stage2: lower float negation explicitly

Rather than lowering float negation as `0.0 - x`.

 * Add AIR instruction for float negation.
 * Add compiler-rt functions for f128, f80 negation

closes #11853

src/codegen/c.zig

@@ -1755,6 +1755,8 @@ fn genBody(f: *Function, body: []const Air.Inst.Index) error{ AnalysisFail, OutO
             .mul_sat => try airSatOp(f, inst, "muls_"),
             .shl_sat => try airSatOp(f, inst, "shls_"),
 
+            .neg => try airNeg(f, inst),
+
             .sqrt,
             .sin,
             .cos,
@@ -4098,6 +4100,20 @@ fn airWasmMemoryGrow(f: *Function, inst: Air.Inst.Index) !CValue {
     return local;
 }
 
+fn airNeg(f: *Function, inst: Air.Inst.Index) !CValue {
+    if (f.liveness.isUnused(inst)) return CValue.none;
+
+    const un_op = f.air.instructions.items(.data)[inst].un_op;
+    const writer = f.object.writer();
+    const inst_ty = f.air.typeOfIndex(inst);
+    const operand = try f.resolveInst(un_op);
+    const local = try f.allocLocal(inst_ty, .Const);
+    try writer.writeAll("-");
+    try f.writeCValue(writer, operand);
+    try writer.writeAll(";\n");
+    return local;
+}
+
 fn airMulAdd(f: *Function, inst: Air.Inst.Index) !CValue {
     if (f.liveness.isUnused(inst)) return CValue.none;
     const pl_op = f.air.instructions.items(.data)[inst].pl_op;

src/codegen/llvm.zig

@@ -4019,6 +4019,7 @@ pub const FuncGen = struct {
                 .ceil         => try self.airUnaryOp(inst, .ceil),
                 .round        => try self.airUnaryOp(inst, .round),
                 .trunc_float  => try self.airUnaryOp(inst, .trunc),
+                .neg          => try self.airUnaryOp(inst, .neg),
 
                 .cmp_eq  => try self.airCmp(inst, .eq),
                 .cmp_gt  => try self.airCmp(inst, .gt),
@@ -6545,13 +6546,14 @@ pub const FuncGen = struct {
         fabs,
         floor,
         fma,
+        fmax,
+        fmin,
+        fmod,
         log,
         log10,
         log2,
-        fmax,
-        fmin,
         mul,
-        fmod,
+        neg,
         round,
         sin,
         sqrt,
@@ -6584,6 +6586,7 @@ pub const FuncGen = struct {
         var fn_name_buf: [64]u8 = undefined;
         const strat: FloatOpStrat = if (intrinsics_allowed) switch (op) {
             // Some operations are dedicated LLVM instructions, not available as intrinsics
+            .neg => return self.builder.buildFNeg(params[0], ""),
             .add => return self.builder.buildFAdd(params[0], params[1], ""),
             .sub => return self.builder.buildFSub(params[0], params[1], ""),
             .mul => return self.builder.buildFMul(params[0], params[1], ""),
@@ -6595,6 +6598,11 @@ pub const FuncGen = struct {
         } else b: {
             const float_bits = scalar_ty.floatBits(target);
             break :b switch (op) {
+                .neg => FloatOpStrat{
+                    .libc = std.fmt.bufPrintZ(&fn_name_buf, "__neg{s}f2", .{
+                        compilerRtFloatAbbrev(float_bits),
+                    }) catch unreachable,
+                },
                 .add, .sub, .div, .mul => FloatOpStrat{
                     .libc = std.fmt.bufPrintZ(&fn_name_buf, "__{s}{s}f3", .{
                         @tagName(op), compilerRtFloatAbbrev(float_bits),

src/codegen/llvm/bindings.zig

@@ -549,6 +549,9 @@ pub const Builder = opaque {
     pub const buildFSub = LLVMBuildFSub;
     extern fn LLVMBuildFSub(*const Builder, LHS: *const Value, RHS: *const Value, Name: [*:0]const u8) *const Value;
 
+    pub const buildFNeg = LLVMBuildFNeg;
+    extern fn LLVMBuildFNeg(*const Builder, V: *const Value, Name: [*:0]const u8) *const Value;
+
     pub const buildSub = LLVMBuildSub;
     extern fn LLVMBuildSub(*const Builder, LHS: *const Value, RHS: *const Value, Name: [*:0]const u8) *const Value;