Merge pull request #10295 from ifreund/prefetch

Implement @prefetch()

Author: andrewrk

doc/langref.html.in

@@ -8674,6 +8674,50 @@ test "@wasmMemoryGrow" {
       {#see_also|@ctz|@clz#}
       {#header_close#}
 
+      {#header_open|@prefetch#}
+      <pre>{#syntax#}@prefetch(ptr: anytype, comptime options: std.builtin.PrefetchOptions){#endsyntax#}</pre>
+      <p>
+      This builtin tells the compiler to emit a prefetch instruction if supported by the
+      target CPU. If the target CPU does not support the requested prefetch instruction,
+      this builtin is a noop. This function has no effect on the behavior of the program,
+      only on the performance characteristics.
+      </p>
+      <p>
+      The {#syntax#}ptr{#endsyntax#} argument may be any pointer type and determines the memory
+      address to prefetch. This function does not dereference the pointer, it is perfectly legal
+      to pass a pointer to invalid memory to this function and no illegal behavior will result.
+      </p>
+      <p>
+      The {#syntax#}options{#endsyntax#} argument is the following struct:
+      </p>
+      {#code_begin|syntax|builtin#}
+/// This data structure is used by the Zig language code generation and
+/// therefore must be kept in sync with the compiler implementation.
+pub const PrefetchOptions = struct {
+    /// Whether the prefetch should prepare for a read or a write.
+    rw: Rw = .read,
+    /// 0 means no temporal locality. That is, the data can be immediately
+    /// dropped from the cache after it is accessed.
+    ///
+    /// 3 means high temporal locality. That is, the data should be kept in
+    /// the cache as it is likely to be accessed again soon.
+    locality: u2 = 3,
+    /// The cache that the prefetch should be preformed on.
+    cache: Cache = .data,
+
+    pub const Rw = enum {
+        read,
+        write,
+    };
+
+    pub const Cache = enum {
+        instruction,
+        data,
+    };
+};
+      {#code_end#}
+      {#header_close#}
+
       {#header_open|@ptrCast#}
       <pre>{#syntax#}@ptrCast(comptime DestType: type, value: anytype) DestType{#endsyntax#}</pre>
       <p>

lib/std/builtin.zig

@@ -651,6 +651,31 @@ pub const CallOptions = struct {
     };
 };
 
+/// This data structure is used by the Zig language code generation and
+/// therefore must be kept in sync with the compiler implementation.
+pub const PrefetchOptions = struct {
+    /// Whether the prefetch should prepare for a read or a write.
+    rw: Rw = .read,
+    /// 0 means no temporal locality. That is, the data can be immediately
+    /// dropped from the cache after it is accessed.
+    ///
+    /// 3 means high temporal locality. That is, the data should be kept in
+    /// the cache as it is likely to be accessed again soon.
+    locality: u2 = 3,
+    /// The cache that the prefetch should be preformed on.
+    cache: Cache = .data,
+
+    pub const Rw = enum {
+        read,
+        write,
+    };
+
+    pub const Cache = enum {
+        instruction,
+        data,
+    };
+};
+
 /// This data structure is used by the Zig language code generation and
 /// therefore must be kept in sync with the compiler implementation.
 pub const ExportOptions = struct {

src/AstGen.zig

@@ -6739,7 +6739,6 @@ fn builtinCall(
         }
     }
 
-    // zig fmt: off
     switch (info.tag) {
         .import => {
             const node_tags = tree.nodes.items(.tag);
@@ -6768,7 +6767,7 @@ fn builtinCall(
                 astgen.extra.items[extra_index] = @enumToInt(param_ref);
                 extra_index += 1;
             }
-            const result = try gz.addExtendedMultiOpPayloadIndex(.compile_log,payload_index, params.len);
+            const result = try gz.addExtendedMultiOpPayloadIndex(.compile_log, payload_index, params.len);
             return rvalue(gz, rl, result, node);
         },
         .field => {
@@ -6784,11 +6783,14 @@ fn builtinCall(
             });
             return rvalue(gz, rl, result, node);
         },
+
+        // zig fmt: off
         .as         => return as(       gz, scope, rl, node, params[0], params[1]),
         .bit_cast   => return bitCast(  gz, scope, rl, node, params[0], params[1]),
         .TypeOf     => return typeOf(   gz, scope, rl, node, params),
         .union_init => return unionInit(gz, scope, rl, node, params),
         .c_import   => return cImport(  gz, scope,     node, params[0]),
+        // zig fmt: on
 
         .@"export" => {
             const node_tags = tree.nodes.items(.tag);
@@ -6858,9 +6860,7 @@ fn builtinCall(
                     const field_ident = dot_token + 1;
                     decl_name = try astgen.identAsString(field_ident);
                 },
-                else => return astgen.failNode(
-                    params[0], "symbol to export must identify a declaration", .{},
-                ),
+                else => return astgen.failNode(params[0], "symbol to export must identify a declaration", .{}),
             }
             const options = try comptimeExpr(gz, scope, .{ .ty = .export_options_type }, params[1]);
             _ = try gz.addPlNode(.@"export", node, Zir.Inst.Export{
@@ -6888,6 +6888,7 @@ fn builtinCall(
 
