AMDGPU: mark ldexp LibCalls as unavailable

[nhaehnle]

Summary

The LibCallSimplifier will turn llvm.exp2.* intrinsics into ldexp* libcalls
which do not make sense with the AMDGPU backend.

In the long run, we'll want an llvm.ldexp.* intrinsic to properly make use of
this optimization, but this works around the problem for now.

See also: http://reviews.llvm.org/D14327 (suggested llvm.ldexp.* implementation)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92709

[• tstellarAMD]

test/Transforms/InstCombine/exp2-1.ll

84–89 ↗

(On Diff #41155)

I'm confused about this test. amdgcn does not support ldexp, so why is it being checked here?

[arsenm]

test/Transforms/InstCombine/exp2-1.ll

5 ↗

(On Diff #41155)

the -mcpu is unnecessary

[nhaehnle]

Fix test: remove unneeded -mcpu; also, win and amdgpu behave differently

[• tstellarAMD]

test/Transforms/InstCombine/exp2-1.ll

84–86 ↗

(On Diff #41606)

If you add an additional -check-prefix=CHECK option to all of the RUN lines, then these can be consolidated into a single CHECK-LABEL: line. In fact I think CHECK is a default prefix, so you may not even have to add the extra -check-prefix to consolidate these.

96–98 ↗

(On Diff #41606)

These can be consolidated too.

[arsenm]

test/Transforms/InstCombine/exp2-1.ll

84–86 ↗

(On Diff #41606)

the added CHECK is needed. It's only implicitly there if no other check prefix is specified

[nhaehnle]

Consolidate test labels

I left the first batch of tests (where the input uses libcalls) untouched.
The win32 run actually doesn't apply the transformation for those, so
things would have gotten rather messy.

[• tstellarAMD]

LGTM.

[nhaehnle]

I don't have commit access, could you push this out?

llvm/trunk/lib/Analysis/TargetLibraryInfo.cpp

 #ifndef NDEBUG
   // Verify that the StandardNames array is in alphabetical order.
   for (unsigned F = 1; F < LibFunc::NumLibFuncs; ++F) {
     if (strcmp(StandardNames[F-1], StandardNames[F]) >= 0)
       llvm_unreachable("TargetLibraryInfoImpl function names must be sorted");
   }
 #endif // !NDEBUG
 
+  if (T.getArch() == Triple::r600 ||
+      T.getArch() == Triple::amdgcn) {
+    TLI.setUnavailable(LibFunc::ldexp);
+    TLI.setUnavailable(LibFunc::ldexpf);
+    TLI.setUnavailable(LibFunc::ldexpl);
+  }
+
   // There are no library implementations of mempcy and memset for AMD gpus and
   // these can be difficult to lower in the backend.
   if (T.getArch() == Triple::r600 ||
       T.getArch() == Triple::amdgcn ||
       T.getArch() == Triple::wasm32 ||
       T.getArch() == Triple::wasm64) {
     TLI.setUnavailable(LibFunc::memcpy);
     TLI.setUnavailable(LibFunc::memset);

llvm/trunk/test/Transforms/InstCombine/exp2-1.ll

 ; Test that the exp2 library call simplifier works correctly.
 ;
-; RUN: opt < %s -instcombine -S | FileCheck %s
-; RUN: opt < %s -instcombine -S -mtriple=i386-pc-win32 | FileCheck %s -check-prefix=CHECK-WIN
+; RUN: opt < %s -instcombine -S | FileCheck %s -check-prefix=CHECK -check-prefix=INTRINSIC -check-prefix=LDEXP -check-prefix=LDEXPF
+; RUN: opt < %s -instcombine -S -mtriple=i386-pc-win32 | FileCheck %s -check-prefix=INTRINSIC -check-prefix=LDEXP -check-prefix=NOLDEXPF
+; RUN: opt < %s -instcombine -S -mtriple=amdgcn-unknown-unknown | FileCheck %s -check-prefix=INTRINSIC -check-prefix=NOLDEXP -check-prefix=NOLDEXPF
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
 
 declare double @exp2(double)
 declare float @exp2f(float)
 
 ; Check exp2(sitofp(x)) -> ldexp(1.0, sext(x)).
 

 ; CHECK: call float @ldexpf
   ret float %ret
 }
 
 declare double @llvm.exp2.f64(double)
 declare float @llvm.exp2.f32(float)
 
 define double @test_simplify9(i8 zeroext %x) {
-; CHECK-LABEL: @test_simplify9(
-; CHECK-WIN-LABEL: @test_simplify9(
+; INTRINSIC-LABEL: @test_simplify9(
   %conv = uitofp i8 %x to double
   %ret = call double @llvm.exp2.f64(double %conv)
-; CHECK: call double @ldexp
-; CHECK-WIN: call double @ldexp
+; LDEXP: call double @ldexp
+; NOLDEXP-NOT: call double @ldexp
   ret double %ret
 }
 
 define float @test_simplify10(i8 zeroext %x) {
-; CHECK-LABEL: @test_simplify10(
-; CHECK-WIN-LABEL: @test_simplify10(
+; INTRINSIC-LABEL: @test_simplify10(
   %conv = uitofp i8 %x to float
   %ret = call float @llvm.exp2.f32(float %conv)
-; CHECK: call float @ldexpf
-; CHECK-WIN-NOT: call float @ldexpf
+; LDEXPF: call float @ldexpf
+; NOLDEXPF-NOT: call float @ldexpf
   ret float %ret
 }