From 096a80ff5183955bd7754a5d80245cba0382222c Mon Sep 17 00:00:00 2001 From: Austin Clements Date: Tue, 13 Jan 2026 14:18:14 -0500 Subject: [PATCH] [dev.simd] simd/archsimd: 128- and 256-bit FMA operations do not require AVX-512 Currently, all FMA operations are marked as requiring AVX512, even on smaller vector widths. This is happening because the narrower FMA operations are marked as extension "FMA" in the XED. Since this extension doesn't start with "AVX", we filter them out very early in the XED process. However, this is just a quirk of naming: the FMA feature depends on the AVX feature, so it is part of AVX, even if it doesn't say so on the tin. Fix this by accepting the FMA extension and adding FMA to the table of CPU features. We also tweak internal/cpu slightly do it correctly enforces that the logical FMA feature depends on both the FMA and AVX CPUID flags. This actually *deletes* a lot of generated code because we no longer need the AVX-512 encoding of these 128- and 256-bit operations. Change-Id: I744a18d0be888f536ac034fe88b110347622be7e Reviewed-on: https://go-review.googlesource.com/c/go/+/736160 Auto-Submit: Austin Clements Reviewed-by: Junyang Shao LUCI-TryBot-Result: Go LUCI --- src/cmd/compile/internal/amd64/simdssa.go | 12 - .../compile/internal/ssa/_gen/simdAMD64.rules | 12 - .../compile/internal/ssa/_gen/simdAMD64ops.go | 36 +- src/cmd/compile/internal/ssa/opGen.go | 324 +++------------ src/cmd/compile/internal/ssa/rewriteAMD64.go | 372 ------------------ src/internal/cpu/cpu_x86.go | 14 +- src/simd/archsimd/_gen/simdgen/xed.go | 4 +- src/simd/archsimd/cpu.go | 10 + src/simd/archsimd/ops_amd64.go | 24 +- 9 files changed, 93 insertions(+), 715 deletions(-) diff --git a/src/cmd/compile/internal/amd64/simdssa.go b/src/cmd/compile/internal/amd64/simdssa.go index c4d0fd69c6b..d14badc3955 100644 --- a/src/cmd/compile/internal/amd64/simdssa.go +++ b/src/cmd/compile/internal/amd64/simdssa.go @@ -1959,23 +1959,11 @@ func ssaGenSIMDValue(s *ssagen.State, v *ssa.Value) bool { ssa.OpAMD64VPERMI2Q256load, ssa.OpAMD64VPERMI2PD512load, ssa.OpAMD64VPERMI2Q512load, - ssa.OpAMD64VFMADD213PS128load, - ssa.OpAMD64VFMADD213PS256load, ssa.OpAMD64VFMADD213PS512load, - ssa.OpAMD64VFMADD213PD128load, - ssa.OpAMD64VFMADD213PD256load, ssa.OpAMD64VFMADD213PD512load, - ssa.OpAMD64VFMADDSUB213PS128load, - ssa.OpAMD64VFMADDSUB213PS256load, ssa.OpAMD64VFMADDSUB213PS512load, - ssa.OpAMD64VFMADDSUB213PD128load, - ssa.OpAMD64VFMADDSUB213PD256load, ssa.OpAMD64VFMADDSUB213PD512load, - ssa.OpAMD64VFMSUBADD213PS128load, - ssa.OpAMD64VFMSUBADD213PS256load, ssa.OpAMD64VFMSUBADD213PS512load, - ssa.OpAMD64VFMSUBADD213PD128load, - ssa.OpAMD64VFMSUBADD213PD256load, ssa.OpAMD64VFMSUBADD213PD512load, ssa.OpAMD64VPSHLDVD128load, ssa.OpAMD64VPSHLDVD256load, diff --git a/src/cmd/compile/internal/ssa/_gen/simdAMD64.rules b/src/cmd/compile/internal/ssa/_gen/simdAMD64.rules index 5c83f39a1fe..dcc160bbf9d 100644 --- a/src/cmd/compile/internal/ssa/_gen/simdAMD64.rules +++ b/src/cmd/compile/internal/ssa/_gen/simdAMD64.rules @@ -2771,11 +2771,7 @@ (VPMULLQ128 x l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VPMULLQ128load {sym} [off] x ptr mem) (VPMULLQ256 x l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VPMULLQ256load {sym} [off] x ptr mem) (VPMULLQ512 x l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VPMULLQ512load {sym} [off] x ptr mem) -(VFMADD213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PS128load {sym} [off] x y ptr mem) -(VFMADD213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PS256load {sym} [off] x y ptr mem) (VFMADD213PS512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PS512load {sym} [off] x y ptr mem) -(VFMADD213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PD128load {sym} [off] x y ptr mem) -(VFMADD213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PD256load {sym} [off] x y ptr mem) (VFMADD213PD512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PD512load {sym} [off] x y ptr mem) (VFMADD213PSMasked128 x y l:(VMOVDQUload128 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PSMasked128load {sym} [off] x y ptr mask mem) (VFMADD213PSMasked256 x y l:(VMOVDQUload256 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PSMasked256load {sym} [off] x y ptr mask mem) @@ -2783,11 +2779,7 @@ (VFMADD213PDMasked128 x y l:(VMOVDQUload128 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PDMasked128load {sym} [off] x y ptr mask mem) (VFMADD213PDMasked256 x y l:(VMOVDQUload256 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PDMasked256load {sym} [off] x y ptr mask mem) (VFMADD213PDMasked512 x y l:(VMOVDQUload512 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADD213PDMasked512load {sym} [off] x y ptr mask mem) -(VFMADDSUB213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PS128load {sym} [off] x y ptr mem) -(VFMADDSUB213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PS256load {sym} [off] x y ptr mem) (VFMADDSUB213PS512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PS512load {sym} [off] x y ptr mem) -(VFMADDSUB213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PD128load {sym} [off] x y ptr mem) -(VFMADDSUB213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PD256load {sym} [off] x y ptr mem) (VFMADDSUB213PD512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PD512load {sym} [off] x y ptr mem) (VFMADDSUB213PSMasked128 x y l:(VMOVDQUload128 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PSMasked128load {sym} [off] x y ptr mask mem) (VFMADDSUB213PSMasked256 x y l:(VMOVDQUload256 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMADDSUB213PSMasked256load {sym} [off] x y ptr mask mem) @@ -2807,11 +2799,7 @@ (VPMULLQMasked128 x l:(VMOVDQUload128 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VPMULLQMasked128load {sym} [off] x ptr mask mem) (VPMULLQMasked256 x l:(VMOVDQUload256 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VPMULLQMasked256load {sym} [off] x ptr mask mem) (VPMULLQMasked512 x l:(VMOVDQUload512 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VPMULLQMasked512load {sym} [off] x ptr mask mem) -(VFMSUBADD213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PS128load {sym} [off] x y ptr mem) -(VFMSUBADD213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PS256load {sym} [off] x y ptr mem) (VFMSUBADD213PS512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PS512load {sym} [off] x y ptr mem) -(VFMSUBADD213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PD128load {sym} [off] x y ptr mem) -(VFMSUBADD213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PD256load {sym} [off] x y ptr mem) (VFMSUBADD213PD512 x y l:(VMOVDQUload512 {sym} [off] ptr mem)) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PD512load {sym} [off] x y ptr mem) (VFMSUBADD213PSMasked128 x y l:(VMOVDQUload128 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PSMasked128load {sym} [off] x y ptr mask mem) (VFMSUBADD213PSMasked256 x y l:(VMOVDQUload256 {sym} [off] ptr mem) mask) && canMergeLoad(v, l) && clobber(l) => (VFMSUBADD213PSMasked256load {sym} [off] x y ptr mask mem) diff --git a/src/cmd/compile/internal/ssa/_gen/simdAMD64ops.go b/src/cmd/compile/internal/ssa/_gen/simdAMD64ops.go index 648e372fb4a..339ec7fce39 100644 --- a/src/cmd/compile/internal/ssa/_gen/simdAMD64ops.go +++ b/src/cmd/compile/internal/ssa/_gen/simdAMD64ops.go @@ -172,38 +172,38 @@ func simdAMD64Ops(v11, v21, v2k, vkv, v2kv, v2kk, v31, v3kv, vgpv, vgp, vfpv, vf {name: "VEXPANDPSMasked128", argLength: 2, reg: wkw, asm: "VEXPANDPS", commutative: false, typ: "Vec128", resultInArg0: false}, {name: "VEXPANDPSMasked256", argLength: 2, reg: wkw, asm: "VEXPANDPS", commutative: false, typ: "Vec256", resultInArg0: false}, {name: "VEXPANDPSMasked512", argLength: 2, reg: wkw, asm: "VEXPANDPS", commutative: false, typ: "Vec512", resultInArg0: false}, - {name: "VFMADD213PD128", argLength: 3, reg: w31, asm: "VFMADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMADD213PD256", argLength: 3, reg: w31, asm: "VFMADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMADD213PD128", argLength: 3, reg: v31, asm: "VFMADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMADD213PD256", argLength: 3, reg: v31, asm: "VFMADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADD213PD512", argLength: 3, reg: w31, asm: "VFMADD213PD", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMADD213PDMasked128", argLength: 4, reg: w3kw, asm: "VFMADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMADD213PDMasked256", argLength: 4, reg: w3kw, asm: "VFMADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADD213PDMasked512", argLength: 4, reg: w3kw, asm: "VFMADD213PD", commutative: false, typ: "Vec512", resultInArg0: true}, - {name: "VFMADD213PS128", argLength: 3, reg: w31, asm: "VFMADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMADD213PS256", argLength: 3, reg: w31, asm: "VFMADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMADD213PS128", argLength: 3, reg: v31, asm: "VFMADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMADD213PS256", argLength: 3, reg: v31, asm: "VFMADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADD213PS512", argLength: 3, reg: w31, asm: "VFMADD213PS", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMADD213PSMasked128", argLength: 4, reg: w3kw, asm: "VFMADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMADD213PSMasked256", argLength: 