authorgravatar for jacobly@ziglang.orgJacob Young <jacobly@ziglang.org> 2023-05-03 02:37:48-04:00
committergravatar for jacobly@ziglang.orgJacob Young <jacobly@ziglang.org> 2023-05-03 04:25:14-04:00
log6893f90887836584f9377793cca7235d8947a326
treeff0680c19ee97c6521b92fccf324d6f21df65ad7
parenta5e50891cbd66c740afba722e94f841a2b43c1fc

x86_64: implement sqrt

5 files changed, 49 insertions(+), 1 deletions(-)

src/arch/x86_64/CodeGen.zig+27-1
...@@ -229,6 +229,7 @@ pub const MCValue = union(enum) {...@@ -229,6 +229,7 @@ pub const MCValue = union(enum) {
229 fn isRegister(mcv: MCValue) bool {229 fn isRegister(mcv: MCValue) bool {
230 return switch (mcv) {230 return switch (mcv) {
231 .register => true,231 .register => true,
232 .register_offset => |reg_off| return reg_off.off == 0,
232 else => false,233 else => false,
233 };234 };
234 }235 }
...@@ -1449,7 +1450,6 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {...@@ -1449,7 +1450,6 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
1449 .shl_sat => try self.airShlSat(inst),1450 .shl_sat => try self.airShlSat(inst),
1450 .slice => try self.airSlice(inst),1451 .slice => try self.airSlice(inst),
14511452
1452 .sqrt,
1453 .sin,1453 .sin,
1454 .cos,1454 .cos,
1455 .tan,1455 .tan,
...@@ -1464,6 +1464,7 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {...@@ -1464,6 +1464,7 @@ fn genBody(self: *Self, body: []const Air.Inst.Index) InnerError!void {
1464 .trunc_float,1464 .trunc_float,
1465 => try self.airUnaryMath(inst),1465 => try self.airUnaryMath(inst),
14661466
1467 .sqrt => try self.airSqrt(inst),
1467 .neg, .fabs => try self.airFloatSign(inst),1468 .neg, .fabs => try self.airFloatSign(inst),
14681469
1469 .add_with_overflow => try self.airAddSubWithOverflow(inst),1470 .add_with_overflow => try self.airAddSubWithOverflow(inst),
...@@ -4242,6 +4243,31 @@ fn airFloatSign(self: *Self, inst: Air.Inst.Index) !void {...@@ -4242,6 +4243,31 @@ fn airFloatSign(self: *Self, inst: Air.Inst.Index) !void {
4242 return self.finishAir(inst, dst_mcv, .{ un_op, .none, .none });4243 return self.finishAir(inst, dst_mcv, .{ un_op, .none, .none });
4243}4244}
42444245
4246fn airSqrt(self: *Self, inst: Air.Inst.Index) !void {
4247 const un_op = self.air.instructions.items(.data)[inst].un_op;
4248 const ty = self.air.typeOf(un_op);
4249
4250 const src_mcv = try self.resolveInst(un_op);
4251 const dst_mcv = if (src_mcv.isRegister() and self.reuseOperand(inst, un_op, 0, src_mcv))
4252 src_mcv
4253 else
4254 try self.copyToRegisterWithInstTracking(inst, ty, src_mcv);
4255
4256 try self.genBinOpMir(switch (ty.zigTypeTag()) {
4257 .Float => switch (ty.floatBits(self.target.*)) {
4258 32 => .sqrtss,
4259 64 => .sqrtsd,
4260 else => return self.fail("TODO implement airSqrt for {}", .{
4261 ty.fmt(self.bin_file.options.module.?),
4262 }),
4263 },
4264 else => return self.fail("TODO implement airSqrt for {}", .{
4265 ty.fmt(self.bin_file.options.module.?),
4266 }),
4267 }, ty, dst_mcv, src_mcv);
4268 return self.finishAir(inst, dst_mcv, .{ un_op, .none, .none });
4269}
4270
4245fn airUnaryMath(self: *Self, inst: Air.Inst.Index) !void {4271fn airUnaryMath(self: *Self, inst: Air.Inst.Index) !void {
4246 const un_op = self.air.instructions.items(.data)[inst].un_op;4272 const un_op = self.air.instructions.items(.data)[inst].un_op;
4247 _ = un_op;4273 _ = un_op;
src/arch/x86_64/Encoding.zig+4
...@@ -277,6 +277,8 @@ pub const Mnemonic = enum {...@@ -277,6 +277,8 @@ pub const Mnemonic = enum {
277 movss,277 movss,
278 mulss,278 mulss,
279 orps,279 orps,
280 sqrtps,
281 sqrtss,
280 subss,282 subss,
281 ucomiss,283 ucomiss,
282 xorps,284 xorps,
...@@ -291,6 +293,8 @@ pub const Mnemonic = enum {...@@ -291,6 +293,8 @@ pub const Mnemonic = enum {
291 movq, //movd, movsd,293 movq, //movd, movsd,
292 mulsd,294 mulsd,
293 orpd,295 orpd,
296 sqrtpd,
297 sqrtsd,
