PR #18948: [GPU][NFC] Improve error messages.

Imported from GitHub PR #18948

Copybara import of the project:

--
80e717c by Ilia Sergachev <isergachev@nvidia.com>:

[GPU][NFC] Improve error messages.

Merging this change closes #18948

FUTURE_COPYBARA_INTEGRATE_REVIEW=#18948 from openxla:improve_error_messages 80e717c
PiperOrigin-RevId: 693279083

xla/service/gpu/autotuning/gemm_fusion_autotuner.cc

@@ -1039,7 +1039,7 @@ GemmFusionAutotunerImpl::CompileAll(AutotunerCompileUtil& compile_util,
           absl::StatusOr<bool> has_executable =
               compile(fusion, config, gemm_config_set.size() > 1);
           TF_CHECK_OK(has_executable.status())
-              << "Failure occured when compiling fusion " << fusion->name()
+              << " - Failure occured when compiling fusion " << fusion->name()
               << " with config '" << ToString(config)
               << "'\nFused HLO computation:\n"
               << fusion->fused_instructions_computation()->ToString();

xla/service/gpu/triton_fusion_analysis.cc

@@ -162,7 +162,8 @@ absl::Status FusionContext::PropagateDimensionOrdersToParameters(
 
     if (!std::holds_alternative<DimOrdersAndReqs>(result)) {
       return FailedPrecondition(
-          "Can not propagate dim orders and requirements.");
+          "Can not propagate dim orders and requirements: %s",
+          std::get<FusionDecision>(result).Explain());
     }
 
     if (!CombineDimOrdersAndReqs(std::get<DimOrdersAndReqs>(result))) {