         .breakpoint => return simpleNoOpVoid(gz, rl, node, .breakpoint),
 
+        // zig fmt: off
         .This               => return rvalue(gz, rl, try gz.addNodeExtended(.this,               node), node),
         .return_address     => return rvalue(gz, rl, try gz.addNodeExtended(.ret_addr,           node), node),
         .src                => return rvalue(gz, rl, try gz.addNodeExtended(.builtin_src,        node), node),
@@ -6942,6 +6943,8 @@ fn builtinCall(
         .err_set_cast => return typeCast(gz, scope, rl, node, params[0], params[1], .err_set_cast),
         .ptr_cast     => return typeCast(gz, scope, rl, node, params[0], params[1], .ptr_cast),
         .truncate     => return typeCast(gz, scope, rl, node, params[0], params[1], .truncate),
+        // zig fmt: on
+
         .align_cast => {
             const dest_align = try comptimeExpr(gz, scope, align_rl, params[0]);
             const rhs = try expr(gz, scope, .none, params[1]);
@@ -6952,6 +6955,7 @@ fn builtinCall(
             return rvalue(gz, rl, result, node);
         },
 
+        // zig fmt: off
         .has_decl  => return hasDeclOrField(gz, scope, rl, node, params[0], params[1], .has_decl),
         .has_field => return hasDeclOrField(gz, scope, rl, node, params[0], params[1], .has_field),
 
@@ -6978,6 +6982,7 @@ fn builtinCall(
 
         .cmpxchg_strong => return cmpxchg(gz, scope, rl, node, params, .cmpxchg_strong),
         .cmpxchg_weak   => return cmpxchg(gz, scope, rl, node, params, .cmpxchg_weak),
+        // zig fmt: on
 
         .wasm_memory_size => {
             const operand = try expr(gz, scope, .{ .ty = .u32_type }, params[0]);
@@ -7221,8 +7226,17 @@ fn builtinCall(
             });
             return rvalue(gz, rl, result, node);
         },
+        .prefetch => {
+            const ptr = try expr(gz, scope, .none, params[0]);
+            const options = try comptimeExpr(gz, scope, .{ .ty = .prefetch_options_type }, params[1]);
+            const result = try gz.addExtendedPayload(.prefetch, Zir.Inst.BinNode{
+                .node = gz.nodeIndexToRelative(node),
+                .lhs = ptr,
+                .rhs = options,
+            });
+            return rvalue(gz, rl, result, node);
+        },
     }
-    // zig fmt: on
 }
 
 fn simpleNoOpVoid(

src/BuiltinFn.zig

@@ -67,6 +67,7 @@ pub const Tag = enum {
     mul_with_overflow,
     panic,
     pop_count,
+    prefetch,
     ptr_cast,
     ptr_to_int,
     rem,
@@ -615,6 +616,13 @@ pub const list = list: {
                 .param_count = 2,
             },
         },
+        .{
+            "@prefetch",
+            .{
+                .tag = .prefetch,
+                .param_count = 2,
+            },
+        },
         .{
             "@ptrCast",
             .{

src/Sema.zig

@@ -1042,6 +1042,7 @@ fn zirExtended(sema: *Sema, block: *Block, inst: Zir.Inst.Index) CompileError!Ai
         .c_define           => return sema.zirCDefine(           block, extended),
         .wasm_memory_size   => return sema.zirWasmMemorySize(    block, extended),
         .wasm_memory_grow   => return sema.zirWasmMemoryGrow(    block, extended),
+        .prefetch           => return sema.zirPrefetch(          block, extended),
         // zig fmt: on
     }
 }
@@ -11104,6 +11105,16 @@ fn zirWasmMemoryGrow(
     return sema.fail(block, src, "TODO: implement Sema.zirWasmMemoryGrow", .{});
 }
 
+fn zirPrefetch(
+    sema: *Sema,
+    block: *Block,
+    extended: Zir.Inst.Extended.InstData,
+) CompileError!Air.Inst.Ref {
+    const extra = sema.code.extraData(Zir.Inst.BinNode, extended.operand).data;
+    const src: LazySrcLoc = .{ .node_offset = extra.node };
+    return sema.fail(block, src, "TODO: implement Sema.zirPrefetch", .{});
+}
+
 fn zirBuiltinExtern(
     sema: *Sema,
     block: *Block,
@@ -14231,6 +14242,7 @@ fn resolveTypeFields(sema: *Sema, block: *Block, src: LazySrcLoc, ty: Type) Comp
         .float_mode => return sema.resolveBuiltinTypeFields(block, src, "FloatMode"),
         .reduce_op => return sema.resolveBuiltinTypeFields(block, src, "ReduceOp"),
         .call_options => return sema.resolveBuiltinTypeFields(block, src, "CallOptions"),
+        .prefetch_options => return sema.resolveBuiltinTypeFields(block, src, "PrefetchOptions"),
 
