[CIR] Added codegen support for builtin_mempcpy

[badumbatish]

Related to #1817

[badumbatish]

Input is

char* a(void *A, void *B, long x) {
  return __builtin_mempcpy(A, B, x);
}

CIR is

 ninja -C build && ./build/bin/clang --target=x86_64-unknown-linux  -S scratch/builtin_mempcpy.c  -o - -emit-llvm -g --sysroot=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX15.4.sdk  -emit-cir
ninja: Entering directory `build'
[126/126] Linking CXX shared library lib/libclang-cpp.dylib

module @"/Users/jjasmine/Developer/igalia/clangir/scratch/builtin_mempcpy.c" attributes {cir.lang = #cir.lang<c>, cir.sob = #cir.signed_overflow_behavior<undefined>, cir.triple = "x86_64-unknown-linux", cir.type_size_info = #cir.type_size_info<char = 8, int = 32, size_t = 64>, cir.uwtable = #cir.uwtable<async>, dlti.dl_spec = #dlti.dl_spec<!llvm.ptr<270> = dense<32> : vector<4xi64>, !llvm.ptr<271> = dense<32> : vector<4xi64>, !llvm.ptr<272> = dense<64> : vector<4xi64>, i64 = dense<64> : vector<2xi64>, i128 = dense<128> : vector<2xi64>, f80 = dense<128> : vector<2xi64>, !llvm.ptr = dense<64> : vector<4xi64>, i1 = dense<8> : vector<2xi64>, i8 = dense<8> : vector<2xi64>, i16 = dense<16> : vector<2xi64>, i32 = dense<32> : vector<2xi64>, f16 = dense<16> : vector<2xi64>, f64 = dense<64> : vector<2xi64>, f128 = dense<128> : vector<2xi64>, "dlti.endianness" = "little", "dlti.mangling_mode" = "e", "dlti.legal_int_widths" = array<i32: 8, 16, 32, 64>, "dlti.stack_alignment" = 128 : i64>} {
  cir.func dso_local @a(%arg0: !cir.ptr<!void> loc(fused[#loc3, #loc4]), %arg1: !cir.ptr<!void> loc(fused[#loc5, #loc6]), %arg2: !s64i loc(fused[#loc7, #loc8])) -> !cir.ptr<!s8i> extra(#fn_attr) {
    %0 = cir.alloca !cir.ptr<!void>, !cir.ptr<!cir.ptr<!void>>, ["A", init] {alignment = 8 : i64} loc(#loc17)
    %1 = cir.alloca !cir.ptr<!void>, !cir.ptr<!cir.ptr<!void>>, ["B", init] {alignment = 8 : i64} loc(#loc18)
    %2 = cir.alloca !s64i, !cir.ptr<!s64i>, ["x", init] {alignment = 8 : i64} loc(#loc19)
    %3 = cir.alloca !cir.ptr<!s8i>, !cir.ptr<!cir.ptr<!s8i>>, ["__retval"] {alignment = 8 : i64} loc(#loc2)
    cir.store %arg0, %0 : !cir.ptr<!void>, !cir.ptr<!cir.ptr<!void>> loc(#loc9)
    cir.store %arg1, %1 : !cir.ptr<!void>, !cir.ptr<!cir.ptr<!void>> loc(#loc9)
    cir.store %arg2, %2 : !s64i, !cir.ptr<!s64i> loc(#loc9)
    %4 = cir.load align(8) %0 : !cir.ptr<!cir.ptr<!void>>, !cir.ptr<!void> loc(#loc10)
    %5 = cir.load align(8) %1 : !cir.ptr<!cir.ptr<!void>>, !cir.ptr<!void> loc(#loc11)
    %6 = cir.load align(8) %2 : !cir.ptr<!s64i>, !s64i loc(#loc12)
    %7 = cir.cast(integral, %6 : !s64i), !u64i loc(#loc12)
    cir.libc.memcpy %7 bytes from %5 to %4 : !u64i, !cir.ptr<!void> -> !cir.ptr<!void> loc(#loc20)
    %8 = cir.cast(bitcast, %4 : !cir.ptr<!void>), !cir.ptr<!cir.ptr<!u8i>> loc(#loc10)
    %9 = cir.ptr_stride(%8 : !cir.ptr<!cir.ptr<!u8i>>, %7 : !u64i), !cir.ptr<!cir.ptr<!u8i>> loc(#loc20)
    %10 = cir.cast(bitcast, %9 : !cir.ptr<!cir.ptr<!u8i>>), !cir.ptr<!s8i> loc(#loc20)
    cir.store %10, %3 : !cir.ptr<!s8i>, !cir.ptr<!cir.ptr<!s8i>> loc(#loc21)
    %11 = cir.load %3 : !cir.ptr<!cir.ptr<!s8i>>, !cir.ptr<!s8i> loc(#loc21)
    cir.return %11 : !cir.ptr<!s8i> loc(#loc21)
  } loc(#loc16)
} loc(#loc)

