[AMDGPU] Fix DynLDS causing crash when LowerLDS is run at fullLTO pip…

…eline (#96038)

Direct mapped dynamic LDS is not lowered in the LowerLDSModule pass.
Hence it is not marked with an absolute symbol. When the LowerLDS pass is
rerun in LTO, compilation fails with an assert "cannot mix abs and non-abs LDVs".
This patch adds an additional check for direct mapped dynLDS to skip the assert.

Fixes SWDEV-454281

llvm/lib/Target/AMDGPU/Utils/AMDGPUMemoryUtils.cpp

@@ -207,14 +207,19 @@ LDSUsesInfoTy getTransitiveUsesOfLDS(const CallGraph &CG, Module &M) {
   }
 
   // Verify that we fall into one of 2 cases:
-  //    - All variables are absolute: this is a re-run of the pass
+  //    - All variables are either absolute 
+  //      or direct mapped dynamic LDS that is not lowered.
+  //      this is a re-run of the pass
   //      so we don't have anything to do.
   //    - No variables are absolute.
   std::optional<bool> HasAbsoluteGVs;
   for (auto &Map : {DirectMapKernel, IndirectMapKernel}) {
     for (auto &[Fn, GVs] : Map) {
       for (auto *GV : GVs) {
         bool IsAbsolute = GV->isAbsoluteSymbolRef();
+        bool IsDirectMapDynLDSGV = AMDGPU::isDynamicLDS(*GV) && DirectMapKernel.contains(Fn);
+        if (IsDirectMapDynLDSGV)
+          continue;
         if (HasAbsoluteGVs.has_value()) {
           if (*HasAbsoluteGVs != IsAbsolute) {
             report_fatal_error(

llvm/test/CodeGen/AMDGPU/lds-mixed-absolute-dynlds.ll

@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds < %s 2>&1 | FileCheck %s
+; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s 2>&1 | FileCheck %s
+
+; Dynamic LDS that are direct mapped are not lowered in LowerModuleLDS pass.
+; In such cases, LowerModuleLDS is free to leave it in and ignore it, and we want to make sure
+; LowerModuleLDS doesn't crash if it re-runs on such modules.
+
+@loweredlds = addrspace(3) global i32 poison, !absolute_symbol !0
+@dynlds = external addrspace(3) global [0 x i32]
+
+define amdgpu_kernel void @kern(i32 %val0) {
+; CHECK-LABEL: define amdgpu_kernel void @kern(
+; CHECK-SAME: i32 [[VAL0:%.*]]) {
+; CHECK-NEXT:    store i32 0, ptr addrspace(3) @loweredlds, align 4
+; CHECK-NEXT:    store i32 1, ptr addrspace(3) @dynlds, align 4
+; CHECK-NEXT:    ret void
+;
+  store i32 0, ptr addrspace(3) @loweredlds
+  store i32 1, ptr addrspace(3) @dynlds
+  ret void
+}
+
+
+!0 = !{i32 0, i32 1}

llvm/test/CodeGen/AMDGPU/lds-run-twice.ll

@@ -7,10 +7,12 @@
 ; Check AMDGPULowerModuleLDS can run more than once on the same module, and that
 ; the second run is a no-op.
 
+@dynlds = external addrspace(3) global [0 x i32], align 4
 @lds = internal unnamed_addr addrspace(3) global i32 undef, align 4
 
 define amdgpu_kernel void @test() {
 entry:
+  store i32 0, ptr addrspace(3) @dynlds
   store i32 1, ptr addrspace(3) @lds
   ret void
 }