Rollup merge of rust-lang#97097 - chorman0773:add_tmm_clobers, r=joshtriplett

JohnTitor · web-flow · commit 7361b22b4c57 · 2022-05-18T07:40:59.000+09:00
Add tmm_reg clobbers

This adds support for naming the 8 tile registers from intel AMX as clobbers from `asm!` invocations on x86_64 (only). It does not add the registers as input or output operands.
diff --git a/compiler/rustc_codegen_gcc/src/asm.rs b/compiler/rustc_codegen_gcc/src/asm.rs
@@ -592,7 +592,7 @@ fn reg_to_gcc(reg: InlineAsmRegOrRegClass) -> ConstraintOrRegister {
             InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => unimplemented!(),
             InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => unimplemented!(),
             InlineAsmRegClass::X86(
-                X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg,
+                X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg | X86InlineAsmRegClass::tmm_reg,
             ) => unreachable!("clobber-only"),
             InlineAsmRegClass::SpirV(SpirVInlineAsmRegClass::reg) => {
                 bug!("GCC backend does not support SPIR-V")
@@ -656,6 +656,7 @@ fn dummy_output_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, reg: InlineAsmRegCl
         InlineAsmRegClass::X86(X86InlineAsmRegClass::x87_reg) => unimplemented!(),
         InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => cx.type_i16(),
         InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => cx.type_i16(),
+        InlineAsmRegClass::X86(X86InlineAsmRegClass::tmm_reg) => unimplemented!(),
         InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => cx.type_i32(),
         InlineAsmRegClass::SpirV(SpirVInlineAsmRegClass::reg) => {
             bug!("LLVM backend does not support SPIR-V")
@@ -787,7 +788,7 @@ fn modifier_to_gcc(arch: InlineAsmArch, reg: InlineAsmRegClass, modifier: Option
         },
         InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => None,
         InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => None,
-        InlineAsmRegClass::X86(X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg) => {
+        InlineAsmRegClass::X86(X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg | X86InlineAsmRegClass::tmm_reg) => {
             unreachable!("clobber-only")
         }
         InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => unimplemented!(),
diff --git a/compiler/rustc_codegen_llvm/src/asm.rs b/compiler/rustc_codegen_llvm/src/asm.rs
@@ -604,7 +604,8 @@ fn reg_to_llvm(reg: InlineAsmRegOrRegClass, layout: Option<&TyAndLayout<'_>>) ->
             InlineAsmRegClass::X86(
                 X86InlineAsmRegClass::x87_reg
                 | X86InlineAsmRegClass::mmx_reg
-                | X86InlineAsmRegClass::kreg0,
+                | X86InlineAsmRegClass::kreg0
+                | X86InlineAsmRegClass::tmm_reg,
             ) => unreachable!("clobber-only"),
             InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => "r",
             InlineAsmRegClass::Bpf(BpfInlineAsmRegClass::reg) => "r",
@@ -692,7 +693,8 @@ fn modifier_to_llvm(
         InlineAsmRegClass::X86(
             X86InlineAsmRegClass::x87_reg
             | X86InlineAsmRegClass::mmx_reg
-            | X86InlineAsmRegClass::kreg0,
+            | X86InlineAsmRegClass::kreg0
+            | X86InlineAsmRegClass::tmm_reg,
         ) => {
             unreachable!("clobber-only")
         }
@@ -766,7 +768,8 @@ fn dummy_output_type<'ll>(cx: &CodegenCx<'ll, '_>, reg: InlineAsmRegClass) -> &'
         InlineAsmRegClass::X86(
             X86InlineAsmRegClass::x87_reg
             | X86InlineAsmRegClass::mmx_reg
-            | X86InlineAsmRegClass::kreg0,
+            | X86InlineAsmRegClass::kreg0
+            | X86InlineAsmRegClass::tmm_reg,
         ) => {
             unreachable!("clobber-only")
         }
diff --git a/compiler/rustc_span/src/symbol.rs b/compiler/rustc_span/src/symbol.rs
@@ -1408,6 +1408,7 @@ symbols! {
         thread_local_macro,
         thumb2,
         thumb_mode: "thumb-mode",
+        tmm_reg,
         todo_macro,
         tool_attributes,
         tool_lints,
diff --git a/compiler/rustc_target/src/asm/mod.rs b/compiler/rustc_target/src/asm/mod.rs
@@ -912,6 +912,7 @@ impl InlineAsmClobberAbi {
 
