diff options
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrAVX512.td | 53 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrSSE.td | 35 |
2 files changed, 35 insertions, 53 deletions
diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index 9f4a75c6689..8315b867316 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -4286,15 +4286,6 @@ let Predicates = [HasAVX512, OptForSize] in { (v4i32 (VMOVSSZrr (v4i32 (AVX512_128_SET0)), (v4i32 (EXTRACT_SUBREG (v8i32 VR256X:$src), sub_xmm)))), sub_xmm)>; - def : Pat<(v4f64 (X86vzmovl (v4f64 VR256X:$src))), - (SUBREG_TO_REG (i32 0), - (v2f64 (VMOVSDZrr (v2f64 (AVX512_128_SET0)), - (v2f64 (EXTRACT_SUBREG (v4f64 VR256X:$src), sub_xmm)))), sub_xmm)>; - def : Pat<(v4i64 (X86vzmovl (v4i64 VR256X:$src))), - (SUBREG_TO_REG (i32 0), - (v2i64 (VMOVSDZrr (v2i64 (AVX512_128_SET0)), - (v2i64 (EXTRACT_SUBREG (v4i64 VR256X:$src), sub_xmm)))), sub_xmm)>; - def : Pat<(v16f32 (X86vzmovl (v16f32 VR512:$src))), (SUBREG_TO_REG (i32 0), (v4f32 (VMOVSSZrr (v4f32 (AVX512_128_SET0)), @@ -4303,17 +4294,6 @@ let Predicates = [HasAVX512, OptForSize] in { (SUBREG_TO_REG (i32 0), (v4i32 (VMOVSSZrr (v4i32 (AVX512_128_SET0)), (v4i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_xmm)))), sub_xmm)>; - - def : Pat<(v8f64 (X86vzmovl (v8f64 VR512:$src))), - (SUBREG_TO_REG (i32 0), - (v2f64 (VMOVSDZrr (v2f64 (AVX512_128_SET0)), - (v2f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm)))), sub_xmm)>; - - def : Pat<(v8i64 (X86vzmovl (v8i64 VR512:$src))), - (SUBREG_TO_REG (i32 0), - (v2i64 (VMOVSDZrr (v2i64 (AVX512_128_SET0)), - (v2i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm)))), sub_xmm)>; - } // Use 128-bit blends for OptForSpeed since BLENDs have better throughput than @@ -4329,17 +4309,6 @@ let Predicates = [HasAVX512, OptForSpeed] in { (v4i32 (VPBLENDWrri (v4i32 (V_SET0)), (v4i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_xmm)), (i8 3))), sub_xmm)>; - - def : Pat<(v8f64 (X86vzmovl (v8f64 VR512:$src))), - (SUBREG_TO_REG (i32 0), - (v2f64 (VBLENDPDrri (v2f64 (V_SET0)), - (v2f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm)), - (i8 1))), sub_xmm)>; - def : Pat<(v8i64 (X86vzmovl (v8i64 VR512:$src))), - (SUBREG_TO_REG (i32 0), - (v2i64 (VPBLENDWrri (v2i64 (V_SET0)), - (v2i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm)), - (i8 0xf))), sub_xmm)>; } let Predicates = [HasAVX512] in { @@ -4452,6 +4421,28 @@ let Predicates = [HasAVX512] in { (SUBREG_TO_REG (i32 0), (v4i32 (VMOVDI2PDIZrm addr:$src)), sub_xmm)>; def : Pat<(v8i64 (X86vzload addr:$src)), (SUBREG_TO_REG (i64 0), (v2i64 (VMOVQI2PQIZrm addr:$src)), sub_xmm)>; + + def : Pat<(v4f64 (X86vzmovl (v4f64 VR256X:$src))), + (SUBREG_TO_REG (i32 0), + (v2f64 (VMOVZPQILo2PQIZrr + (v2f64 (EXTRACT_SUBREG (v4f64 VR256X:$src), sub_xmm)))), + sub_xmm)>; + def : Pat<(v4i64 (X86vzmovl (v4i64 VR256X:$src))), + (SUBREG_TO_REG (i32 0), + (v2i64 (VMOVZPQILo2PQIZrr + (v2i64 (EXTRACT_SUBREG (v4i64 VR256X:$src), sub_xmm)))), + sub_xmm)>; + + def : Pat<(v8f64 (X86vzmovl (v8f64 VR512:$src))), + (SUBREG_TO_REG (i32 0), + (v2f64 (VMOVZPQILo2PQIZrr + (v2f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm)))), + sub_xmm)>; + def : Pat<(v8i64 (X86vzmovl (v8i64 VR512:$src))), + (SUBREG_TO_REG (i32 0), + (v2i64 (VMOVZPQILo2PQIZrr + (v2i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm)))), + sub_xmm)>; } //===----------------------------------------------------------------------===// diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td index c96bac6828f..e25d2dca404 100644 --- a/llvm/lib/Target/X86/X86InstrSSE.td +++ b/llvm/lib/Target/X86/X86InstrSSE.td @@ -312,17 +312,6 @@ let Predicates = [UseAVX, OptForSize] in { (SUBREG_TO_REG (i32 0), (v4i32 (VMOVSSrr (v4i32 (V_SET0)), (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)))), sub_xmm)>; - - def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))), - (SUBREG_TO_REG (i32 0), - (v2f64 (VMOVSDrr (v2f64 (V_SET0)), - (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)))), - sub_xmm)>; - def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))), - (SUBREG_TO_REG (i32 0), - (v2i64 (VMOVSDrr (v2i64 (V_SET0)), - (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)))), - sub_xmm)>; } let Predicates = [UseSSE1] in { @@ -4307,6 +4296,19 @@ let Predicates = [UseSSE2] in { (MOVZPQILo2PQIrr VR128:$src)>; } +let Predicates = [UseAVX] in { + def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))), + (SUBREG_TO_REG (i32 0), + (v2f64 (VMOVZPQILo2PQIrr + (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)))), + sub_xmm)>; + def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))), + (SUBREG_TO_REG (i32 0), + (v2i64 (VMOVZPQILo2PQIrr + (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)))), + sub_xmm)>; +} + //===---------------------------------------------------------------------===// // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP //===---------------------------------------------------------------------===// @@ -6319,17 +6321,6 @@ let Predicates = [HasAVX, OptForSpeed] in { (v4i32 (VPBLENDWrri (v4i32 (V_SET0)), (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), (i8 3))), sub_xmm)>; - - def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))), - (SUBREG_TO_REG (i32 0), - (v2f64 (VBLENDPDrri (v2f64 (V_SET0)), - (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), - (i8 1))), sub_xmm)>; - def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))), - (SUBREG_TO_REG (i32 0), - (v2i64 (VPBLENDWrri (v2i64 (V_SET0)), - (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), - (i8 0xf))), sub_xmm)>; } // Prefer a movss or movsd over a blendps when optimizing for size. these were |
