[SeparateConstOffsetFromGEP] Fixed a bug related to unsigned modulo

[oota-llvm.git] / test / Transforms / SeparateConstOffsetFromGEP / NVPTX / split-gep.ll

diff --git a/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll b/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll
index 31989dd691604e783e3631641403b5a5a393db5c..ea0d3f5673a59594efd7296f914bf85500e2704f 100644 (file)
--- a/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll
+++ b/test/Transforms/SeparateConstOffsetFromGEP/NVPTX/split-gep.ll
@@ -26,21 +26,23 @@ entry:
 ; CHECK-LABEL: @struct(
 ; CHECK: getelementptr [1024 x %struct.S]* @struct_array, i64 0, i64 %{{[a-zA-Z0-9]+}}, i32 1
 
-; We should be able to trace into s/zext(a + b) if a + b is non-negative
+; We should be able to trace into sext(a + b) if a + b is non-negative
 ; (e.g., used as an index of an inbounds GEP) and one of a and b is
 ; non-negative.
 define float* @sext_add(i32 %i, i32 %j) {
 entry:
   %0 = add i32 %i, 1
   %1 = sext i32 %0 to i64  ; inbound sext(i + 1) = sext(i) + 1
-  %2 = sub i32 %j, 2
-  ; However, inbound sext(j - 2) != sext(j) - 2, e.g., j = INT_MIN
+  %2 = add i32 %j, -2
+  ; However, inbound sext(j + -2) != sext(j) + -2, e.g., j = INT_MIN
   %3 = sext i32 %2 to i64
   %p = getelementptr inbounds [32 x [32 x float]]* @float_2d_array, i64 0, i64 %1, i64 %3
   ret float* %p
 }
 ; CHECK-LABEL: @sext_add(
 ; CHECK-NOT: = add
+; CHECK: add i32 %j, -2
+; CHECK: sext
 ; CHECK: getelementptr [32 x [32 x float]]* @float_2d_array, i64 0, i64 %{{[a-zA-Z0-9]+}}, i64 %{{[a-zA-Z0-9]+}}
 ; CHECK: getelementptr float* %{{[a-zA-Z0-9]+}}, i64 32
 
@@ -73,16 +75,16 @@ entry:
   %2 = zext i48 %1 to i64    ; zext(sext(a +nsw nuw 1)) = zext(sext(a)) + 1
   %3 = add nsw i32 %b, 2
   %4 = sext i32 %3 to i48
-  %5 = zext i48 %4 to i64    ; zext(sext(a +nsw 2)) != zext(sext(a)) + 2
-  %p1 = getelementptr inbounds [32 x [32 x float]]* @float_2d_array, i64 0, i64 %2, i64 %5
+  %5 = zext i48 %4 to i64    ; zext(sext(b +nsw 2)) != zext(sext(b)) + 2
+  %p1 = getelementptr [32 x [32 x float]]* @float_2d_array, i64 0, i64 %2, i64 %5
   store float* %p1, float** %out1
   %6 = add nuw i32 %a, 3
   %7 = zext i32 %6 to i48
-  %8 = sext i48 %7 to i64 ; sext(zext(b +nuw 3)) = zext(b +nuw 3) = zext(b) + 3
+  %8 = sext i48 %7 to i64 ; sext(zext(a +nuw 3)) = zext(a +nuw 3) = zext(a) + 3
   %9 = add nsw i32 %b, 4
   %10 = zext i32 %9 to i48
   %11 = sext i48 %10 to i64  ; sext(zext(b +nsw 4)) != zext(b) + 4
-  %p2 = getelementptr inbounds [32 x [32 x float]]* @float_2d_array, i64 0, i64 %8, i64 %11
+  %p2 = getelementptr [32 x [32 x float]]* @float_2d_array, i64 0, i64 %8, i64 %11
   store float* %p2, float** %out2
   ret void
 }
@@ -232,3 +234,46 @@ entry:
 ; CHECK-LABEL: @and(
 ; CHECK: getelementptr [32 x [32 x float]]* @float_2d_array
 ; CHECK-NOT: getelementptr
+
+; The code that rebuilds an OR expression used to be buggy, and failed on this
+; test.
+define float* @shl_add_or(i64 %a, float* %ptr) {
+; CHECK-LABEL: @shl_add_or(
+entry:
+  %shl = shl i64 %a, 2
+  %add = add i64 %shl, 12
+  %or = or i64 %add, 1
+; CHECK: [[OR:%or[0-9]*]] = add i64 %shl, 1
+  ; ((a << 2) + 12) and 1 have no common bits. Therefore,
+  ; SeparateConstOffsetFromGEP is able to extract the 12.
+  ; TODO(jingyue): We could reassociate the expression to combine 12 and 1.
+  %p = getelementptr float* %ptr, i64 %or
+; CHECK: [[PTR:%[a-zA-Z0-9]+]] = getelementptr float* %ptr, i64 [[OR]]
+; CHECK: getelementptr float* [[PTR]], i64 12
+  ret float* %p
+; CHECK-NEXT: ret
+}
+
+; The source code used to be buggy in checking
+; (AccumulativeByteOffset % ElementTypeSizeOfGEP == 0)
+; where AccumulativeByteOffset is signed but ElementTypeSizeOfGEP is unsigned.
+; The compiler would promote AccumulativeByteOffset to unsigned, causing
+; unexpected results. For example, while -64 % (int64_t)24 != 0,
+; -64 % (uint64_t)24 == 0.
+%struct3 = type { i64, i32 }
+%struct2 = type { %struct3, i32 }
+%struct1 = type { i64, %struct2 }
+%struct0 = type { i32, i32, i64*, [100 x %struct1] }
+define %struct2* @sign_mod_unsign(%struct0* %ptr, i64 %idx) {
+; CHECK-LABEL: @sign_mod_unsign(
+entry:
+  %arrayidx = add nsw i64 %idx, -2
+; CHECK-NOT: add
+  %ptr2 = getelementptr inbounds %struct0* %ptr, i64 0, i32 3, i64 %arrayidx, i32 1
+; CHECK: [[PTR:%[a-zA-Z0-9]+]] = getelementptr %struct0* %ptr, i64 0, i32 3, i64 %idx, i32 1
+; CHECK: [[PTR1:%[a-zA-Z0-9]+]] = bitcast %struct2* [[PTR]] to i8*
+; CHECK: getelementptr i8* [[PTR1]], i64 -64
+; CHECK: bitcast
+  ret %struct2* %ptr2
+; CHECK-NEXT: ret
+}