4, reg: w3kw, asm: "VFMADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADD213PSMasked512", argLength: 4, reg: w3kw, asm: "VFMADD213PS", commutative: false, typ: "Vec512", resultInArg0: true}, - {name: "VFMADDSUB213PD128", argLength: 3, reg: w31, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMADDSUB213PD256", argLength: 3, reg: w31, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMADDSUB213PD128", argLength: 3, reg: v31, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMADDSUB213PD256", argLength: 3, reg: v31, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADDSUB213PD512", argLength: 3, reg: w31, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMADDSUB213PDMasked128", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMADDSUB213PDMasked256", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADDSUB213PDMasked512", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec512", resultInArg0: true}, - {name: "VFMADDSUB213PS128", argLength: 3, reg: w31, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMADDSUB213PS256", argLength: 3, reg: w31, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMADDSUB213PS128", argLength: 3, reg: v31, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMADDSUB213PS256", argLength: 3, reg: v31, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADDSUB213PS512", argLength: 3, reg: w31, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMADDSUB213PSMasked128", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMADDSUB213PSMasked256", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMADDSUB213PSMasked512", argLength: 4, reg: w3kw, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec512", resultInArg0: true}, - {name: "VFMSUBADD213PD128", argLength: 3, reg: w31, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMSUBADD213PD256", argLength: 3, reg: w31, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMSUBADD213PD128", argLength: 3, reg: v31, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMSUBADD213PD256", argLength: 3, reg: v31, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMSUBADD213PD512", argLength: 3, reg: w31, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMSUBADD213PDMasked128", argLength: 4, reg: w3kw, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMSUBADD213PDMasked256", argLength: 4, reg: w3kw, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMSUBADD213PDMasked512", argLength: 4, reg: w3kw, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec512", resultInArg0: true}, - {name: "VFMSUBADD213PS128", argLength: 3, reg: w31, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, - {name: "VFMSUBADD213PS256", argLength: 3, reg: w31, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, + {name: "VFMSUBADD213PS128", argLength: 3, reg: v31, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, + {name: "VFMSUBADD213PS256", argLength: 3, reg: v31, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, {name: "VFMSUBADD213PS512", argLength: 3, reg: w31, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec512", resultInArg0: true}, {name: "VFMSUBADD213PSMasked128", argLength: 4, reg: w3kw, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec128", resultInArg0: true}, {name: "VFMSUBADD213PSMasked256", argLength: 4, reg: w3kw, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec256", resultInArg0: true}, @@ -1594,38 +1594,26 @@ func simdAMD64Ops(v11, v21, v2k, vkv, v2kv, v2kk, v31, v3kv, vgpv, vgp, vfpv, vf {name: "VDIVPSMasked128load", argLength: 4, reg: w2kwload, asm: "VDIVPS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: false}, {name: "VDIVPSMasked256load", argLength: 4, reg: w2kwload, asm: "VDIVPS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: false}, {name: "VDIVPSMasked512load", argLength: 4, reg: w2kwload, asm: "VDIVPS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: false}, - {name: "VFMADD213PD128load", argLength: 4, reg: w31load, asm: "VFMADD213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADD213PD256load", argLength: 4, reg: w31load, asm: "VFMADD213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PD512load", argLength: 4, reg: w31load, asm: "VFMADD213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PDMasked128load", argLength: 5, reg: w3kwload, asm: "VFMADD213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PDMasked256load", argLength: 5, reg: w3kwload, asm: "VFMADD213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PDMasked512load", argLength: 5, reg: w3kwload, asm: "VFMADD213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADD213PS128load", argLength: 4, reg: w31load, asm: "VFMADD213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADD213PS256load", argLength: 4, reg: w31load, asm: "VFMADD213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PS512load", argLength: 4, reg: w31load, asm: "VFMADD213PS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PSMasked128load", argLength: 5, reg: w3kwload, asm: "VFMADD213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PSMasked256load", argLength: 5, reg: w3kwload, asm: "VFMADD213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADD213PSMasked512load", argLength: 5, reg: w3kwload, asm: "VFMADD213PS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADDSUB213PD128load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADDSUB213PD256load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PD512load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PDMasked128load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PDMasked256load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PDMasked512load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADDSUB213PS128load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMADDSUB213PS256load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PS512load", argLength: 4, reg: w31load, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PSMasked128load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PSMasked256load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMADDSUB213PSMasked512load", argLength: 5, reg: w3kwload, asm: "VFMADDSUB213PS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMSUBADD213PD128load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMSUBADD213PD256load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PD512load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PDMasked128load", argLength: 5, reg: w3kwload, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PDMasked256load", argLength: 5, reg: w3kwload, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PDMasked512load", argLength: 5, reg: w3kwload, asm: "VFMSUBADD213PD", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMSUBADD213PS128load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, - {name: "VFMSUBADD213PS256load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PS512load", argLength: 4, reg: w31load, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec512", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PSMasked128load", argLength: 5, reg: w3kwload, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec128", aux: "SymOff", symEffect: "Read", resultInArg0: true}, {name: "VFMSUBADD213PSMasked256load", argLength: 5, reg: w3kwload, asm: "VFMSUBADD213PS", commutative: false, typ: "Vec256", aux: "SymOff", symEffect: "Read", resultInArg0: true}, diff --git a/src/cmd/compile/internal/ssa/opGen.go b/src/cmd/compile/internal/ssa/opGen.go index abaf7911d47..e685235769f 100644 --- a/src/cmd/compile/internal/ssa/opGen.go +++ b/src/cmd/compile/internal/ssa/opGen.go @@ -2835,38 +2835,26 @@ const ( OpAMD64VDIVPSMasked128load OpAMD64VDIVPSMasked256load OpAMD64VDIVPSMasked512load - OpAMD64VFMADD213PD128load - OpAMD64VFMADD213PD256load OpAMD64VFMADD213PD512load OpAMD64VFMADD213PDMasked128load OpAMD64VFMADD213PDMasked256load OpAMD64VFMADD213PDMasked512load - OpAMD64VFMADD213PS128load - OpAMD64VFMADD213PS256load OpAMD64VFMADD213PS512load OpAMD64VFMADD213PSMasked128load OpAMD64VFMADD213PSMasked256load OpAMD64VFMADD213PSMasked512load - OpAMD64VFMADDSUB213PD128load - OpAMD64VFMADDSUB213PD256load OpAMD64VFMADDSUB213PD512load OpAMD64VFMADDSUB213PDMasked128load OpAMD64VFMADDSUB213PDMasked256load OpAMD64VFMADDSUB213PDMasked512load - OpAMD64VFMADDSUB213PS128load - OpAMD64VFMADDSUB213PS256load OpAMD64VFMADDSUB213PS512load OpAMD64VFMADDSUB213PSMasked128load OpAMD64VFMADDSUB213PSMasked256load OpAMD64VFMADDSUB213PSMasked512load - OpAMD64VFMSUBADD213PD128load - OpAMD64VFMSUBADD213PD256load OpAMD64VFMSUBADD213PD512load OpAMD64VFMSUBADD213PDMasked128load OpAMD64VFMSUBADD213PDMasked256load OpAMD64VFMSUBADD213PDMasked512load - OpAMD64VFMSUBADD213PS128load - OpAMD64VFMSUBADD213PS256load OpAMD64VFMSUBADD213PS512load OpAMD64VFMSUBADD213PSMasked128load OpAMD64VFMSUBADD213PSMasked256load @@ -23095,12 +23083,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADD213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23111,12 +23099,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADD213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23194,12 +23182,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADD213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23210,12 +23198,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADD213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23293,12 +23281,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADDSUB213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23309,12 +23297,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADDSUB213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23392,12 +23380,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADDSUB213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23408,12 +23396,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMADDSUB213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23491,12 +23479,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMSUBADD213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23507,12 +23495,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMSUBADD213PD, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23590,12 +23578,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMSUBADD213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -23606,12 +23594,12 @@ var opcodeTable = [...]opInfo{ asm: x86.AVFMSUBADD213PS, reg: regInfo{ inputs: []inputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {2, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 + {1, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 + {2, 4294901760}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 }, outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 + {0, 2147418112}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 }, }, }, @@ -44109,42 +44097,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMADD213PD128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADD213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMADD213PD256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADD213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMADD213PD512load", auxType: auxSymOff, @@ -44220,42 +44172,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMADD213PS128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADD213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMADD213PS256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADD213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMADD213PS512load", auxType: auxSymOff, @@ -44331,42 +44247,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMADDSUB213PD128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADDSUB213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMADDSUB213PD256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADDSUB213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMADDSUB213PD512load", auxType: auxSymOff, @@ -44442,42 +44322,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMADDSUB213PS128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADDSUB213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMADDSUB213PS256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMADDSUB213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMADDSUB213PS512load", auxType: auxSymOff, @@ -44553,42 +44397,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMSUBADD213PD128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMSUBADD213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMSUBADD213PD256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMSUBADD213PD, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMSUBADD213PD512load", auxType: auxSymOff, @@ -44664,42 +44472,6 @@ var opcodeTable = [...]opInfo{ }, }, }, - { - name: "VFMSUBADD213PS128load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMSUBADD213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, - { - name: "VFMSUBADD213PS256load", - auxType: auxSymOff, - argLen: 4, - resultInArg0: true, - symEffect: SymRead, - asm: x86.AVFMSUBADD213PS, - reg: regInfo{ - inputs: []inputInfo{ - {2, 72057594037977087}, // AX CX DX BX SP BP SI DI R8 R9 R10 R11 R12 R13 R15 SB - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - {1, 281474976645120}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - outputs: []outputInfo{ - {0, 281472829161472}, // X0 X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X16 X17 X18 X19 X20 X21 X22 X23 X24 X25 X26 X27 X28 X29 X30 X31 - }, - }, - }, { name: "VFMSUBADD213PS512load", auxType: auxSymOff, diff --git a/src/cmd/compile/internal/ssa/rewriteAMD64.go b/src/cmd/compile/internal/ssa/rewriteAMD64.go index 3eb2a6278b4..b33698c3e5c 100644 --- a/src/cmd/compile/internal/ssa/rewriteAMD64.go +++ b/src/cmd/compile/internal/ssa/rewriteAMD64.go @@ -782,10 +782,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VDIVPSMasked256(v) case OpAMD64VDIVPSMasked512: return rewriteValueAMD64_OpAMD64VDIVPSMasked512(v) - case OpAMD64VFMADD213PD128: - return rewriteValueAMD64_OpAMD64VFMADD213PD128(v) - case OpAMD64VFMADD213PD256: - return rewriteValueAMD64_OpAMD64VFMADD213PD256(v) case