[WebAssembly] Additional patterns for pmin/pax

[samparker]

Summary

Each operation was missing their inverted condition using olt or ogt. Also, as we don't need to discern +/-0, I think we should also be able to use ole and oge.

[luke]

Nice, LGTM

[tlively]

I don't think this is correct with respect to NaN vector elements. SIMDe is getting test failures starting with this commit, reported here: https://github.com/emscripten-core/emscripten/issues/18759

llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td

 
 // NaN-propagating minimum: min
 defm MIN : SIMDBinaryFP<fminimum, "min", 232>;
 
 // NaN-propagating maximum: max
 defm MAX : SIMDBinaryFP<fmaximum, "max", 233>;
 
 // Pseudo-minimum: pmin
-def pmin : PatFrag<(ops node:$lhs, node:$rhs),
-                   (vselect (setolt $rhs, $lhs), $rhs, $lhs)>;
+def pmin : PatFrags<(ops node:$lhs, node:$rhs), [
+                    (vselect (setolt $lhs, $rhs), $lhs, $rhs),
+                    (vselect (setole $lhs, $rhs), $lhs, $rhs),
+                    (vselect (setogt $rhs, $lhs), $lhs, $rhs),
+                    (vselect (setoge $rhs, $lhs), $lhs, $rhs)
+]>;
 defm PMIN : SIMDBinaryFP<pmin, "pmin", 234>;
 
 // Pseudo-maximum: pmax
-def pmax : PatFrag<(ops node:$lhs, node:$rhs),
-                   (vselect (setolt $lhs, $rhs), $rhs, $lhs)>;
+def pmax : PatFrags<(ops node:$lhs, node:$rhs), [
+                    (vselect (setogt $lhs, $rhs), $lhs, $rhs),
+                    (vselect (setoge $lhs, $rhs), $lhs, $rhs),
+                    (vselect (setolt $rhs, $lhs), $lhs, $rhs),
+                    (vselect (setole $rhs, $lhs), $lhs, $rhs)
+]>;
 defm PMAX : SIMDBinaryFP<pmax, "pmax", 235>;
 
 // Also match the pmin/pmax cases where the operands are int vectors (but the
 // comparison is still a floating point comparison). This can happen when using
 // the wasm_simd128.h intrinsics because v128_t is an integer vector.
 foreach vec = [F32x4, F64x2] in {
 defvar pmin = !cast<NI>("PMIN_"#vec);
 defvar pmax = !cast<NI>("PMAX_"#vec);

llvm/test/CodeGen/WebAssembly/simd-arith.ll

   %a = call <4 x float> @llvm.fabs.v4f32(<4 x float> %x)
   ret <4 x float> %a
 }
 
 define <4 x float> @min_unordered_v4f32(<4 x float> %x) {
 ; SIMD128-LABEL: min_unordered_v4f32:
 ; SIMD128:         .functype min_unordered_v4f32 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f32x4.gt $push0=, $0, $1
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f32x4.pmin $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: min_unordered_v4f32:
 ; SIMD128-FAST:         .functype min_unordered_v4f32 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f32x4.gt $push1=, $0, $1
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f32x4.pmin $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: min_unordered_v4f32:
 ; NO-SIMD128:         .functype min_unordered_v4f32 (i32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.const $push17=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.gt $push1=, $3, $pop17

   ret <4 x float> %a
 }
 
 define <4 x float> @max_unordered_v4f32(<4 x float> %x) {
 ; SIMD128-LABEL: max_unordered_v4f32:
 ; SIMD128:         .functype max_unordered_v4f32 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
 ; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    f32x4.pmax $push1=, $0, $pop0
+; SIMD128-NEXT:    f32x4.pmax $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: max_unordered_v4f32:
 ; SIMD128-FAST:         .functype max_unordered_v4f32 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
 ; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    f32x4.pmax $push0=, $0, $pop1
+; SIMD128-FAST-NEXT:    f32x4.pmax $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: max_unordered_v4f32:
 ; NO-SIMD128:         .functype max_unordered_v4f32 (i32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.const $push17=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.lt $push1=, $3, $pop17

     <4 x float> <float 5., float 5., float 5., float 5.>
   ret <4 x float> %a
 }
 
 define <4 x float> @min_ordered_v4f32(<4 x float> %x) {
 ; SIMD128-LABEL: min_ordered_v4f32:
 ; SIMD128:         .functype min_ordered_v4f32 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f32x4.le $push0=, $1, $0
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f32x4.pmin $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: min_ordered_v4f32:
 ; SIMD128-FAST:         .functype min_ordered_v4f32 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f32x4.le $push1=, $1, $0
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f32x4.pmin $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: min_ordered_v4f32:
 ; NO-SIMD128:         .functype min_ordered_v4f32 (i32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.const $push17=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.ge $push1=, $3, $pop17