                     mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7,
                     st0, st1, st2, st3, st4, st5, st6, st7,
+                    tmm0, tmm1, tmm2, tmm3, tmm4, tmm5, tmm6, tmm7,
                 }
             },
             InlineAsmClobberAbi::X86_64Win => clobbered_regs! {
@@ -931,6 +932,7 @@ impl InlineAsmClobberAbi {
 
                     mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7,
                     st0, st1, st2, st3, st4, st5, st6, st7,
+                    tmm0, tmm1, tmm2, tmm3, tmm4, tmm5, tmm6, tmm7,
                 }
             },
             InlineAsmClobberAbi::AArch64 => clobbered_regs! {
diff --git a/compiler/rustc_target/src/asm/x86.rs b/compiler/rustc_target/src/asm/x86.rs
@@ -17,6 +17,7 @@ def_reg_class! {
         kreg0,
         mmx_reg,
         x87_reg,
+        tmm_reg,
     }
 }
 
@@ -41,6 +42,7 @@ impl X86InlineAsmRegClass {
             Self::xmm_reg | Self::ymm_reg | Self::zmm_reg => &['x', 'y', 'z'],
             Self::kreg | Self::kreg0 => &[],
             Self::mmx_reg | Self::x87_reg => &[],
+            Self::tmm_reg => &[],
         }
     }
 
@@ -80,6 +82,7 @@ impl X86InlineAsmRegClass {
             },
             Self::kreg | Self::kreg0 => None,
             Self::mmx_reg | Self::x87_reg => None,
+            Self::tmm_reg => None,
         }
     }
 
@@ -98,6 +101,7 @@ impl X86InlineAsmRegClass {
             Self::zmm_reg => Some(('z', "zmm0")),
             Self::kreg | Self::kreg0 => None,
             Self::mmx_reg | Self::x87_reg => None,
+            Self::tmm_reg => None,
         }
     }
 
@@ -135,6 +139,7 @@ impl X86InlineAsmRegClass {
             },
             Self::kreg0 => &[],
             Self::mmx_reg | Self::x87_reg => &[],
+            Self::tmm_reg => &[],
         }
     }
 }
@@ -320,6 +325,14 @@ def_regs! {
         st5: x87_reg = ["st(5)"],
         st6: x87_reg = ["st(6)"],
         st7: x87_reg = ["st(7)"],
+        tmm0: tmm_reg = ["tmm0"] % x86_64_only,
+        tmm1: tmm_reg = ["tmm1"] % x86_64_only,
+        tmm2: tmm_reg = ["tmm2"] % x86_64_only,
+        tmm3: tmm_reg = ["tmm3"] % x86_64_only,
+        tmm4: tmm_reg = ["tmm4"] % x86_64_only,
+        tmm5: tmm_reg = ["tmm5"] % x86_64_only,
+        tmm6: tmm_reg = ["tmm6"] % x86_64_only,
+        tmm7: tmm_reg = ["tmm7"] % x86_64_only,
         #error = ["bp", "bpl", "ebp", "rbp"] =>
             "the frame pointer cannot be used as an operand for inline asm",
         #error = ["sp", "spl", "esp", "rsp"] =>
diff --git a/src/test/codegen/asm-clobber_abi.rs b/src/test/codegen/asm-clobber_abi.rs
@@ -6,29 +6,29 @@
 use std::arch::asm;
 
