feasible is it that a ~15% slowdown in a program with a very hot loop is due to incidentally inhibiting some caching behavior (instr? data?)? Or perhaps effecting alignment? FTR my CPU is a Core i7-2620M, Sandy Bridge.

Thanks all.

On Wed, Jun 19, 2013 at 9:27 AM, Jan Stolarek wrote:

...
...
If it's not sorted out, can you open a ticket, put in the relevant info

(so

...
we don't need to look at the email trail), and we can tackle it when you get here.

Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly.

Janek

...
Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org]

On

...
Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call

to

...
quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to

work

...
on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for

later.

...
Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

Nicolas Frisby

4:24 p.m.

The relief comes when we can confirm, explain, and hopefully avoid it :) On Jun 19, 2013 3:20 PM, "Jan Stolarek" wrote:

...

Nicolas, I kinda like that explanation, because it relieves me of any responsibility for this problem :) Still, I have reasons to suspect that this might actually be my fault. Generated Core is slightly different - the generated worker function accepts parameters in different order - and I don't know why that happens. I also don't see why this would impact performance. Looks like I will need to become familiar with the profiling tools that you mentioned.

Janek

Dnia środa, 19 czerwca 2013, Nicolas Frisby napisał:

...
I'm also seeing performance regressions in the shootout benchmarks that I can't identify in the asm. The new asm looks better but performs worse, with a ~15% slowdown.

I fired up the performance counters in my CPU and the free Intel code for inspecting them showed that my CPU utilization took about a 10% hit, even while executing fewer total instructions.

1) Jan, perhaps we're seeing the same sort of behavior -- the shootout benchmarks have extremely hot loops (hundreds of millions of iterations IIRC). I used ticky profiling too, and saw no suspicious changes in any counters.

2) Dear Low-level Gurus: How feasible is it that a ~15% slowdown in a program with a very hot loop is due to incidentally inhibiting some caching behavior (instr? data?)? Or perhaps effecting alignment? FTR my CPU is a Core i7-2620M, Sandy Bridge.

Thanks all.

On Wed, Jun 19, 2013 at 9:27 AM, Jan Stolarek
...
...
If it's not sorted out, can you open a ticket, put in the relevant info

(so

...
we don't need to look at the email trail), and we can tackle it when you get here.

Currently there's a temporary workaround: I'm using new folding rules for all primitive types, except for Integer, in which case I left the old folding rules unchanged. This of course should be modified to make all rules uniform, but for now it at least passes validation. I didn't fill the ticket, because the bug does not exist yet :) It only manifests itself in my patches, which have not been applied yet. I'll add all the information from this discussion to my github fork of GHC and then move it to Trac once the bug makes it to HEAD.

What worries me more about my patches is the performance regression in kahan, because I see no obvious differences in the generated assembly.

Janek

...
Simon

-----Original Message----- From: ghc-devs-bounces@haskell.org [mailto:ghc-devs-bounces@haskell.org]

On

...
Behalf Of Jan Stolarek Sent: 20 May 2013 12:35 To: Ian Lynagh Cc: ghc-devs@haskell.org Subject: Re: Integer constant folding in the presence of new primops

...
If you remove everything but the quotInteger test from integerConstantFolding and compile with -ddump-rule-rewrites then you'll see that the eqInteger rule fires before quotInteger. This is presumably comparing against 0, as the definition of quot for Integer (in GHC.Real) is _ `quot` 0 = divZeroError n `quot` d = n `quotInteger` d

Yes, I noticed these two rules firing together - perhaps that's the explanation why. I created a small program for testing:

main = print quotInt quotInt :: Integer quotInt = 100063 `quot` 156

I noticed that when I define eqInteger wrapper to be NOINLINE, the call

to

...
quot is translated to Core as:

Main.quotInt = GHC.Real.$fIntegralInteger_$cquot (__integer 100063) (__integer 156)

but when I change the wrapper to INLINE I get:

Main.quotInt = GHC.Real.$fNumRatio_$cquot <-------- NumRatio instead of IntegralInteger (__integer 100063) (__integer 156)

All rule firing happens later (I used -ddump-simpl-iterations -ddump-rule-firings), except that for $fNumRatio_$cquot the quot rules don't fire.

...
Do you also still have eqInteger wired in? It sounds like you might have given them both the same unique?

No, they didn't have the same unique. I modified the existing rules to

work

...
on the new primops and ignore their wrappers. At the moment I reverted these changes so that I can make progress and leave this problem for

later.

...
Janek

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

4354

Age (days ago)

4388

Last active (days ago)

List overview

Download

11 comments

5 participants

participants (5)

Austin Seipp
Ian Lynagh
Jan Stolarek
Nicolas Frisby
Simon Peyton-Jones