OpAMD64VFMADD213PD512: return rewriteValueAMD64_OpAMD64VFMADD213PD512(v) case OpAMD64VFMADD213PDMasked128: @@ -794,10 +790,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VFMADD213PDMasked256(v) case OpAMD64VFMADD213PDMasked512: return rewriteValueAMD64_OpAMD64VFMADD213PDMasked512(v) - case OpAMD64VFMADD213PS128: - return rewriteValueAMD64_OpAMD64VFMADD213PS128(v) - case OpAMD64VFMADD213PS256: - return rewriteValueAMD64_OpAMD64VFMADD213PS256(v) case OpAMD64VFMADD213PS512: return rewriteValueAMD64_OpAMD64VFMADD213PS512(v) case OpAMD64VFMADD213PSMasked128: @@ -806,10 +798,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VFMADD213PSMasked256(v) case OpAMD64VFMADD213PSMasked512: return rewriteValueAMD64_OpAMD64VFMADD213PSMasked512(v) - case OpAMD64VFMADDSUB213PD128: - return rewriteValueAMD64_OpAMD64VFMADDSUB213PD128(v) - case OpAMD64VFMADDSUB213PD256: - return rewriteValueAMD64_OpAMD64VFMADDSUB213PD256(v) case OpAMD64VFMADDSUB213PD512: return rewriteValueAMD64_OpAMD64VFMADDSUB213PD512(v) case OpAMD64VFMADDSUB213PDMasked128: @@ -818,10 +806,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VFMADDSUB213PDMasked256(v) case OpAMD64VFMADDSUB213PDMasked512: return rewriteValueAMD64_OpAMD64VFMADDSUB213PDMasked512(v) - case OpAMD64VFMADDSUB213PS128: - return rewriteValueAMD64_OpAMD64VFMADDSUB213PS128(v) - case OpAMD64VFMADDSUB213PS256: - return rewriteValueAMD64_OpAMD64VFMADDSUB213PS256(v) case OpAMD64VFMADDSUB213PS512: return rewriteValueAMD64_OpAMD64VFMADDSUB213PS512(v) case OpAMD64VFMADDSUB213PSMasked128: @@ -830,10 +814,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VFMADDSUB213PSMasked256(v) case OpAMD64VFMADDSUB213PSMasked512: return rewriteValueAMD64_OpAMD64VFMADDSUB213PSMasked512(v) - case OpAMD64VFMSUBADD213PD128: - return rewriteValueAMD64_OpAMD64VFMSUBADD213PD128(v) - case OpAMD64VFMSUBADD213PD256: - return rewriteValueAMD64_OpAMD64VFMSUBADD213PD256(v) case OpAMD64VFMSUBADD213PD512: return rewriteValueAMD64_OpAMD64VFMSUBADD213PD512(v) case OpAMD64VFMSUBADD213PDMasked128: @@ -842,10 +822,6 @@ func rewriteValueAMD64(v *Value) bool { return rewriteValueAMD64_OpAMD64VFMSUBADD213PDMasked256(v) case OpAMD64VFMSUBADD213PDMasked512: return rewriteValueAMD64_OpAMD64VFMSUBADD213PDMasked512(v) - case OpAMD64VFMSUBADD213PS128: - return rewriteValueAMD64_OpAMD64VFMSUBADD213PS128(v) - case OpAMD64VFMSUBADD213PS256: - return rewriteValueAMD64_OpAMD64VFMSUBADD213PS256(v) case OpAMD64VFMSUBADD213PS512: return rewriteValueAMD64_OpAMD64VFMSUBADD213PS512(v) case OpAMD64VFMSUBADD213PSMasked128: @@ -31486,64 +31462,6 @@ func rewriteValueAMD64_OpAMD64VDIVPSMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMADD213PD128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADD213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADD213PD128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADD213PD128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMADD213PD256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADD213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADD213PD256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADD213PD256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMADD213PD512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] @@ -31666,64 +31584,6 @@ func rewriteValueAMD64_OpAMD64VFMADD213PDMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMADD213PS128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADD213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADD213PS128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADD213PS128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMADD213PS256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADD213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADD213PS256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADD213PS256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMADD213PS512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] @@ -31846,64 +31706,6 @@ func rewriteValueAMD64_OpAMD64VFMADD213PSMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMADDSUB213PD128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADDSUB213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADDSUB213PD128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADDSUB213PD128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMADDSUB213PD256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADDSUB213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADDSUB213PD256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADDSUB213PD256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMADDSUB213PD512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] @@ -32026,64 +31828,6 @@ func