Comments (5)
wsmoses
commented on August 16, 2024
Can you paste the gemm.mlir output in both case?
from polygeist.
ThanHenderson
commented on August 16, 2024
Function specified (post mlir-clang):
module attributes {llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} {
func @kernel_gemm(%arg0: i32, %arg1: i32, %arg2: i32, %arg3: f64, %arg4: f64, %arg5: memref<1000x1100xf64>, %arg6: memref<1000x1200xf64>, %arg7: memref<1200x1100xf64>) attributes {llvm.linkage = #llvm.linkage<external>} {
affine.for %arg8 = 0 to 1000 {
affine.for %arg9 = 0 to 1100 {
%0 = affine.load %arg5[%arg8, %arg9] : memref<1000x1100xf64>
%1 = arith.mulf %0, %arg4 : f64
affine.store %1, %arg5[%arg8, %arg9] : memref<1000x1100xf64>
}
affine.for %arg9 = 0 to 1200 {
affine.for %arg10 = 0 to 1100 {
%0 = affine.load %arg6[%arg8, %arg9] : memref<1000x1200xf64>
%1 = arith.mulf %arg3, %0 : f64
%2 = affine.load %arg7[%arg9, %arg10] : memref<1200x1100xf64>
%3 = arith.mulf %1, %2 : f64
%4 = affine.load %arg5[%arg8, %arg10] : memref<1000x1100xf64>
%5 = arith.addf %4, %3 : f64
affine.store %5, %arg5[%arg8, %arg10] : memref<1000x1100xf64>
}
}
}
return
}
}
Full file (post mlir-clang):
module attributes {llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} {
llvm.mlir.global internal constant @str7("==END DUMP_ARRAYS==\0A\00")
llvm.mlir.global internal constant @str6("\0Aend dump: %s\0A\00")
llvm.mlir.global internal constant @str5("%0.2lf \00")
llvm.mlir.global internal constant @str4("\0A\00")
llvm.mlir.global internal constant @str3("C\00")
llvm.mlir.global internal constant @str2("begin dump: %s\00")
llvm.func @fprintf(!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, ...) -> i32
llvm.mlir.global internal constant @str1("==BEGIN DUMP_ARRAYS==\0A\00")
llvm.mlir.global external @stderr() : !llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>
llvm.func @strcmp(!llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
llvm.mlir.global internal constant @str0("\00")
func @main(%arg0: i32, %arg1: !llvm.ptr<ptr<i8>>) -> i32 attributes {llvm.linkage = #llvm.linkage<external>} {
%cst = arith.constant 1.100000e+03 : f64
%c1200 = arith.constant 1200 : index
%cst_0 = arith.constant 1.200000e+03 : f64
%cst_1 = arith.constant 1.000000e+03 : f64
%c1100 = arith.constant 1100 : index
%c1000 = arith.constant 1000 : index
%cst_2 = arith.constant 1.500000e+00 : f64
%cst_3 = arith.constant 1.200000e+00 : f64
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c1_i32 = arith.constant 1 : i32
%c2_i32 = arith.constant 2 : i32
%c1000_i32 = arith.constant 1000 : i32
%c1100_i32 = arith.constant 1100 : i32
%c1200_i32 = arith.constant 1200 : i32
%c42_i32 = arith.constant 42 : i32
%c0_i32 = arith.constant 0 : i32
%0 = memref.alloca() : memref<1xf64>
%1 = llvm.mlir.undef : f64
affine.store %1, %0[0] : memref<1xf64>
%2 = memref.alloca() : memref<1xf64>
affine.store %1, %2[0] : memref<1xf64>
%3 = memref.alloc() : memref<1000x1100xf64>
%4 = memref.alloc() : memref<1000x1200xf64>
%5 = memref.alloc() : memref<1200x1100xf64>
affine.store %cst_2, %2[0] : memref<1xf64>
affine.store %cst_3, %0[0] : memref<1xf64>
scf.for %arg2 = %c0 to %c1000 step %c1 {
%9 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1100 step %c1 {
%10 = arith.index_cast %arg3 : index to i32
%11 = arith.muli %9, %10 : i32
