[flang] Interpret 'Q' exponent letter as kind=16 even on x86

[klausler]

The compiler was interpreting 'Q' as an exponent letter in a literal real constant as meaning real(kind=10) on x86-64, which is the legacy 80387 80-bit extended precision floating-point type. It turns out that 'Q' means kind=16 with all other compilers, even for x86-64 targets. Change to conform.

[llvmbot]

@llvm/pr-subscribers-flang-semantics

@llvm/pr-subscribers-flang-driver

Author: Peter Klausler (klausler)

Changes

The compiler was interpreting 'Q' as an exponent letter in a literal real constant as meaning real(kind=10) on x86-64, which is the legacy 80387 80-bit extended precision floating-point type. It turns out that 'Q' means kind=16 with all other compilers, even for x86-64 targets. Change to conform.

---

Full diff: https://github.com/llvm/llvm-project/pull/124158.diff

3 Files Affected:

• (modified) flang/lib/Common/default-kinds.cpp (+1-5)
• (modified) flang/lib/Frontend/CompilerInstance.cpp (-3)
• (modified) flang/test/Semantics/kinds04_q10.f90 (+4-9)

diff --git a/flang/lib/Common/default-kinds.cpp b/flang/lib/Common/default-kinds.cpp
index fbafd827ff0d0b..d2ca9103513617 100644
--- a/flang/lib/Common/default-kinds.cpp
+++ b/flang/lib/Common/default-kinds.cpp
@@ -11,11 +11,7 @@
 
 namespace Fortran::common {
 
-IntrinsicTypeDefaultKinds::IntrinsicTypeDefaultKinds() {
-#if __x86_64__
-  quadPrecisionKind_ = 10;
-#endif
-}
+IntrinsicTypeDefaultKinds::IntrinsicTypeDefaultKinds() {}
 
 IntrinsicTypeDefaultKinds &IntrinsicTypeDefaultKinds::set_defaultIntegerKind(
     int k) {
diff --git a/flang/lib/Frontend/CompilerInstance.cpp b/flang/lib/Frontend/CompilerInstance.cpp
index 298790bae66558..dfd15b9c3c39d9 100644
--- a/flang/lib/Frontend/CompilerInstance.cpp
+++ b/flang/lib/Frontend/CompilerInstance.cpp
@@ -153,9 +153,6 @@ bool CompilerInstance::executeAction(FrontendAction &act) {
   CompilerInvocation &invoc = this->getInvocation();
 
   llvm::Triple targetTriple{llvm::Triple(invoc.getTargetOpts().triple)};
-  if (targetTriple.getArch() == llvm::Triple::ArchType::x86_64) {
-    invoc.getDefaultKinds().set_quadPrecisionKind(10);
-  }
 
   // Set some sane defaults for the frontend.
   invoc.setDefaultFortranOpts();
diff --git a/flang/test/Semantics/kinds04_q10.f90 b/flang/test/Semantics/kinds04_q10.f90
index d352daa1cbbf06..aa5c4abe2f1dfc 100644
--- a/flang/test/Semantics/kinds04_q10.f90
+++ b/flang/test/Semantics/kinds04_q10.f90
@@ -1,14 +1,9 @@
-! RUN: %python %S/test_errors.py %s %flang_fc1 -triple x86_64-unknown-linux-gnu
+! RUN: %python %S/test_errors.py %s %flang_fc1
 ! C716 If both kind-param and exponent-letter appear, exponent-letter
 ! shall be E. (As an extension we also allow an exponent-letter which matches
 ! the kind-param)
 ! C717 The value of kind-param shall specify an approximation method that
 ! exists on the processor.
-!
-! This test is for x86_64, where exponent-letter 'q' is for
-! 10-byte extended precision
-! UNSUPPORTED: system-windows, system-aix
-! REQUIRES: x86-registered-target
 
 subroutine s(var)
   real :: realvar1 = 4.0E6_4
@@ -16,9 +11,9 @@ subroutine s(var)
   real :: realvar3 = 4.0Q6
   !PORTABILITY: Explicit kind parameter together with non-'E' exponent letter is not standard
   real :: realvar4 = 4.0D6_8
-  !PORTABILITY: Explicit kind parameter together with non-'E' exponent letter is not standard
-  real :: realvar5 = 4.0Q6_10
   !WARNING: Explicit kind parameter on real constant disagrees with exponent letter 'q'
+  real :: realvar5 = 4.0Q6_10
+  !PORTABILITY: Explicit kind parameter together with non-'E' exponent letter is not standard
   real :: realvar6 = 4.0Q6_16
   real :: realvar7 = 4.0E6_8
   real :: realvar8 = 4.0E6_10
@@ -31,7 +26,7 @@ subroutine s(var)
   double precision :: doublevar3 = 4.0Q6
   !PORTABILITY: Explicit kind parameter together with non-'E' exponent letter is not standard
   double precision :: doublevar4 = 4.0D6_8
-  !WARNING: Explicit kind parameter on real constant disagrees with exponent letter 'q'
+  !PORTABILITY: Explicit kind parameter together with non-'E' exponent letter is not standard
   double precision :: doublevar5 = 4.0Q6_16
   double precision :: doublevar6 = 4.0E6_8
   double precision :: doublevar7 = 4.0E6_10

[llvm-ci]

LLVM Buildbot has detected a new failure on builder ppc64-flang-aix running on ppc64-flang-aix-test while building flang at step 6 "test-build-unified-tree-check-flang".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/201/builds/2291

Here is the relevant piece of the build log for the reference

Step 6 (test-build-unified-tree-check-flang) failure: test (failure)
******************** TEST 'Flang :: Semantics/kinds04_q10.f90' FAILED ********************
Exit Code: 1

Command Output (stdout):
--
--- 
+++ 
@@ -3 +2,0 @@

actual at 15: underflow on REAL(10) to REAL(4) conversion@@ -5 +3,0 @@

actual at 19: underflow on REAL(10) to REAL(4) conversion@@ -9 +6,0 @@

actual at 32: underflow on REAL(10) to REAL(8) conversion

FAIL

--
Command Output (stderr):
--
RUN: at line 1: "/home/llvm/llvm-external-buildbots/workers/env/bin/python3.11" /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/llvm-project/flang/test/Semantics/test_errors.py /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/llvm-project/flang/test/Semantics/kinds04_q10.f90 /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/build/bin/flang -fc1
+ /home/llvm/llvm-external-buildbots/workers/env/bin/python3.11 /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/llvm-project/flang/test/Semantics/test_errors.py /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/llvm-project/flang/test/Semantics/kinds04_q10.f90 /home/llvm/llvm-external-buildbots/workers/ppc64-flang-aix-test/ppc64-flang-aix-build/build/bin/flang -fc1

--

********************