[flang][OpenMP] Fix regression in !$ continuation

[klausler]

A recent patch that obviated the need to use -fopenmp when using the compiler to preprocess in -E mode broke a case of Fortran line continuation when using OpenMP conditional compilation lines (!$) when not in -E mode. Fix.

[llvmbot]

@llvm/pr-subscribers-flang-openmp

@llvm/pr-subscribers-flang-parser

Author: Peter Klausler (klausler)

Changes

A recent patch that obviated the need to use -fopenmp when using the compiler to preprocess in -E mode broke a case of Fortran line continuation when using OpenMP conditional compilation lines (!$) when not in -E mode. Fix.

---

Full diff: https://github.com/llvm/llvm-project/pull/134756.diff

1 Files Affected:

• (modified) flang/lib/Parser/prescan.cpp (+23-12)

diff --git a/flang/lib/Parser/prescan.cpp b/flang/lib/Parser/prescan.cpp
index 31fdadeddef53..a55a610a1f4f2 100644
--- a/flang/lib/Parser/prescan.cpp
+++ b/flang/lib/Parser/prescan.cpp
@@ -1351,26 +1351,37 @@ const char *Prescanner::FreeFormContinuationLine(bool ampersand) {
   }
   p = SkipWhiteSpaceIncludingEmptyMacros(p);
   if (InCompilerDirective()) {
-    if (preprocessingOnly_ && directiveSentinel_[0] == '$' &&
-        directiveSentinel_[1] == '\0') {
-      // in -E mode, don't treat !$ as a continuation
+    if (directiveSentinel_[0] == '$' && directiveSentinel_[1] == '\0') {
+      if (preprocessingOnly_) {
+        // in -E mode, don't treat !$ as a continuation
+        return nullptr;
+      } else if (p[0] == '!' && p[1] == '$') {
+        // accept but do not require a matching sentinel
+        if (IsLetter(p[2])) {
+          return nullptr; // not !$
+        }
+        p += 2;
+      }
     } else if (*p++ == '!') {
       for (const char *s{directiveSentinel_}; *s != '\0'; ++p, ++s) {
         if (*s != ToLowerCaseLetter(*p)) {
           return nullptr; // not the same directive class
         }
       }
-      p = SkipWhiteSpace(p);
-      if (*p == '&') {
-        if (!ampersand) {
-          insertASpace_ = true;
-        }
-        return p + 1;
-      } else if (ampersand) {
-        return p;
+    } else {
+      return nullptr;
+    }
+    p = SkipWhiteSpace(p);
+    if (*p == '&') {
+      if (!ampersand) {
+        insertASpace_ = true;
       }
+      return p + 1;
+    } else if (ampersand) {
+      return p;
+    } else {
+      return nullptr;
     }
-    return nullptr;
   }
   if (p[0] == '!' && p[1] == '$' && !preprocessingOnly_ &&
       features_.IsEnabled(LanguageFeature::OpenMP)) {

[eugeneepshteyn]

Maybe some tests should be added for the failure fixed by this PR? (I don't remember seeing regression in check-flang, so perhaps there are no tests there that cover this condition...)

[klausler]

Added tests, and verified changes against test suite failures.

[ro-i]

This commit (0ae9bb9) breaks building a SPEC openmp benchmark (https://www.spec.org/auto/omp2012/Docs/357.bt331.html). A reduced test case:

program BT
!$     integer  omp_get_max_threads
!$     external omp_get_max_threads

c---------------------------------------------------------------------
       write(*, '(A1)') ' '
       end

This leads to the following compile error:

$ flang -fopenmp -c bt.condcomp.f
error: Could not parse bt.condcomp.f
./bt.condcomp.f:1:1: warning: Character in fixed-form label field must be a digit
  program BT
  ^
./bt.condcomp.f:5:6: error: expected end of statement
  c---------------------------------------------------------------------
       ^
./bt.condcomp.f:3:8: in the context: specification construct
  !$     external omp_get_max_threads
         ^
./bt.condcomp.f:2:8: in the context: specification part
  !$     integer  omp_get_max_threads
         ^

[klausler]

Fix is ready, will merge today. From: Robert Imschweiler ***@***.***> Date: Monday, April 14, 2025 at 08:45 To: llvm/llvm-project ***@***.***> Cc: Peter Klausler ***@***.***>, State change ***@***.***> Subject: Re: [llvm/llvm-project] [flang][OpenMP] Fix regression in !$ continuation (PR #134756) This commit (0ae9bb9<0ae9bb9>) breaks building a SPEC openmp benchmark (https://www.spec.org/auto/omp2012/Docs/357.bt331.html). A reduced test case: program BT !$ integer omp_get_max_threads !$ external omp_get_max_threads c--------------------------------------------------------------------- write(*, '(A1)') ' ' end This leads to the following compile error: $ flang -fopenmp -c bt.condcomp.f error: Could not parse bt.condcomp.f ./bt.condcomp.f:1:1: warning: Character in fixed-form label field must be a digit program BT ^ ./bt.condcomp.f:5:6: error: expected end of statement c--------------------------------------------------------------------- ^ ./bt.condcomp.f:3:8: in the context: specification construct !$ external omp_get_max_threads ^ ./bt.condcomp.f:2:8: in the context: specification part !$ integer omp_get_max_threads ^ — Reply to this email directly, view it on GitHub<#134756 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AIRI5XPBYKGM6DDMNCF3Q4D2ZPJZRAVCNFSM6AAAAAB2U2RVDGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQMBSGE2DINBSGY>. You are receiving this because you modified the open/close state.Message ID: ***@***.***> [Image removed by sender.]ro-i left a comment (llvm/llvm-project#134756)<#134756 (comment)> This commit (0ae9bb9<0ae9bb9>) breaks building a SPEC openmp benchmark (https://www.spec.org/auto/omp2012/Docs/357.bt331.html). A reduced test case: program BT !$ integer omp_get_max_threads !$ external omp_get_max_threads c--------------------------------------------------------------------- write(*, '(A1)') ' ' end This leads to the following compile error: $ flang -fopenmp -c bt.condcomp.f error: Could not parse bt.condcomp.f ./bt.condcomp.f:1:1: warning: Character in fixed-form label field must be a digit program BT ^ ./bt.condcomp.f:5:6: error: expected end of statement c--------------------------------------------------------------------- ^ ./bt.condcomp.f:3:8: in the context: specification construct !$ external omp_get_max_threads ^ ./bt.condcomp.f:2:8: in the context: specification part !$ integer omp_get_max_threads ^ — Reply to this email directly, view it on GitHub<#134756 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AIRI5XPBYKGM6DDMNCF3Q4D2ZPJZRAVCNFSM6AAAAAB2U2RVDGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQMBSGE2DINBSGY>. You are receiving this because you modified the open/close state.Message ID: ***@***.***>