%12 = arith.remsi %11, %c1000_i32 : i32
%13 = arith.sitofp %12 : i32 to f64
%14 = arith.divf %13, %cst_1 : f64
memref.store %14, %3[%arg2, %arg3] : memref<1000x1100xf64>
}
}
scf.for %arg2 = %c0 to %c1000 step %c1 {
%9 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1200 step %c1 {
%10 = arith.index_cast %arg3 : index to i32
%11 = arith.addi %10, %c1_i32 : i32
%12 = arith.muli %9, %11 : i32
%13 = arith.remsi %12, %c1200_i32 : i32
%14 = arith.sitofp %13 : i32 to f64
%15 = arith.divf %14, %cst_0 : f64
memref.store %15, %4[%arg2, %arg3] : memref<1000x1200xf64>
}
}
scf.for %arg2 = %c0 to %c1200 step %c1 {
%9 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1100 step %c1 {
%10 = arith.index_cast %arg3 : index to i32
%11 = arith.addi %10, %c2_i32 : i32
%12 = arith.muli %9, %11 : i32
%13 = arith.remsi %12, %c1100_i32 : i32
%14 = arith.sitofp %13 : i32 to f64
%15 = arith.divf %14, %cst : f64
memref.store %15, %5[%arg2, %arg3] : memref<1200x1100xf64>
}
}
%6 = affine.load %2[0] : memref<1xf64>
%7 = affine.load %0[0] : memref<1xf64>
affine.for %arg2 = 0 to 1000 {
affine.for %arg3 = 0 to 1100 {
%9 = affine.load %3[%arg2, %arg3] : memref<1000x1100xf64>
%10 = arith.mulf %9, %7 : f64
affine.store %10, %3[%arg2, %arg3] : memref<1000x1100xf64>
}
affine.for %arg3 = 0 to 1200 {
affine.for %arg4 = 0 to 1100 {
%9 = affine.load %4[%arg2, %arg3] : memref<1000x1200xf64>
%10 = arith.mulf %6, %9 : f64
%11 = affine.load %5[%arg3, %arg4] : memref<1200x1100xf64>
%12 = arith.mulf %10, %11 : f64
%13 = affine.load %3[%arg2, %arg4] : memref<1000x1100xf64>
%14 = arith.addf %13, %12 : f64
affine.store %14, %3[%arg2, %arg4] : memref<1000x1100xf64>
}
}
}
%8 = arith.cmpi sgt, %arg0, %c42_i32 : i32
scf.if %8 {
%9 = llvm.load %arg1 : !llvm.ptr<ptr<i8>>
%10 = llvm.mlir.addressof @str0 : !llvm.ptr<array<1 x i8>>
%11 = llvm.getelementptr %10[%c0_i32, %c0_i32] : (!llvm.ptr<array<1 x i8>>, i32, i32) -> !llvm.ptr<i8>
%12 = llvm.call @strcmp(%9, %11) : (!llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%13 = arith.cmpi eq, %12, %c0_i32 : i32
scf.if %13 {
call @print_array(%c1000_i32, %c1100_i32, %3) : (i32, i32, memref<1000x1100xf64>) -> ()
}
}
memref.dealloc %3 : memref<1000x1100xf64>
memref.dealloc %4 : memref<1000x1200xf64>
memref.dealloc %5 : memref<1200x1100xf64>
return %c0_i32 : i32
}
func @kernel_gemm(%arg0: i32, %arg1: i32, %arg2: i32, %arg3: f64, %arg4: f64, %arg5: memref<1000x1100xf64>, %arg6: memref<1000x1200xf64>, %arg7: memref<1200x1100xf64>) attributes {llvm.linkage = #llvm.linkage<external>} {
affine.for %arg8 = 0 to 1000 {
affine.for %arg9 = 0 to 1100 {
%0 = affine.load %arg5[%arg8, %arg9] : memref<1000x1100xf64>
%1 = arith.mulf %0, %arg4 : f64
affine.store %1, %arg5[%arg8, %arg9] : memref<1000x1100xf64>
}
affine.for %arg9 = 0 to 1200 {
affine.for %arg10 = 0 to 1100 {
%0 = affine.load %arg6[%arg8, %arg9] : memref<1000x1200xf64>
%1 = arith.mulf %arg3, %0 : f64
%2 = affine.load %arg7[%arg9, %arg10] : memref<1200x1100xf64>
%3 = arith.mulf %1, %2 : f64
%4 = affine.load %arg5[%arg8, %arg10] : memref<1000x1100xf64>
%5 = arith.addf %4, %3 : f64
affine.store %5, %arg5[%arg8, %arg10] : memref<1000x1100xf64>
}
}
}
return
}
func private @print_array(%arg0: i32, %arg1: i32, %arg2: memref<1000x1100xf64>) attributes {llvm.linkage = #llvm.linkage<internal>} {