     <4 x float> <float 5., float 5., float 5., float 5.>, <4 x float> %x
   ret <4 x float> %a
 }
 
 define <4 x float> @max_ordered_v4f32(<4 x float> %x) {
 ; SIMD128-LABEL: max_ordered_v4f32:
 ; SIMD128:         .functype max_ordered_v4f32 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f32x4.ge $push0=, $1, $0
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f32x4.pmax $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: max_ordered_v4f32:
 ; SIMD128-FAST:         .functype max_ordered_v4f32 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f32x4.ge $push1=, $1, $0
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f32x4.pmax $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: max_ordered_v4f32:
 ; NO-SIMD128:         .functype max_ordered_v4f32 (i32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.const $push17=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f32.le $push1=, $3, $pop17

   )
   ret <4 x float> %a
 }
 
 define <4 x float> @pmin_v4f32(<4 x float> %x, <4 x float> %y) {
 ; SIMD128-LABEL: pmin_v4f32:
 ; SIMD128:         .functype pmin_v4f32 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f32x4.pmin $push0=, $0, $1
+; SIMD128-NEXT:    f32x4.pmin $push0=, $1, $0
 ; SIMD128-NEXT:    return $pop0
 ;
 ; SIMD128-FAST-LABEL: pmin_v4f32:
 ; SIMD128-FAST:         .functype pmin_v4f32 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f32x4.pmin $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f32x4.pmin $push0=, $1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: pmin_v4f32:
 ; NO-SIMD128:         .functype pmin_v4f32 (i32, f32, f32, f32, f32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.lt $push0=, $7, $3
 ; NO-SIMD128-NEXT:    f32.select $push1=, $7, $3, $pop0
 ; NO-SIMD128-NEXT:    f32.store 8($0), $pop1

   %a = select <4 x i1> %c, <4 x i32> %y, <4 x i32> %x
   ret <4 x i32> %a
 }
 
 define <4 x float> @pmax_v4f32(<4 x float> %x, <4 x float> %y) {
 ; SIMD128-LABEL: pmax_v4f32:
 ; SIMD128:         .functype pmax_v4f32 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f32x4.pmax $push0=, $0, $1
+; SIMD128-NEXT:    f32x4.pmax $push0=, $1, $0
 ; SIMD128-NEXT:    return $pop0
 ;
 ; SIMD128-FAST-LABEL: pmax_v4f32:
 ; SIMD128-FAST:         .functype pmax_v4f32 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f32x4.pmax $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f32x4.pmax $push0=, $1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: pmax_v4f32:
 ; NO-SIMD128:         .functype pmax_v4f32 (i32, f32, f32, f32, f32, f32, f32, f32, f32) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f32.lt $push0=, $3, $7
 ; NO-SIMD128-NEXT:    f32.select $push1=, $7, $3, $pop0
 ; NO-SIMD128-NEXT:    f32.store 8($0), $pop1

   %a = call <2 x double> @llvm.fabs.v2f64(<2 x double> %x)
   ret <2 x double> %a
 }
 
 define <2 x double> @min_unordered_v2f64(<2 x double> %x) {
 ; SIMD128-LABEL: min_unordered_v2f64:
 ; SIMD128:         .functype min_unordered_v2f64 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f64x2.gt $push0=, $0, $1
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f64x2.pmin $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: min_unordered_v2f64:
 ; SIMD128-FAST:         .functype min_unordered_v2f64 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f64x2.gt $push1=, $0, $1
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f64x2.pmin $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: min_unordered_v2f64:
 ; NO-SIMD128:         .functype min_unordered_v2f64 (i32, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.const $push7=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.gt $push1=, $2, $pop7

   ret <2 x double> %a
 }
 
 define <2 x double> @max_unordered_v2f64(<2 x double> %x) {
 ; SIMD128-LABEL: max_unordered_v2f64:
 ; SIMD128:         .functype max_unordered_v2f64 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
 ; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    f64x2.pmax $push1=, $0, $pop0
+; SIMD128-NEXT:    f64x2.pmax $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: max_unordered_v2f64:
 ; SIMD128-FAST:         .functype max_unordered_v2f64 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
 ; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    f64x2.pmax $push0=, $0, $pop1
+; SIMD128-FAST-NEXT:    f64x2.pmax $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: max_unordered_v2f64:
 ; NO-SIMD128:         .functype max_unordered_v2f64 (i32, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.const $push7=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.lt $push1=, $2, $pop7