rewriteValueAMD64_OpAMD64VFMADDSUB213PDMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMADDSUB213PS128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADDSUB213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADDSUB213PS128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADDSUB213PS128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMADDSUB213PS256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMADDSUB213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMADDSUB213PS256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMADDSUB213PS256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMADDSUB213PS512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] @@ -32206,64 +31950,6 @@ func rewriteValueAMD64_OpAMD64VFMADDSUB213PSMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMSUBADD213PD128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMSUBADD213PD128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMSUBADD213PD128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMSUBADD213PD128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMSUBADD213PD256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMSUBADD213PD256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMSUBADD213PD256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMSUBADD213PD256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMSUBADD213PD512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] @@ -32386,64 +32072,6 @@ func rewriteValueAMD64_OpAMD64VFMSUBADD213PDMasked512(v *Value) bool { } return false } -func rewriteValueAMD64_OpAMD64VFMSUBADD213PS128(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMSUBADD213PS128 x y l:(VMOVDQUload128 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMSUBADD213PS128load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload128 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMSUBADD213PS128load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} -func rewriteValueAMD64_OpAMD64VFMSUBADD213PS256(v *Value) bool { - v_2 := v.Args[2] - v_1 := v.Args[1] - v_0 := v.Args[0] - // match: (VFMSUBADD213PS256 x y l:(VMOVDQUload256 {sym} [off] ptr mem)) - // cond: canMergeLoad(v, l) && clobber(l) - // result: (VFMSUBADD213PS256load {sym} [off] x y ptr mem) - for { - x := v_0 - y := v_1 - l := v_2 - if l.Op != OpAMD64VMOVDQUload256 { - break - } - off := auxIntToInt32(l.AuxInt) - sym := auxToSym(l.Aux) - mem := l.Args[1] - ptr := l.Args[0] - if !(canMergeLoad(v, l) && clobber(l)) { - break - } - v.reset(OpAMD64VFMSUBADD213PS256load) - v.AuxInt = int32ToAuxInt(off) - v.Aux = symToAux(sym) - v.AddArg4(x, y, ptr, mem) - return true - } - return false -} func rewriteValueAMD64_OpAMD64VFMSUBADD213PS512(v *Value) bool { v_2 := v.Args[2] v_1 := v.Args[1] diff --git a/src/internal/cpu/cpu_x86.go b/src/internal/cpu/cpu_x86.go index 711fb045c33..0e97651e3d2 100644 --- a/src/internal/cpu/cpu_x86.go +++ b/src/internal/cpu/cpu_x86.go @@ -136,12 +136,6 @@ func doinit() { // e.g. setting the xsavedisable boot option on Windows 10. X86.HasOSXSAVE = isSet(ecx1, cpuid_OSXSAVE) - // The FMA instruction set extension only has VEX prefixed instructions. - // VEX prefixed instructions require OSXSAVE to be enabled. - // See Intel 64 and IA-32 Architecture Software Developer’s Manual Volume 2 - // Section 2.4 "AVX and SSE Instruction Exception Specification" - X86.HasFMA = isSet(ecx1, cpuid_FMA) && X86.HasOSXSAVE - osSupportsAVX := false osSupportsAVX512 := false // For XGETBV, OSXSAVE bit is required and sufficient. @@ -159,6 +153,14 @@ func doinit() { X86.HasAVX = isSet(ecx1, cpuid_AVX) && osSupportsAVX + // The FMA instruction set extension requires both the FMA and AVX flags. + // + // Furthermore, the FMA instructions are all VEX prefixed instructions. + // VEX prefixed instructions require OSXSAVE to be enabled. + // See Intel 64 and IA-32 Architecture Software Developer’s Manual Volume 2 + // Section 2.4 "AVX and SSE Instruction Exception Specification" + X86.HasFMA = isSet(ecx1, cpuid_FMA) && X86.HasAVX && X86.HasOSXSAVE + if maxID < 7 { osInit() return diff --git a/src/simd/archsimd/_gen/simdgen/xed.go b/src/simd/archsimd/_gen/simdgen/xed.go index 49bedc9b28b..5d6fac64d0b 100644 --- a/src/simd/archsimd/_gen/simdgen/xed.go +++ b/src/simd/archsimd/_gen/simdgen/xed.go @@ -77,7 +77,7 @@ func loadXED(xedPath string) []*unify.Value { switch { case inst.RealOpcode == "N": return // Skip unstable instructions - case !(strings.HasPrefix(inst.Extension, "AVX") || strings.HasPrefix(inst.Extension, "SHA")): + case !