         .@"union", .union_tagged => {
             const union_obj = ty.cast(Type.Payload.Union).?.data;
@@ -14819,6 +14831,7 @@ fn typeHasOnePossibleValue(
         .float_mode,
         .reduce_op,
         .call_options,
+        .prefetch_options,
         .export_options,
         .extern_options,
         .type_info,
@@ -15032,6 +15045,7 @@ pub fn addType(sema: *Sema, ty: Type) !Air.Inst.Ref {
         .float_mode => return .float_mode_type,
         .reduce_op => return .reduce_op_type,
         .call_options => return .call_options_type,
+        .prefetch_options => return .prefetch_options_type,
         .export_options => return .export_options_type,
         .extern_options => return .extern_options_type,
         .type_info => return .type_info_type,

src/Zir.zig

@@ -1572,6 +1572,9 @@ pub const Inst = struct {
         wasm_memory_size,
         /// `operand` is payload index to `BinNode`.
         wasm_memory_grow,
+        /// The `@prefetch` builtin.
+        /// `operand` is payload index to `BinNode`.
+        prefetch,
 
         pub const InstData = struct {
             opcode: Extended,
@@ -1648,6 +1651,7 @@ pub const Inst = struct {
         float_mode_type,
         reduce_op_type,
         call_options_type,
+        prefetch_options_type,
         export_options_type,
         extern_options_type,
         type_info_type,
@@ -1917,6 +1921,10 @@ pub const Inst = struct {
                 .ty = Type.initTag(.type),
                 .val = Value.initTag(.call_options_type),
             },
+            .prefetch_options_type = .{
+                .ty = Type.initTag(.type),
+                .val = Value.initTag(.prefetch_options_type),
+            },
             .export_options_type = .{
                 .ty = Type.initTag(.type),
                 .val = Value.initTag(.export_options_type),

src/print_zir.zig

@@ -477,7 +477,7 @@ const Writer = struct {
                 try self.writeSrc(stream, src);
             },
 
-            .builtin_extern, .c_define, .wasm_memory_grow => {
+            .builtin_extern, .c_define, .wasm_memory_grow, .prefetch => {
                 const inst_data = self.code.extraData(Zir.Inst.BinNode, extended.operand).data;
                 const src: LazySrcLoc = .{ .node_offset = inst_data.node };
                 try self.writeInstRef(stream, inst_data.lhs);

src/stage1/all_types.hpp

@@ -898,6 +898,18 @@ struct AstNodeFnCallExpr {
     bool seen; // used by @compileLog
 };
 
+// Must be kept in sync with std.builtin.PrefetchOptions.Rw
+enum PrefetchRw {
+    PrefetchRwRead,
+    PrefetchRwWrite,
+};
+
+// Must be kept in sync with std.builtin.PrefetchOptions.Cache
+enum PrefetchCache {
+    PrefetchCacheInstruction,
+    PrefetchCacheData,
+};
+
 struct AstNodeArrayAccessExpr {
     AstNode *array_ref_expr;
     AstNode *subscript;
@@ -1818,6 +1830,7 @@ enum BuiltinFnId {
     BuiltinFnIdReduce,
     BuiltinFnIdMaximum,
     BuiltinFnIdMinimum,
+    BuiltinFnIdPrefetch,
 };
 
 struct BuiltinFnEntry {
@@ -2021,6 +2034,7 @@ struct CodeGen {
     LLVMValueRef return_err_fn;
     LLVMValueRef wasm_memory_size;
     LLVMValueRef wasm_memory_grow;
+    LLVMValueRef prefetch;
     LLVMTypeRef anyframe_fn_type;
 
     // reminder: hash tables must be initialized before use
@@ -2647,6 +2661,7 @@ enum Stage1ZirInstId : uint8_t {
     Stage1ZirInstIdWasmMemorySize,
     Stage1ZirInstIdWasmMemoryGrow,
     Stage1ZirInstIdSrc,
+    Stage1ZirInstIdPrefetch,
 };
 
 // ir_render_* functions in codegen.cpp consume Gen instructions and produce LLVM IR.
@@ -2743,6 +2758,7 @@ enum Stage1AirInstId : uint8_t {
     Stage1AirInstIdWasmMemorySize,
     Stage1AirInstIdWasmMemoryGrow,
     Stage1AirInstIdExtern,
+    Stage1AirInstIdPrefetch,
 };
 
 struct Stage1ZirInst {
@@ -3683,6 +3699,24 @@ struct Stage1ZirInstSrc {
     Stage1ZirInst base;
 };
 
+struct Stage1ZirInstPrefetch {
+    Stage1ZirInst base;
+
+    Stage1ZirInst *ptr;
+    Stage1ZirInst *options;
+};
+
+struct Stage1AirInstPrefetch {
+    Stage1AirInst base;
+
+    Stage1AirInst *ptr;
+    PrefetchRw rw;
+    // Must be in the range 0-3 inclusive
+    uint8_t locality;
+    PrefetchCache cache;
+};
+
+
 struct Stage1ZirInstSlice {
     Stage1ZirInst base;