SLSR: Pass address space to isLegalAddressingMode

This only updates one of the uses. The other is used in cases
that may never touch memory, so I'm not sure why this is even
calling it. Perhaps there should be a new, similar hook for such
cases or pass -1 for unknown address space.

llvm-svn: 239540

diff --git a/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
index 453503a..8e5cf46 100644 (file)
--- a/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
+++ b/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
@@ -265,8 +265,10 @@ static bool isGEPFoldable(GetElementPtrInst *GEP,
       BaseOffset += DL->getStructLayout(STy)->getElementOffset(Field);
     }
   }
+
+  unsigned AddrSpace = GEP->getPointerAddressSpace();
   return TTI->isLegalAddressingMode(GEP->getType()->getElementType(), BaseGV,
-                                    BaseOffset, HasBaseReg, Scale);
+                                    BaseOffset, HasBaseReg, Scale, AddrSpace);
 }
 
 // Returns whether (Base + Index * Stride) can be folded to an addressing mode.

diff --git a/llvm/test/Transforms/StraightLineStrengthReduce/R600/lit.local.cfg b/llvm/test/Transforms/StraightLineStrengthReduce/R600/lit.local.cfg
new file mode 100644 (file)
index 0000000..ad9ce25
--- /dev/null
+++ b/llvm/test/Transforms/StraightLineStrengthReduce/R600/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'R600' in config.root.targets:
+    config.unsupported = True