(strings.HasPrefix(inst.Extension, "AVX") || strings.HasPrefix(inst.Extension, "SHA") || inst.Extension == "FMA"): // We're only interested in AVX and SHA instructions. return } @@ -795,6 +795,7 @@ var cpuFeatureMap = map[string]string{ "AVX2": "AVX2", "AVXAES": "AVXAES", "SHA": "SHA", + "FMA": "FMA", // AVX-512 foundational features. We combine all of these into one "AVX512" feature. "AVX512F": "AVX512", @@ -827,6 +828,7 @@ func init() { "AVX512": {Implies: []string{"AVX2"}}, "AVXAES": {Virtual: true, Implies: []string{"AVX", "AES"}}, + "FMA": {Implies: []string{"AVX"}}, // AVX-512 subfeatures. "AVX512BITALG": {Implies: []string{"AVX512"}}, diff --git a/src/simd/archsimd/cpu.go b/src/simd/archsimd/cpu.go index 1cd765268f3..8069ee7f269 100644 --- a/src/simd/archsimd/cpu.go +++ b/src/simd/archsimd/cpu.go @@ -141,6 +141,16 @@ func (X86Features) AVXVNNI() bool { return cpu.X86.HasAVXVNNI } +// FMA returns whether the CPU supports the FMA feature. +// +// If it returns true, then the CPU also supports AVX. +// +// FMA is defined on all GOARCHes, but will only return true on +// GOARCH amd64. +func (X86Features) FMA() bool { + return cpu.X86.HasFMA +} + // SHA returns whether the CPU supports the SHA feature. // // SHA is defined on all GOARCHes, but will only return true on diff --git a/src/simd/archsimd/ops_amd64.go b/src/simd/archsimd/ops_amd64.go index 294c472fdf1..85df2a17699 100644 --- a/src/simd/archsimd/ops_amd64.go +++ b/src/simd/archsimd/ops_amd64.go @@ -4082,12 +4082,12 @@ func (x Uint64x8) Mul(y Uint64x8) Uint64x8 // MulAdd performs a fused (x * y) + z. // -// Asm: VFMADD213PS, CPU Feature: AVX512 +// Asm: VFMADD213PS, CPU Feature: FMA func (x Float32x4) MulAdd(y Float32x4, z Float32x4) Float32x4 // MulAdd performs a fused (x * y) + z. // -// Asm: VFMADD213PS, CPU Feature: AVX512 +// Asm: VFMADD213PS, CPU Feature: FMA func (x Float32x8) MulAdd(y Float32x8, z Float32x8) Float32x8 // MulAdd performs a fused (x * y) + z. @@ -4097,12 +4097,12 @@ func (x Float32x16) MulAdd(y Float32x16, z Float32x16) Float32x16 // MulAdd performs a fused (x * y) + z. // -// Asm: VFMADD213PD, CPU Feature: AVX512 +// Asm: VFMADD213PD, CPU Feature: FMA func (x Float64x2) MulAdd(y Float64x2, z Float64x2) Float64x2 // MulAdd performs a fused (x * y) + z. // -// Asm: VFMADD213PD, CPU Feature: AVX512 +// Asm: VFMADD213PD, CPU Feature: FMA func (x Float64x4) MulAdd(y Float64x4, z Float64x4) Float64x4 // MulAdd performs a fused (x * y) + z. @@ -4114,12 +4114,12 @@ func (x Float64x8) MulAdd(y Float64x8, z Float64x8) Float64x8 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. // -// Asm: VFMADDSUB213PS, CPU Feature: AVX512 +// Asm: VFMADDSUB213PS, CPU Feature: FMA func (x Float32x4) MulAddSub(y Float32x4, z Float32x4) Float32x4 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. // -// Asm: VFMADDSUB213PS, CPU Feature: AVX512 +// Asm: VFMADDSUB213PS, CPU Feature: FMA func (x Float32x8) MulAddSub(y Float32x8, z Float32x8) Float32x8 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. @@ -4129,12 +4129,12 @@ func (x Float32x16) MulAddSub(y Float32x16, z Float32x16) Float32x16 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. // -// Asm: VFMADDSUB213PD, CPU Feature: AVX512 +// Asm: VFMADDSUB213PD, CPU Feature: FMA func (x Float64x2) MulAddSub(y Float64x2, z Float64x2) Float64x2 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. // -// Asm: VFMADDSUB213PD, CPU Feature: AVX512 +// Asm: VFMADDSUB213PD, CPU Feature: FMA func (x Float64x4) MulAddSub(y Float64x4, z Float64x4) Float64x4 // MulAddSub performs a fused (x * y) - z for odd-indexed elements, and (x * y) + z for even-indexed elements. @@ -4204,12 +4204,12 @@ func (x Uint16x32) MulHigh(y Uint16x32) Uint16x32 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. // -// Asm: VFMSUBADD213PS, CPU Feature: AVX512 +// Asm: VFMSUBADD213PS, CPU Feature: FMA func (x Float32x4) MulSubAdd(y Float32x4, z Float32x4) Float32x4 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. // -// Asm: VFMSUBADD213PS, CPU Feature: AVX512 +// Asm: VFMSUBADD213PS, CPU Feature: FMA func (x Float32x8) MulSubAdd(y Float32x8, z Float32x8) Float32x8 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. @@ -4219,12 +4219,12 @@ func (x Float32x16) MulSubAdd(y Float32x16, z Float32x16) Float32x16 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. // -// Asm: VFMSUBADD213PD, CPU Feature: AVX512 +// Asm: VFMSUBADD213PD, CPU Feature: FMA func (x Float64x2) MulSubAdd(y Float64x2, z Float64x2) Float64x2 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements. // -// Asm: VFMSUBADD213PD, CPU Feature: AVX512 +// Asm: VFMSUBADD213PD, CPU Feature: FMA func (x Float64x4) MulSubAdd(y Float64x4, z Float64x4) Float64x4 // MulSubAdd performs a fused (x * y) + z for odd-indexed elements, and (x * y) - z for even-indexed elements.