%c1 = arith.constant 1 : index
%c0 = arith.constant 0 : index
%c20_i32 = arith.constant 20 : i32
%c0_i32 = arith.constant 0 : i32
%0 = llvm.mlir.addressof @stderr : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%1 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%2 = llvm.mlir.addressof @str1 : !llvm.ptr<array<23 x i8>>
%3 = llvm.getelementptr %2[%c0_i32, %c0_i32] : (!llvm.ptr<array<23 x i8>>, i32, i32) -> !llvm.ptr<i8>
%4 = llvm.call @fprintf(%1, %3) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
%5 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%6 = llvm.mlir.addressof @str2 : !llvm.ptr<array<15 x i8>>
%7 = llvm.getelementptr %6[%c0_i32, %c0_i32] : (!llvm.ptr<array<15 x i8>>, i32, i32) -> !llvm.ptr<i8>
%8 = llvm.mlir.addressof @str3 : !llvm.ptr<array<2 x i8>>
%9 = llvm.getelementptr %8[%c0_i32, %c0_i32] : (!llvm.ptr<array<2 x i8>>, i32, i32) -> !llvm.ptr<i8>
%10 = llvm.call @fprintf(%5, %7, %9) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%11 = arith.index_cast %arg0 : i32 to index
scf.for %arg3 = %c0 to %11 step %c1 {
%20 = arith.index_cast %arg3 : index to i32
%21 = arith.index_cast %arg1 : i32 to index
scf.for %arg4 = %c0 to %21 step %c1 {
%22 = arith.index_cast %arg4 : index to i32
%23 = arith.muli %20, %arg0 : i32
%24 = arith.addi %23, %22 : i32
%25 = arith.remsi %24, %c20_i32 : i32
%26 = arith.cmpi eq, %25, %c0_i32 : i32
scf.if %26 {
%32 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%33 = llvm.mlir.addressof @str4 : !llvm.ptr<array<2 x i8>>
%34 = llvm.getelementptr %33[%c0_i32, %c0_i32] : (!llvm.ptr<array<2 x i8>>, i32, i32) -> !llvm.ptr<i8>
%35 = llvm.call @fprintf(%32, %34) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
}
%27 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%28 = llvm.mlir.addressof @str5 : !llvm.ptr<array<8 x i8>>
%29 = llvm.getelementptr %28[%c0_i32, %c0_i32] : (!llvm.ptr<array<8 x i8>>, i32, i32) -> !llvm.ptr<i8>
%30 = memref.load %arg2[%arg3, %arg4] : memref<1000x1100xf64>
%31 = llvm.call @fprintf(%27, %29, %30) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, f64) -> i32
}
}
%12 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%13 = llvm.mlir.addressof @str6 : !llvm.ptr<array<17 x i8>>
%14 = llvm.getelementptr %13[%c0_i32, %c0_i32] : (!llvm.ptr<array<17 x i8>>, i32, i32) -> !llvm.ptr<i8>
%15 = llvm.call @fprintf(%12, %14, %9) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%16 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%17 = llvm.mlir.addressof @str7 : !llvm.ptr<array<23 x i8>>
%18 = llvm.getelementptr %17[%c0_i32, %c0_i32] : (!llvm.ptr<array<23 x i8>>, i32, i32) -> !llvm.ptr<i8>
%19 = llvm.call @fprintf(%16, %18) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
return
}
}
And here is after running polymer-opt on the full file MLIR with --extract-scop-stmt (this is what then goes to --pluto-opt at which point it fails):
module attributes {llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", llvm.target_triple = "x86_64-unknown-linux-gnu"} {
llvm.mlir.global internal constant @str7("==END DUMP_ARRAYS==\0A\00")
llvm.mlir.global internal constant @str6("\0Aend dump: %s\0A\00")
llvm.mlir.global internal constant @str5("%0.2lf \00")
llvm.mlir.global internal constant @str4("\0A\00")
llvm.mlir.global internal constant @str3("C\00")