 // CHECK-LABEL: @clobber_sysv64
-// CHECK: ={ax},={cx},={dx},={si},={di},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{dirflag},~{fpsr},~{flags},~{memory}
+// CHECK: ={ax},={cx},={dx},={si},={di},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{tmm0},~{tmm1},~{tmm2},~{tmm3},~{tmm4},~{tmm5},~{tmm6},~{tmm7},~{dirflag},~{fpsr},~{flags},~{memory}
 #[no_mangle]
 pub unsafe fn clobber_sysv64() {
     asm!("", clobber_abi("sysv64"));
 }
 
 // CHECK-LABEL: @clobber_win64
-// CHECK: ={ax},={cx},={dx},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{dirflag},~{fpsr},~{flags},~{memory}
+// CHECK: ={ax},={cx},={dx},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{tmm0},~{tmm1},~{tmm2},~{tmm3},~{tmm4},~{tmm5},~{tmm6},~{tmm7},~{dirflag},~{fpsr},~{flags},~{memory}
 #[no_mangle]
 pub unsafe fn clobber_win64() {
     asm!("", clobber_abi("win64"));
 }
 
 // CHECK-LABEL: @clobber_sysv64
-// CHECK: =&{dx},={ax},={cx},={si},={di},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{dirflag},~{fpsr},~{flags},~{memory}
+// CHECK: =&{dx},={ax},={cx},={si},={di},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{tmm0},~{tmm1},~{tmm2},~{tmm3},~{tmm4},~{tmm5},~{tmm6},~{tmm7},~{dirflag},~{fpsr},~{flags},~{memory}
 #[no_mangle]
 pub unsafe fn clobber_sysv64_edx() {
     let foo: i32;
     asm!("", out("edx") foo, clobber_abi("sysv64"));
 }
 
 // CHECK-LABEL: @clobber_win64
-// CHECK: =&{dx},={ax},={cx},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{dirflag},~{fpsr},~{flags},~{memory}
+// CHECK: =&{dx},={ax},={cx},={r8},={r9},={r10},={r11},={xmm0},={xmm1},={xmm2},={xmm3},={xmm4},={xmm5},={xmm6},={xmm7},={xmm8},={xmm9},={xmm10},={xmm11},={xmm12},={xmm13},={xmm14},={xmm15},~{xmm16},~{xmm17},~{xmm18},~{xmm19},~{xmm20},~{xmm21},~{xmm22},~{xmm23},~{xmm24},~{xmm25},~{xmm26},~{xmm27},~{xmm28},~{xmm29},~{xmm30},~{xmm31},~{k0},~{k1},~{k2},~{k3},~{k4},~{k5},~{k6},~{k7},~{st},~{st(1)},~{st(2)},~{st(3)},~{st(4)},~{st(5)},~{st(6)},~{st(7)},~{tmm0},~{tmm1},~{tmm2},~{tmm3},~{tmm4},~{tmm5},~{tmm6},~{tmm7},~{dirflag},~{fpsr},~{flags},~{memory}
 #[no_mangle]
 pub unsafe fn clobber_win64_edx() {
     let foo: i32;
diff --git a/src/test/codegen/asm-target-clobbers.rs b/src/test/codegen/asm-target-clobbers.rs
@@ -6,6 +6,13 @@
 
 use std::arch::asm;
 
+// CHECK-LABEL: @amx_clobber
+// base: call void asm sideeffect inteldialect "", "~{tmm0}"()
+#[no_mangle]
+pub unsafe fn amx_clobber() {
+    asm!("", out("tmm0") _, options(nostack, nomem, preserves_flags));
+}
+
 // CHECK-LABEL: @avx512_clobber
 // base: call void asm sideeffect inteldialect "", "~{xmm31}"()
 // avx512: call float asm sideeffect inteldialect "", "=&{xmm31}"()

