fix detection to run ThreeElementVectorLoweringPass (#1367)

* fix detection to run ThreeElementVectorLoweringPass

If a gep ends up trying to access the 4th element of a vec3, we need
to lower vec3 to vec4.

This is fixing a regression on CTS test_half
Ref kpet/clvk#698

* do not consider gep with only 1 indice

lib/ThreeElementVectorLoweringPass.cpp

@@ -269,22 +269,46 @@ bool clspv::ThreeElementVectorLoweringPass::vec3ShouldBeLowered(Module &M) {
     return false;
   default:
     for (auto &F : M.functions()) {
-      if (vec3BitcastInFunction(F))
+      if (vec3ShouldBeLowered(F))
         return true;
     }
     return false;
   }
 }
 
-bool clspv::ThreeElementVectorLoweringPass::vec3BitcastInFunction(Function &F) {
+bool clspv::ThreeElementVectorLoweringPass::vec3ShouldBeLowered(Function &F) {
   for (Instruction &I : instructions(F)) {
     if (haveImplicitCast(&I)) {
       return true;
+    } else if (haveInvalidVec3GEP(&I)) {
+      return true;
     }
   }
   return false;
 }
 
+bool clspv::ThreeElementVectorLoweringPass::haveInvalidVec3GEP(Value *Value) {
+  auto gep = dyn_cast<GetElementPtrInst>(Value);
+  if (!gep || gep->getNumIndices() <= 1) {
+    return false;
+  }
+
+  SmallVector<llvm::Value *> idxs(gep->idx_begin(), gep->idx_end() - 1);
+  auto last_type =
+      GetElementPtrInst::getIndexedType(gep->getSourceElementType(), idxs);
+  auto vec_type = dyn_cast<FixedVectorType>(last_type);
+  if (!vec_type || vec_type->getNumElements() != 3) {
+    return false;
+  }
+  auto last_idx = gep->getOperand(gep->getNumOperands() - 1);
+  auto cst_idx = dyn_cast<ConstantInt>(last_idx);
+  if (!cst_idx || cst_idx->getZExtValue() >= 3) {
+    return true;
+  }
+
+  return false;
+}
+
 bool clspv::ThreeElementVectorLoweringPass::haveImplicitCast(Value *Value) {
   Type *source_ty = nullptr;
   Type *dest_ty = nullptr;

lib/ThreeElementVectorLoweringPass.h

@@ -104,8 +104,9 @@ struct ThreeElementVectorLoweringPass
 private:
   // High-level implementation details of runOnModule.
 
-  /// Look for bitcast of vec3 inside the function
-  bool vec3BitcastInFunction(llvm::Function &F);
+  /// Look for vec3 patterns inside the function that requires lowering vec3 to
+  /// vec4.
+  bool vec3ShouldBeLowered(llvm::Function &F);
 
   /// Returns whether the vec3 should be transformed into vec4
   bool vec3ShouldBeLowered(llvm::Module &M);
@@ -114,6 +115,10 @@ struct ThreeElementVectorLoweringPass
   /// opaque pointers
   bool haveImplicitCast(llvm::Value *Value);
 
+  // Returns true if the type before last indice is a vec3 and last indice is
+  // not constant or bigger or equal to 3.
+  bool haveInvalidVec3GEP(llvm::Value *Value);
+
   /// Lower all global variables in the module.
   bool runOnGlobals(llvm::Module &M);
 

test/ThreeElementVectorLowering/invalid_vec3_gep.ll

@@ -0,0 +1,13 @@
+; RUN: clspv-opt %s -o %t.ll --passes=three-element-vector-lowering
+; RUN: FileCheck %s < %t.ll
+
+; CHECK:  getelementptr inbounds <4 x i32>, ptr %a, i32 2, i32 3
+
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+define dso_local spir_kernel void @test1(ptr %a) {
+entry:
+  %gep = getelementptr inbounds <3 x i32>, ptr %a, i32 2, i32 3
+  ret void
+}

test/ThreeElementVectorLowering/invalid_vec3_gep2.ll

@@ -0,0 +1,13 @@
+; RUN: clspv-opt %s -o %t.ll --passes=three-element-vector-lowering
+; RUN: FileCheck %s < %t.ll
+
+; CHECK:  getelementptr inbounds <4 x i32>, ptr %a, i32 2, i32 %i
+
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+define dso_local spir_kernel void @test1(ptr %a, i32 %i) {
+entry:
+  %gep = getelementptr inbounds <3 x i32>, ptr %a, i32 2, i32 %i
+  ret void
+}

test/ThreeElementVectorLowering/valid_vec3_gep.ll

@@ -0,0 +1,13 @@
+; RUN: clspv-opt %s -o %t.ll --passes=three-element-vector-lowering
+; RUN: FileCheck %s < %t.ll
+
+; CHECK:  getelementptr inbounds <3 x i32>, ptr %a, i32 2, i32 2
+
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+define dso_local spir_kernel void @test1(ptr %a) {
+entry:
+  %gep = getelementptr inbounds <3 x i32>, ptr %a, i32 2, i32 2
+  ret void
+}