llvm.mlir.global internal constant @str2("begin dump: %s\00")
llvm.func @fprintf(!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, ...) -> i32
llvm.mlir.global internal constant @str1("==BEGIN DUMP_ARRAYS==\0A\00")
llvm.mlir.global external @stderr() : !llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>
llvm.func @strcmp(!llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
llvm.mlir.global internal constant @str0("\00")
func @main(%arg0: i32, %arg1: !llvm.ptr<ptr<i8>>) -> i32 attributes {llvm.linkage = #llvm.linkage<external>} {
%c0_i32 = arith.constant 0 : i32
%c42_i32 = arith.constant 42 : i32
%c1200_i32 = arith.constant 1200 : i32
%c1100_i32 = arith.constant 1100 : i32
%c1000_i32 = arith.constant 1000 : i32
%c2_i32 = arith.constant 2 : i32
%c1_i32 = arith.constant 1 : i32
%c1 = arith.constant 1 : index
%c0 = arith.constant 0 : index
%c1000 = arith.constant 1000 : index
%c1100 = arith.constant 1100 : index
%cst = arith.constant 1.000000e+03 : f64
%cst_0 = arith.constant 1.200000e+03 : f64
%c1200 = arith.constant 1200 : index
%cst_1 = arith.constant 1.100000e+03 : f64
%0 = memref.alloca() : memref<1xf64>
call @S0(%0) : (memref<1xf64>) -> ()
%1 = memref.alloca() : memref<1xf64>
call @S1(%1) : (memref<1xf64>) -> ()
%2 = memref.alloc() : memref<1000x1100xf64>
%3 = memref.alloc() : memref<1000x1200xf64>
%4 = memref.alloc() : memref<1200x1100xf64>
call @S2(%1) : (memref<1xf64>) -> ()
call @S3(%0) : (memref<1xf64>) -> ()
scf.for %arg2 = %c0 to %c1000 step %c1 {
%6 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1100 step %c1 {
%7 = arith.index_cast %arg3 : index to i32
%8 = arith.muli %6, %7 : i32
%9 = arith.remsi %8, %c1000_i32 : i32
%10 = arith.sitofp %9 : i32 to f64
%11 = arith.divf %10, %cst : f64
memref.store %11, %2[%arg2, %arg3] : memref<1000x1100xf64>
}
}
scf.for %arg2 = %c0 to %c1000 step %c1 {
%6 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1200 step %c1 {
%7 = arith.index_cast %arg3 : index to i32
%8 = arith.addi %7, %c1_i32 : i32
%9 = arith.muli %6, %8 : i32
%10 = arith.remsi %9, %c1200_i32 : i32
%11 = arith.sitofp %10 : i32 to f64
%12 = arith.divf %11, %cst_0 : f64
memref.store %12, %3[%arg2, %arg3] : memref<1000x1200xf64>
}
}
scf.for %arg2 = %c0 to %c1200 step %c1 {
%6 = arith.index_cast %arg2 : index to i32
scf.for %arg3 = %c0 to %c1100 step %c1 {
%7 = arith.index_cast %arg3 : index to i32
%8 = arith.addi %7, %c2_i32 : i32
%9 = arith.muli %6, %8 : i32
%10 = arith.remsi %9, %c1100_i32 : i32
%11 = arith.sitofp %10 : i32 to f64
%12 = arith.divf %11, %cst_1 : f64
memref.store %12, %4[%arg2, %arg3] : memref<1200x1100xf64>
}
}
affine.for %arg2 = 0 to 1000 {
affine.for %arg3 = 0 to 1100 {
call @S4(%arg2, %arg3, %2, %0) : (index, index, memref<1000x1100xf64>, memref<1xf64>) -> ()
}
affine.for %arg3 = 0 to 1200 {
affine.for %arg4 = 0 to 1100 {
call @S5(%arg2, %arg4, %2, %arg3, %4, %3, %1) : (index, index, memref<1000x1100xf64>, index, memref<1200x1100xf64>, memref<1000x1200xf64>, memref<1xf64>) -> ()
}
}
}
%5 = arith.cmpi sgt, %arg0, %c42_i32 : i32
scf.if %5 {
%6 = llvm.load %arg1 : !llvm.ptr<ptr<i8>>
%7 = llvm.mlir.addressof @str0 : !llvm.ptr<array<1 x i8>>
%8 = llvm.getelementptr %7[%c0_i32, %c0_i32] : (!llvm.ptr<array<1 x i8>>, i32, i32) -> !llvm.ptr<i8>