     <2 x double> <double 5., double 5.>
   ret <2 x double> %a
 }
 
 define <2 x double> @min_ordered_v2f64(<2 x double> %x) {
 ; SIMD128-LABEL: min_ordered_v2f64:
 ; SIMD128:         .functype min_ordered_v2f64 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f64x2.le $push0=, $1, $0
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f64x2.pmin $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: min_ordered_v2f64:
 ; SIMD128-FAST:         .functype min_ordered_v2f64 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f64x2.le $push1=, $1, $0
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f64x2.pmin $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: min_ordered_v2f64:
 ; NO-SIMD128:         .functype min_ordered_v2f64 (i32, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.const $push7=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.ge $push1=, $2, $pop7

     <2 x double> %x
   ret <2 x double> %a
 }
 
 define <2 x double> @max_ordered_v2f64(<2 x double> %x) {
 ; SIMD128-LABEL: max_ordered_v2f64:
 ; SIMD128:         .functype max_ordered_v2f64 (v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-NEXT:    f64x2.ge $push0=, $1, $0
-; SIMD128-NEXT:    v128.bitselect $push1=, $pop2, $0, $pop0
+; SIMD128-NEXT:    v128.const $push0=, 0x1.4p2, 0x1.4p2
+; SIMD128-NEXT:    f64x2.pmax $push1=, $pop0, $0
 ; SIMD128-NEXT:    return $pop1
 ;
 ; SIMD128-FAST-LABEL: max_ordered_v2f64:
 ; SIMD128-FAST:         .functype max_ordered_v2f64 (v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    v128.const $push3=, 0x1.4p2, 0x1.4p2
-; SIMD128-FAST-NEXT:    local.tee $push2=, $1=, $pop3
-; SIMD128-FAST-NEXT:    f64x2.ge $push1=, $1, $0
-; SIMD128-FAST-NEXT:    v128.bitselect $push0=, $pop2, $0, $pop1
+; SIMD128-FAST-NEXT:    v128.const $push1=, 0x1.4p2, 0x1.4p2
+; SIMD128-FAST-NEXT:    f64x2.pmax $push0=, $pop1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: max_ordered_v2f64:
 ; NO-SIMD128:         .functype max_ordered_v2f64 (i32, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.const $push0=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.const $push7=, 0x1.4p2
 ; NO-SIMD128-NEXT:    f64.le $push1=, $2, $pop7

   )
   ret <2 x double> %a
 }
 
 define <2 x double> @pmin_v2f64(<2 x double> %x, <2 x double> %y) {
 ; SIMD128-LABEL: pmin_v2f64:
 ; SIMD128:         .functype pmin_v2f64 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f64x2.pmin $push0=, $0, $1
+; SIMD128-NEXT:    f64x2.pmin $push0=, $1, $0
 ; SIMD128-NEXT:    return $pop0
 ;
 ; SIMD128-FAST-LABEL: pmin_v2f64:
 ; SIMD128-FAST:         .functype pmin_v2f64 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f64x2.pmin $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f64x2.pmin $push0=, $1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: pmin_v2f64:
 ; NO-SIMD128:         .functype pmin_v2f64 (i32, f64, f64, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.lt $push0=, $4, $2
 ; NO-SIMD128-NEXT:    f64.select $push1=, $4, $2, $pop0
 ; NO-SIMD128-NEXT:    f64.store 8($0), $pop1

   %a = select <2 x i1> %c, <2 x i64> %y, <2 x i64> %x
   ret <2 x i64> %a
 }
 
 define <2 x double> @pmax_v2f64(<2 x double> %x, <2 x double> %y) {
 ; SIMD128-LABEL: pmax_v2f64:
 ; SIMD128:         .functype pmax_v2f64 (v128, v128) -> (v128)
 ; SIMD128-NEXT:  # %bb.0:
-; SIMD128-NEXT:    f64x2.pmax $push0=, $0, $1
+; SIMD128-NEXT:    f64x2.pmax $push0=, $1, $0
 ; SIMD128-NEXT:    return $pop0
 ;
 ; SIMD128-FAST-LABEL: pmax_v2f64:
 ; SIMD128-FAST:         .functype pmax_v2f64 (v128, v128) -> (v128)
 ; SIMD128-FAST-NEXT:  # %bb.0:
-; SIMD128-FAST-NEXT:    f64x2.pmax $push0=, $0, $1
+; SIMD128-FAST-NEXT:    f64x2.pmax $push0=, $1, $0
 ; SIMD128-FAST-NEXT:    return $pop0
 ;
 ; NO-SIMD128-LABEL: pmax_v2f64:
 ; NO-SIMD128:         .functype pmax_v2f64 (i32, f64, f64, f64, f64) -> ()
 ; NO-SIMD128-NEXT:  # %bb.0:
 ; NO-SIMD128-NEXT:    f64.lt $push0=, $2, $4
 ; NO-SIMD128-NEXT:    f64.select $push1=, $4, $2, $pop0
 ; NO-SIMD128-NEXT:    f64.store 8($0), $pop1