Original file line number	Diff line number	Diff line change
`@@ -912,6 +912,7 @@ impl InlineAsmClobberAbi {`
`912`	`912`
`913`	`913`	`mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7,`
`914`	`914`	`st0, st1, st2, st3, st4, st5, st6, st7,`
	`915`	`+ tmm0, tmm1, tmm2, tmm3, tmm4, tmm5, tmm6, tmm7,`
`915`	`916`	`}`
`916`	`917`	`},`
`917`	`918`	`InlineAsmClobberAbi::X86_64Win => clobbered_regs! {`
`@@ -931,6 +932,7 @@ impl InlineAsmClobberAbi {`
`931`	`932`
`932`	`933`	`mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7,`
`933`	`934`	`st0, st1, st2, st3, st4, st5, st6, st7,`
	`935`	`+ tmm0, tmm1, tmm2, tmm3, tmm4, tmm5, tmm6, tmm7,`
`934`	`936`	`}`
`935`	`937`	`},`
`936`	`938`	`InlineAsmClobberAbi::AArch64 => clobbered_regs! {`
Original file line number	Diff line number	Diff line change
`@@ -17,6 +17,7 @@ def_reg_class! {`
`17`	`17`	`kreg0,`
`18`	`18`	`mmx_reg,`
`19`	`19`	`x87_reg,`
	`20`	`+ tmm_reg,`
`20`	`21`	`}`
`21`	`22`	`}`
`22`	`23`
`@@ -41,6 +42,7 @@ impl X86InlineAsmRegClass {`
`41`	`42`	`Self::xmm_reg \| Self::ymm_reg \| Self::zmm_reg => &['x', 'y', 'z'],`
`42`	`43`	`Self::kreg \| Self::kreg0 => &[],`
`43`	`44`	`Self::mmx_reg \| Self::x87_reg => &[],`
	`45`	`+ Self::tmm_reg => &[],`
`44`	`46`	`}`
`45`	`47`	`}`
`46`	`48`
`@@ -80,6 +82,7 @@ impl X86InlineAsmRegClass {`
`80`	`82`	`},`
`81`	`83`	`Self::kreg \| Self::kreg0 => None,`
`82`	`84`	`Self::mmx_reg \| Self::x87_reg => None,`
	`85`	`+ Self::tmm_reg => None,`
`83`	`86`	`}`
`84`	`87`	`}`
`85`	`88`
`@@ -98,6 +101,7 @@ impl X86InlineAsmRegClass {`
`98`	`101`	`Self::zmm_reg => Some(('z', "zmm0")),`
`99`	`102`	`Self::kreg \| Self::kreg0 => None,`
`100`	`103`	`Self::mmx_reg \| Self::x87_reg => None,`
	`104`	`+ Self::tmm_reg => None,`
`101`	`105`	`}`
`102`	`106`	`}`
`103`	`107`
`@@ -135,6 +139,7 @@ impl X86InlineAsmRegClass {`
`135`	`139`	`},`
`136`	`140`	`Self::kreg0 => &[],`
`137`	`141`	`Self::mmx_reg \| Self::x87_reg => &[],`
	`142`	`+ Self::tmm_reg => &[],`
`138`	`143`	`}`
`139`	`144`	`}`
`140`	`145`	`}`
`@@ -320,6 +325,14 @@ def_regs! {`
`320`	`325`	`st5: x87_reg = ["st(5)"],`
`321`	`326`	`st6: x87_reg = ["st(6)"],`
`322`	`327`	`st7: x87_reg = ["st(7)"],`
	`328`	`+ tmm0: tmm_reg = ["tmm0"] % x86_64_only,`
	`329`	`+ tmm1: tmm_reg = ["tmm1"] % x86_64_only,`
	`330`	`+ tmm2: tmm_reg = ["tmm2"] % x86_64_only,`
	`331`	`+ tmm3: tmm_reg = ["tmm3"] % x86_64_only,`
	`332`	`+ tmm4: tmm_reg = ["tmm4"] % x86_64_only,`
	`333`	`+ tmm5: tmm_reg = ["tmm5"] % x86_64_only,`
	`334`	`+ tmm6: tmm_reg = ["tmm6"] % x86_64_only,`
	`335`	`+ tmm7: tmm_reg = ["tmm7"] % x86_64_only,`
`323`	`336`	`#error = ["bp", "bpl", "ebp", "rbp"] =>`
`324`	`337`	`"the frame pointer cannot be used as an operand for inline asm",`
`325`	`338`	`#error = ["sp", "spl", "esp", "rsp"] =>`