%9 = llvm.call @strcmp(%6, %8) : (!llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%10 = arith.cmpi eq, %9, %c0_i32 : i32
scf.if %10 {
call @print_array(%c1000_i32, %c1100_i32, %2) : (i32, i32, memref<1000x1100xf64>) -> ()
}
}
memref.dealloc %2 : memref<1000x1100xf64>
memref.dealloc %3 : memref<1000x1200xf64>
memref.dealloc %4 : memref<1200x1100xf64>
return %c0_i32 : i32
}
func @kernel_gemm(%arg0: i32, %arg1: i32, %arg2: i32, %arg3: f64, %arg4: f64, %arg5: memref<1000x1100xf64>, %arg6: memref<1000x1200xf64>, %arg7: memref<1200x1100xf64>) attributes {llvm.linkage = #llvm.linkage<external>} {
affine.for %arg8 = 0 to 1000 {
affine.for %arg9 = 0 to 1100 {
call @S6(%arg5, %arg8, %arg9, %arg4) : (memref<1000x1100xf64>, index, index, f64) -> ()
}
affine.for %arg9 = 0 to 1200 {
affine.for %arg10 = 0 to 1100 {
call @S7(%arg5, %arg8, %arg10, %arg7, %arg9, %arg3, %arg6) : (memref<1000x1100xf64>, index, index, memref<1200x1100xf64>, index, f64, memref<1000x1200xf64>) -> ()
}
}
}
return
}
func private @S0(%arg0: memref<1xf64>) attributes {scop.stmt} {
%0 = llvm.mlir.undef : f64
affine.store %0, %arg0[0] : memref<1xf64>
return
}
func private @S1(%arg0: memref<1xf64>) attributes {scop.stmt} {
%0 = llvm.mlir.undef : f64
affine.store %0, %arg0[0] : memref<1xf64>
return
}
func private @S2(%arg0: memref<1xf64>) attributes {scop.stmt} {
%cst = arith.constant 1.500000e+00 : f64
affine.store %cst, %arg0[0] : memref<1xf64>
return
}
func private @S3(%arg0: memref<1xf64>) attributes {scop.stmt} {
%cst = arith.constant 1.200000e+00 : f64
affine.store %cst, %arg0[0] : memref<1xf64>
return
}
func private @S4(%arg0: index, %arg1: index, %arg2: memref<1000x1100xf64>, %arg3: memref<1xf64>) attributes {scop.stmt} {
%0 = affine.load %arg2[symbol(%arg0), symbol(%arg1)] : memref<1000x1100xf64>
%1 = affine.load %arg3[0] : memref<1xf64>
%2 = arith.mulf %0, %1 : f64
affine.store %2, %arg2[symbol(%arg0), symbol(%arg1)] : memref<1000x1100xf64>
return
}
func private @S5(%arg0: index, %arg1: index, %arg2: memref<1000x1100xf64>, %arg3: index, %arg4: memref<1200x1100xf64>, %arg5: memref<1000x1200xf64>, %arg6: memref<1xf64>) attributes {scop.stmt} {
%0 = affine.load %arg2[symbol(%arg0), symbol(%arg1)] : memref<1000x1100xf64>
%1 = affine.load %arg6[0] : memref<1xf64>
%2 = affine.load %arg5[symbol(%arg0), symbol(%arg3)] : memref<1000x1200xf64>
%3 = arith.mulf %1, %2 : f64
%4 = affine.load %arg4[symbol(%arg3), symbol(%arg1)] : memref<1200x1100xf64>
%5 = arith.mulf %3, %4 : f64
%6 = arith.addf %0, %5 : f64
affine.store %6, %arg2[symbol(%arg0), symbol(%arg1)] : memref<1000x1100xf64>
return
}
func private @S6(%arg0: memref<1000x1100xf64>, %arg1: index, %arg2: index, %arg3: f64) attributes {scop.stmt} {
%0 = affine.load %arg0[symbol(%arg1), symbol(%arg2)] : memref<1000x1100xf64>
%1 = arith.mulf %0, %arg3 : f64
affine.store %1, %arg0[symbol(%arg1), symbol(%arg2)] : memref<1000x1100xf64>
return
}
func private @S7(%arg0: memref<1000x1100xf64>, %arg1: index, %arg2: index, %arg3: memref<1200x1100xf64>, %arg4: index, %arg5: f64, %arg6: memref<1000x1200xf64>) attributes {scop.stmt} {
%0 = affine.load %arg0[symbol(%arg1), symbol(%arg2)] : memref<1000x1100xf64>
%1 = affine.load %arg6[symbol(%arg1), symbol(%arg4)] : memref<1000x1200xf64>
%2 = arith.mulf %arg5, %1 : f64
%3 = affine.load %arg3[symbol(%arg4), symbol(%arg2)] : memref<1200x1100xf64>