diff --git a/llvm/test/Transforms/StraightLineStrengthReduce/R600/reassociate-geps-and-slsr-addrspace.ll b/llvm/test/Transforms/StraightLineStrengthReduce/R600/reassociate-geps-and-slsr-addrspace.ll
new file mode 100644 (file)
index 0000000..278250a
--- /dev/null
+++ b/llvm/test/Transforms/StraightLineStrengthReduce/R600/reassociate-geps-and-slsr-addrspace.ll
@@ -0,0 +1,107 @@
+; RUN: opt -S -mtriple=amdgcn-- -separate-const-offset-from-gep -slsr -gvn < %s | FileCheck %s
+
+target datalayout = "e-p:32:32-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-p24:64:64-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64"
+
+
+; CHECK-LABEL: @slsr_after_reassociate_global_geps_mubuf_max_offset(
+; CHECK: [[b1:%[0-9]+]] = getelementptr float, float addrspace(1)* %arr, i64 [[bump:%[0-9]+]]
+; CHECK: [[b2:%[0-9]+]] = getelementptr float, float addrspace(1)* [[b1]], i64 [[bump]]
+define void @slsr_after_reassociate_global_geps_mubuf_max_offset(float addrspace(1)* %out, float addrspace(1)* noalias %arr, i32 %i) {
+bb:
+  %i2 = shl nsw i32 %i, 1
+  %j1 = add nsw i32 %i, 1023
+  %tmp = sext i32 %j1 to i64
+  %p1 = getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp
+  %tmp3 = bitcast float addrspace(1)* %p1 to i32 addrspace(1)*
+  %v11 = load i32, i32 addrspace(1)* %tmp3, align 4
+  %tmp4 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v11, i32 addrspace(1)* %tmp4, align 4
+
+  %j2 = add nsw i32 %i2, 1023
+  %tmp5 = sext i32 %j2 to i64
+  %p2 = getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp5
+  %tmp6 = bitcast float addrspace(1)* %p2 to i32 addrspace(1)*
+  %v22 = load i32, i32 addrspace(1)* %tmp6, align 4
+  %tmp7 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v22, i32 addrspace(1)* %tmp7, align 4
+
+  ret void
+}
+
+; CHECK-LABEL: @slsr_after_reassociate_global_geps_over_mubuf_max_offset(
+; CHECK: %j1 = add nsw i32 %i, 1024
+; CHECK: %tmp = sext i32 %j1 to i64
+; CHECK: getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp
+; CHECK: getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp5
+define void @slsr_after_reassociate_global_geps_over_mubuf_max_offset(float addrspace(1)* %out, float addrspace(1)* noalias %arr, i32 %i) {
+bb:
+  %i2 = shl nsw i32 %i, 1
+  %j1 = add nsw i32 %i, 1024
+  %tmp = sext i32 %j1 to i64
+  %p1 = getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp
+  %tmp3 = bitcast float addrspace(1)* %p1 to i32 addrspace(1)*
+  %v11 = load i32, i32 addrspace(1)* %tmp3, align 4
+  %tmp4 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v11, i32 addrspace(1)* %tmp4, align 4
+
+  %j2 = add nsw i32 %i2, 1024
+  %tmp5 = sext i32 %j2 to i64
+  %p2 = getelementptr inbounds float, float addrspace(1)* %arr, i64 %tmp5
+  %tmp6 = bitcast float addrspace(1)* %p2 to i32 addrspace(1)*
+  %v22 = load i32, i32 addrspace(1)* %tmp6, align 4
+  %tmp7 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v22, i32 addrspace(1)* %tmp7, align 4
+
+  ret void
+}
+
+; CHECK-LABEL: @slsr_after_reassociate_lds_geps_ds_max_offset(
+; CHECK: [[B1:%[0-9]+]] = getelementptr float, float addrspace(3)* %arr, i32 %i
+; CHECK: getelementptr float, float addrspace(3)* [[B1]], i32 16383
+
+; CHECK: [[B2:%[0-9]+]] = getelementptr float, float addrspace(3)* [[B1]], i32 %i
+; CHECK: getelementptr float, float addrspace(3)* [[B2]], i32 16383
+define void @slsr_after_reassociate_lds_geps_ds_max_offset(float addrspace(1)* %out, float addrspace(3)* noalias %arr, i32 %i) {
+bb:
+  %i2 = shl nsw i32 %i, 1
+  %j1 = add nsw i32 %i, 16383
+  %p1 = getelementptr inbounds float, float addrspace(3)* %arr, i32 %j1
+  %tmp3 = bitcast float addrspace(3)* %p1 to i32 addrspace(3)*
+  %v11 = load i32, i32 addrspace(3)* %tmp3, align 4
+  %tmp4 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v11, i32 addrspace(1)* %tmp4, align 4
+
+  %j2 = add nsw i32 %i2, 16383
+  %p2 = getelementptr inbounds float, float addrspace(3)* %arr, i32 %j2
+  %tmp6 = bitcast float addrspace(3)* %p2 to i32 addrspace(3)*
+  %v22 = load i32, i32 addrspace(3)* %tmp6, align 4
+  %tmp7 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v22, i32 addrspace(1)* %tmp7, align 4
+
+  ret void
+}
+
+; CHECK-LABEL: @slsr_after_reassociate_lds_geps_over_ds_max_offset(
+; CHECK: %j1 = add nsw i32 %i, 16384
+; CHECK: getelementptr inbounds float, float addrspace(3)* %arr, i32 %j1
+; CHECK: %j2 = add i32 %j1, %i
+; CHECK: getelementptr inbounds float, float addrspace(3)* %arr, i32 %j2
+define void @slsr_after_reassociate_lds_geps_over_ds_max_offset(float addrspace(1)* %out, float addrspace(3)* noalias %arr, i32 %i) {
+bb:
+  %i2 = shl nsw i32 %i, 1
+  %j1 = add nsw i32 %i, 16384
+  %p1 = getelementptr inbounds float, float addrspace(3)* %arr, i32 %j1
+  %tmp3 = bitcast float addrspace(3)* %p1 to i32 addrspace(3)*
+  %v11 = load i32, i32 addrspace(3)* %tmp3, align 4
+  %tmp4 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v11, i32 addrspace(1)* %tmp4, align 4
+
+  %j2 = add nsw i32 %i2, 16384
+  %p2 = getelementptr inbounds float, float addrspace(3)* %arr, i32 %j2
+  %tmp6 = bitcast float addrspace(3)* %p2 to i32 addrspace(3)*
+  %v22 = load i32, i32 addrspace(3)* %tmp6, align 4
+  %tmp7 = bitcast float addrspace(1)* %out to i32 addrspace(1)*
+  store i32 %v22, i32 addrspace(1)* %tmp7, align 4
+
+  ret void
+}