LLVM clangir

 ninja -C build && ./build/bin/clang --target=x86_64-unknown-linux  -S scratch/builtin_mempcpy.c  -o - -emit-llvm -g --sysroot=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX15.4.sdk -fclangir | ./build/bin/opt -S -passes=instcombine,mem2reg,simplifycfg -o -
ninja: Entering directory `build'
ninja: no work to do.

define dso_local ptr @a(ptr %0, ptr %1, i64 %2) #0 !dbg !4 {
  %4 = alloca ptr, i64 1, align 8, !dbg !8
  %5 = alloca ptr, i64 1, align 8, !dbg !8
  %6 = alloca i64, i64 1, align 8, !dbg !8
  %7 = alloca ptr, i64 1, align 8, !dbg !9
  store ptr %0, ptr %4, align 8, !dbg !10
  store ptr %1, ptr %5, align 8, !dbg !10
  store i64 %2, ptr %6, align 8, !dbg !10
  %8 = load ptr, ptr %4, align 8, !dbg !11
  %9 = load ptr, ptr %5, align 8, !dbg !12
  %10 = load i64, ptr %6, align 8, !dbg !13
  call void @llvm.memcpy.p0.p0.i64(ptr %8, ptr %9, i64 %10, i1 false), !dbg !14
  %11 = getelementptr ptr, ptr %8, i64 %10, !dbg !14
  store ptr %11, ptr %7, align 8, !dbg !14
  %12 = load ptr, ptr %7, align 8, !dbg !14
  ret ptr %12, !dbg !14
}

LLVM clang

define dso_local ptr @a(ptr noundef %A, ptr noundef %B, i64 noundef %x) #0 !dbg !10 {
entry:
  %A.addr = alloca ptr, align 8
  %B.addr = alloca ptr, align 8
  %x.addr = alloca i64, align 8
  store ptr %A, ptr %A.addr, align 8
    #dbg_declare(ptr %A.addr, !18, !DIExpression(), !19)
  store ptr %B, ptr %B.addr, align 8
    #dbg_declare(ptr %B.addr, !20, !DIExpression(), !21)
  store i64 %x, ptr %x.addr, align 8
    #dbg_declare(ptr %x.addr, !22, !DIExpression(), !23)
  %0 = load ptr, ptr %A.addr, align 8, !dbg !24
  %1 = load ptr, ptr %B.addr, align 8, !dbg !25
  %2 = load i64, ptr %x.addr, align 8, !dbg !26
  call void @llvm.memcpy.p0.p0.i64(ptr align 1 %0, ptr align 1 %1, i64 %2, i1 false), !dbg !27
  %3 = getelementptr inbounds i8, ptr %0, i64 %2, !dbg !27
  ret ptr %3, !dbg !28
}

[bcardosolopes]

Mostly good, few nits and should be ready!

[bcardosolopes]

Awesome, LGTM