%4 = arith.mulf %2, %3 : f64
%5 = arith.addf %0, %4 : f64
affine.store %5, %arg0[symbol(%arg1), symbol(%arg2)] : memref<1000x1100xf64>
return
}
func private @print_array(%arg0: i32, %arg1: i32, %arg2: memref<1000x1100xf64>) attributes {llvm.linkage = #llvm.linkage<internal>} {
%c0_i32 = arith.constant 0 : i32
%c20_i32 = arith.constant 20 : i32
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%0 = llvm.mlir.addressof @stderr : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%1 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%2 = llvm.mlir.addressof @str1 : !llvm.ptr<array<23 x i8>>
%3 = llvm.getelementptr %2[%c0_i32, %c0_i32] : (!llvm.ptr<array<23 x i8>>, i32, i32) -> !llvm.ptr<i8>
%4 = llvm.call @fprintf(%1, %3) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
%5 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%6 = llvm.mlir.addressof @str2 : !llvm.ptr<array<15 x i8>>
%7 = llvm.getelementptr %6[%c0_i32, %c0_i32] : (!llvm.ptr<array<15 x i8>>, i32, i32) -> !llvm.ptr<i8>
%8 = llvm.mlir.addressof @str3 : !llvm.ptr<array<2 x i8>>
%9 = llvm.getelementptr %8[%c0_i32, %c0_i32] : (!llvm.ptr<array<2 x i8>>, i32, i32) -> !llvm.ptr<i8>
%10 = llvm.call @fprintf(%5, %7, %9) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%11 = arith.index_cast %arg0 : i32 to index
scf.for %arg3 = %c0 to %11 step %c1 {
%20 = arith.index_cast %arg3 : index to i32
%21 = arith.index_cast %arg1 : i32 to index
scf.for %arg4 = %c0 to %21 step %c1 {
%22 = arith.index_cast %arg4 : index to i32
%23 = arith.muli %20, %arg0 : i32
%24 = arith.addi %23, %22 : i32
%25 = arith.remsi %24, %c20_i32 : i32
%26 = arith.cmpi eq, %25, %c0_i32 : i32
scf.if %26 {
%32 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%33 = llvm.mlir.addressof @str4 : !llvm.ptr<array<2 x i8>>
%34 = llvm.getelementptr %33[%c0_i32, %c0_i32] : (!llvm.ptr<array<2 x i8>>, i32, i32) -> !llvm.ptr<i8>
%35 = llvm.call @fprintf(%32, %34) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
}
%27 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%28 = llvm.mlir.addressof @str5 : !llvm.ptr<array<8 x i8>>
%29 = llvm.getelementptr %28[%c0_i32, %c0_i32] : (!llvm.ptr<array<8 x i8>>, i32, i32) -> !llvm.ptr<i8>
%30 = memref.load %arg2[%arg3, %arg4] : memref<1000x1100xf64>
%31 = llvm.call @fprintf(%27, %29, %30) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, f64) -> i32
}
}
%12 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%13 = llvm.mlir.addressof @str6 : !llvm.ptr<array<17 x i8>>
%14 = llvm.getelementptr %13[%c0_i32, %c0_i32] : (!llvm.ptr<array<17 x i8>>, i32, i32) -> !llvm.ptr<i8>
%15 = llvm.call @fprintf(%12, %14, %9) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>, !llvm.ptr<i8>) -> i32
%16 = llvm.load %0 : !llvm.ptr<ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>>
%17 = llvm.mlir.addressof @str7 : !llvm.ptr<array<23 x i8>>
%18 = llvm.getelementptr %17[%c0_i32, %c0_i32] : (!llvm.ptr<array<23 x i8>>, i32, i32) -> !llvm.ptr<i8>
%19 = llvm.call @fprintf(%16, %18) : (!llvm.ptr<struct<(i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<()>>, ptr<struct<()>>, ptr<struct<"struct._IO_FILE", (i32, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<i8>, ptr<struct<"struct._IO_marker", opaque>>, ptr<struct<"struct._IO_FILE">>, i32, i32, i64, i16, i8, array<1 x i8>, ptr<i8>, i64, ptr<struct<"struct._IO_codecvt", opaque>>, ptr<struct<"struct._IO_wide_data", opaque>>, ptr<struct<"struct._IO_FILE">>, ptr<i8>, i64, i32, array<20 x i8>)>>, ptr<i8>, i64, i32, array<20 x i8>)>>, !llvm.ptr<i8>) -> i32
