1 ; RUN: llc < %s -march=x86-64 | grep lea | count 12 2 3 ; This testcase was written to demonstrate an instruction-selection problem, 4 ; however it also happens to expose a limitation in the DAGCombiner's 5 ; expression reassociation which causes it to miss opportunities for 6 ; constant folding due to the intermediate adds having multiple uses. 7 ; The Reassociate pass has similar limitations. If these limitations are 8 ; fixed, the test commands above will need to be updated to expect fewer 9 ; lea instructions. 10 11 @g0 = weak global [1000 x i32] zeroinitializer, align 32 ; <[1000 x i32]*> [#uses=8] 12 @g1 = weak global [1000 x i32] zeroinitializer, align 32 ; <[1000 x i32]*> [#uses=7] 13 14 define void @foo() { 15 entry: 16 %tmp4 = load i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 0) ; <i32> [#uses=1] 17 %tmp8 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 0) ; <i32> [#uses=1] 18 %tmp9 = add i32 %tmp4, 1 ; <i32> [#uses=1] 19 %tmp10 = add i32 %tmp9, %tmp8 ; <i32> [#uses=2] 20 store i32 %tmp10, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 1) 21 %tmp8.1 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 1) ; <i32> [#uses=1] 22 %tmp9.1 = add i32 %tmp10, 1 ; <i32> [#uses=1] 23 %tmp10.1 = add i32 %tmp9.1, %tmp8.1 ; <i32> [#uses=2] 24 store i32 %tmp10.1, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 2) 25 %tmp8.2 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 2) ; <i32> [#uses=1] 26 %tmp9.2 = add i32 %tmp10.1, 1 ; <i32> [#uses=1] 27 %tmp10.2 = add i32 %tmp9.2, %tmp8.2 ; <i32> [#uses=2] 28 store i32 %tmp10.2, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 3) 29 %tmp8.3 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 3) ; <i32> [#uses=1] 30 %tmp9.3 = add i32 %tmp10.2, 1 ; <i32> [#uses=1] 31 %tmp10.3 = add i32 %tmp9.3, %tmp8.3 ; <i32> [#uses=2] 32 store i32 %tmp10.3, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 4) 33 %tmp8.4 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 4) ; <i32> [#uses=1] 34 %tmp9.4 = add i32 %tmp10.3, 1 ; <i32> [#uses=1] 35 %tmp10.4 = add i32 %tmp9.4, %tmp8.4 ; <i32> [#uses=2] 36 store i32 %tmp10.4, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 5) 37 %tmp8.5 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 5) ; <i32> [#uses=1] 38 %tmp9.5 = add i32 %tmp10.4, 1 ; <i32> [#uses=1] 39 %tmp10.5 = add i32 %tmp9.5, %tmp8.5 ; <i32> [#uses=2] 40 store i32 %tmp10.5, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 6) 41 %tmp8.6 = load i32* getelementptr ([1000 x i32]* @g1, i32 0, i32 6) ; <i32> [#uses=1] 42 %tmp9.6 = add i32 %tmp10.5, 1 ; <i32> [#uses=1] 43 %tmp10.6 = add i32 %tmp9.6, %tmp8.6 ; <i32> [#uses=1] 44 store i32 %tmp10.6, i32* getelementptr ([1000 x i32]* @g0, i32 0, i32 7) 45 ret void 46 } 47 48