294 subsd,298 subsd,
295 ucomisd,299 ucomisd,
296 xorpd,300 xorpd,
src/arch/x86_64/Lower.zig+4
...@@ -105,6 +105,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {...@@ -105,6 +105,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {
105 .mulss,105 .mulss,
106 .orps,106 .orps,
107 .roundss,107 .roundss,
108 .sqrtps,
109 .sqrtss,
108 .subss,110 .subss,
109 .ucomiss,111 .ucomiss,
110 .xorps,112 .xorps,
...@@ -122,6 +124,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {...@@ -122,6 +124,8 @@ pub fn lowerMir(lower: *Lower, inst: Mir.Inst) Error![]const Instruction {
122 .mulsd,124 .mulsd,
123 .orpd,125 .orpd,
124 .roundsd,126 .roundsd,
127 .sqrtpd,
128 .sqrtsd,
125 .subsd,129 .subsd,
126 .ucomisd,130 .ucomisd,
127 .xorpd,131 .xorpd,
src/arch/x86_64/Mir.zig+8
...@@ -190,7 +190,11 @@ pub const Inst = struct {...@@ -190,7 +190,11 @@ pub const Inst = struct {
190 orps,190 orps,
191 /// Round scalar single-precision floating-point values191 /// Round scalar single-precision floating-point values
192 roundss,192 roundss,
193 /// Square root of scalar single precision floating-point value
194 sqrtps,
193 /// Subtract scalar single-precision floating-point values195 /// Subtract scalar single-precision floating-point values
196 sqrtss,
197 /// Square root of single precision floating-point values
194 subss,198 subss,
195 /// Unordered compare scalar single-precision floating-point values199 /// Unordered compare scalar single-precision floating-point values
196 ucomiss,200 ucomiss,
...@@ -224,6 +228,10 @@ pub const Inst = struct {...@@ -224,6 +228,10 @@ pub const Inst = struct {
224 orpd,228 orpd,
225 /// Round scalar double-precision floating-point values229 /// Round scalar double-precision floating-point values
226 roundsd,230 roundsd,
231 /// Square root of double precision floating-point values
232 sqrtpd,
233 /// Square root of scalar double precision floating-point value
234 sqrtsd,
227 /// Subtract scalar double-precision floating-point values235 /// Subtract scalar double-precision floating-point values
228 subsd,236 subsd,
229 /// Unordered compare scalar double-precision floating-point values237 /// Unordered compare scalar double-precision floating-point values
src/arch/x86_64/encodings.zig+6
...@@ -856,6 +856,9 @@ pub const table = [_]Entry{...@@ -856,6 +856,9 @@ pub const table = [_]Entry{
856856
857 .{ .subss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x5c }, 0, .sse },857 .{ .subss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x5c }, 0, .sse },
858858
859 .{ .sqrtps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x51 }, 0, .sse },
860 .{ .sqrtss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0xf3, 0x0f, 0x51 }, 0, .sse },
861
859 .{ .ucomiss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0x0f, 0x2e }, 0, .sse },862 .{ .ucomiss, .rm, &.{ .xmm, .xmm_m32 }, &.{ 0x0f, 0x2e }, 0, .sse },
860863
861 .{ .xorps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x57 }, 0, .sse },864 .{ .xorps, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x0f, 0x57 }, 0, .sse },
...@@ -895,6 +898,9 @@ pub const table = [_]Entry{...@@ -895,6 +898,9 @@ pub const table = [_]Entry{
895898
896 .{ .orpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x56 }, 0, .sse2 },899 .{ .orpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x56 }, 0, .sse2 },
897900
901 .{ .sqrtpd, .rm, &.{ .xmm, .xmm_m128 }, &.{ 0x66, 0x0f, 0x51 }, 0, .sse2 },
902 .{ .sqrtsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x51 }, 0, .sse2 },
903
898 .{ .subsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5c }, 0, .sse2 },904 .{ .subsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x5c }, 0, .sse2 },
899905
900 .{ .movsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x10 }, 0, .sse2 },906 .{ .movsd, .rm, &.{ .xmm, .xmm_m64 }, &.{ 0xf2, 0x0f, 0x10 }, 0, .sse2 },