return
}
}
from polygeist.
wsmoses
commented on August 16, 2024
The output here is definitely valid (even running through mlir-opt yields no complaints).
This appears to be a bug in polymer and/or a weird MLIR version mismatch?
from polygeist.
kumasento
commented on August 16, 2024
I think Polymer currently doesn't support detecting which function to
transform. So it would be better to specify the target function.
…On Thu, 24 Feb 2022 at 04:36, William Moses ***@***.***> wrote:
The output here is definitely valid (even running through mlir-opt yields
no complaints).
This appears to be a bug in polymer and/or a weird MLIR version mismatch?
@kumasento <https://github.com/kumasento>
—
Reply to this email directly, view it on GitHub
<#169 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACC42R3RJTVQGLGI7CXADWTU4WYWHANCNFSM5OTTUOFA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
from polygeist.
ThanHenderson
commented on August 16, 2024
@kumasento The interesting thing is that -pluto-opt works (does something but obviously not the desired output) when we just remove the calls to @s# functions from main.
Just to be clear then: there is no way currently to go from a C file with multiple functions -> Polygeist -> Polymer -> executable without specifying specific function in Polygeist first? Or were you saying to specify a target function within Polymer? i.e. set Polygeist not to inline then specify the kernel function through Polymer somehow.
from polygeist.
Related Issues (20)
- Cgeist outputs something strange HOT 6
- How to run PolyBench/C benchmarks with polyhedral pluto optimization using Polygeist? HOT 3
- [Bug] Incorrect scf.for code is generated
- Sign extension not properly propagated to users HOT 2
- declared char constants emit higher bit-width MLIR constants HOT 3
- linker error with cgeist on C++ code HOT 2
- make[1]: *** No rule to make target 'install'. Stop. make: *** [Makefile:555: install-recursive] Error 1 CMake Error at tools/polymer/CMakeLists.txt:28 (execute_process): execute_process failed command indexes: 1: "Child return code: 2" HOT 1
- support for Cuda 12.1 HOT 11
- Incorrect execution result of running CUDA to OpenMP compilation? HOT 1
- Error compiling code with Polygeist: 'affine.for' op operand cannot be used as a symbol
- [bug] Operations such as "|=", "&=" and "^=" lack necessary implicit type conversions
- opaque attribute handling issues
- error when trying to lower .cu file, should I report this here or on LLVM repo issues?
- Problems about fp_16 in CUDA， as half, __hmul.......etc HOT 1
- Is it possible to take MLIR generated by CUDA-to-CPU flow to the cpu-openmp executable? HOT 1
- Add Support for Additional Math Operations in Polygeist
- gitignore bug
- Polymer with -reg2mem -pluto-opt generates IR with out-of-bounds accesses for imperfectly nested loops
- When a class is deleted, the user-defined destructor is not invoked. HOT 3
- Runtime error after compiling CUDA programs using cgeist HOT 2
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from polygeist.