101 lines · plain
1// RUN: mlir-runner %s -e entry -entry-point-result=void \2// RUN: -shared-libs=%mlir_c_runner_utils | \3// RUN: FileCheck %s4 5// End-to-end test of all fp reduction intrinsics (not exhaustive unit tests).6module {7 llvm.func @printNewline()8 llvm.func @printF32(f32)9 llvm.func @entry() {10 // Setup (1,2,3,4).11 %0 = llvm.mlir.constant(1.000000e+00 : f32) : f3212 %1 = llvm.mlir.constant(2.000000e+00 : f32) : f3213 %2 = llvm.mlir.constant(3.000000e+00 : f32) : f3214 %3 = llvm.mlir.constant(4.000000e+00 : f32) : f3215 %4 = llvm.mlir.undef : vector<4xf32>16 %5 = llvm.mlir.constant(0 : index) : i6417 %6 = llvm.insertelement %0, %4[%5 : i64] : vector<4xf32>18 %7 = llvm.shufflevector %6, %4 [0, 0, 0, 0] : vector<4xf32>19 %8 = llvm.mlir.constant(1 : i64) : i6420 %9 = llvm.insertelement %1, %7[%8 : i64] : vector<4xf32>21 %10 = llvm.mlir.constant(2 : i64) : i6422 %11 = llvm.insertelement %2, %9[%10 : i64] : vector<4xf32>23 %12 = llvm.mlir.constant(3 : i64) : i6424 %v = llvm.insertelement %3, %11[%12 : i64] : vector<4xf32>25 26 %max = llvm.intr.vector.reduce.fmax(%v)27 : (vector<4xf32>) -> f3228 llvm.call @printF32(%max) : (f32) -> ()29 llvm.call @printNewline() : () -> ()30 // CHECK: 431 32 %min = llvm.intr.vector.reduce.fmin(%v)33 : (vector<4xf32>) -> f3234 llvm.call @printF32(%min) : (f32) -> ()35 llvm.call @printNewline() : () -> ()36 // CHECK: 137 38 %maximum = llvm.intr.vector.reduce.fmaximum(%v)39 : (vector<4xf32>) -> f3240 llvm.call @printF32(%maximum) : (f32) -> ()41 llvm.call @printNewline() : () -> ()42 // CHECK: 443 44 %minimum = llvm.intr.vector.reduce.fminimum(%v)45 : (vector<4xf32>) -> f3246 llvm.call @printF32(%minimum) : (f32) -> ()47 llvm.call @printNewline() : () -> ()48 // CHECK: 149 50 %add1 = "llvm.intr.vector.reduce.fadd"(%0, %v)51 : (f32, vector<4xf32>) -> f3252 llvm.call @printF32(%add1) : (f32) -> ()53 llvm.call @printNewline() : () -> ()54 // CHECK: 1155 56 %add1r = "llvm.intr.vector.reduce.fadd"(%0, %v)57 {reassoc = true} : (f32, vector<4xf32>) -> f3258 llvm.call @printF32(%add1r) : (f32) -> ()59 llvm.call @printNewline() : () -> ()60 // CHECK: 1161 62 %add2 = "llvm.intr.vector.reduce.fadd"(%1, %v)63 : (f32, vector<4xf32>) -> f3264 llvm.call @printF32(%add2) : (f32) -> ()65 llvm.call @printNewline() : () -> ()66 // CHECK: 1267 68 %add2r = "llvm.intr.vector.reduce.fadd"(%1, %v)69 {reassoc = true} : (f32, vector<4xf32>) -> f3270 llvm.call @printF32(%add2r) : (f32) -> ()71 llvm.call @printNewline() : () -> ()72 // CHECK: 1273 74 %mul1 = "llvm.intr.vector.reduce.fmul"(%0, %v)75 : (f32, vector<4xf32>) -> f3276 llvm.call @printF32(%mul1) : (f32) -> ()77 llvm.call @printNewline() : () -> ()78 // CHECK: 2479 80 %mul1r = "llvm.intr.vector.reduce.fmul"(%0, %v)81 {reassoc = true} : (f32, vector<4xf32>) -> f3282 llvm.call @printF32(%mul1r) : (f32) -> ()83 llvm.call @printNewline() : () -> ()84 // CHECK: 2485 86 %mul2 = "llvm.intr.vector.reduce.fmul"(%1, %v)87 : (f32, vector<4xf32>) -> f3288 llvm.call @printF32(%mul2) : (f32) -> ()89 llvm.call @printNewline() : () -> ()90 // CHECK: 4891 92 %mul2r = "llvm.intr.vector.reduce.fmul"(%1, %v)93 {reassoc = true} : (f32, vector<4xf32>) -> f3294 llvm.call @printF32(%mul2r) : (f32) -> ()95 llvm.call @printNewline() : () -> ()96 // CHECK: 4897 98 llvm.return99 }100}101