llvm/test/CodeGen/X86/avx10_2bf16-arith.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-unknown-unknown --show-mc-encoding -mattr=+avx10.2-256 | FileCheck %s --check-prefixes=CHECK,X64
; RUN: llc < %s -verify-machineinstrs -mtriple=i686-unknown-unknown --show-mc-encoding -mattr=+avx10.2-256 | FileCheck %s --check-prefixes=CHECK,X86

define <16 x bfloat> @test_int_x86_avx10_add_bf16_256(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_add_bf16_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vaddbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x58,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fadd <16 x bfloat> %x1, %x2
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_mask_add_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_add_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vaddbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x58,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_add_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vaddbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x58,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %res0 = fadd <16 x bfloat> %x1, %x2
  %res = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> %src
  ret <16 x bfloat> %res
}
define <16 x bfloat> @test_int_x86_avx10_maskz_add_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_add_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vaddbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x58,0xc2]
; X64-NEXT:    vaddbf16 (%rsi), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x58,0x0e]
; X64-NEXT:    vaddbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x58,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_add_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vaddbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x58,0xc2]
; X86-NEXT:    vaddbf16 (%eax), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x58,0x08]
; X86-NEXT:    vaddbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x58,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %val = load <16 x bfloat>, ptr %ptr
  %res0 = fadd <16 x bfloat> %x1, %x2
  %res1 = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> zeroinitializer
  %t2 = fadd <16 x bfloat> %x1, %val
  %res2 = select <16 x i1> %mask, <16 x bfloat> %t2, <16 x bfloat> zeroinitializer
  %res3  = fadd <16 x bfloat> %res1, %res2
  ret <16 x bfloat> %res3
}

define <8 x bfloat> @test_int_x86_avx10_add_bf16_128(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_add_bf16_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vaddbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x58,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fadd <8 x bfloat> %x1, %x2
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_mask_add_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_add_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vaddbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x58,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_add_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vaddbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x58,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %res0 = fadd <8 x bfloat> %x1, %x2
  %res = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> %src
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_maskz_add_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_add_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vaddbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x58,0xc2]
; X64-NEXT:    vaddbf16 (%rsi), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x58,0x0e]
; X64-NEXT:    vaddbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x58,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_add_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vaddbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x58,0xc2]
; X86-NEXT:    vaddbf16 (%eax), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x58,0x08]
; X86-NEXT:    vaddbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x58,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %val = load <8 x bfloat>, ptr %ptr
  %res0 = fadd <8 x bfloat> %x1, %x2
  %res1 = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> zeroinitializer
  %t2 = fadd <8 x bfloat> %x1, %val
  %res2 = select <8 x i1> %mask, <8 x bfloat> %t2, <8 x bfloat> zeroinitializer
  %res3  = fadd <8 x bfloat> %res1, %res2
  ret <8 x bfloat> %res3
}

define <16 x bfloat> @test_int_x86_avx10_sub_bf16_256(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_sub_bf16_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsubbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x5c,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fsub <16 x bfloat> %x1, %x2
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_mask_sub_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_sub_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsubbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x5c,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_sub_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsubbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x5c,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %res0 = fsub <16 x bfloat> %x1, %x2
  %res = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> %src
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_maskz_sub_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_sub_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsubbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5c,0xc2]
; X64-NEXT:    vsubbf16 (%rsi), %ymm1, %ymm1 # encoding: [0x62,0xf5,0x75,0x28,0x5c,0x0e]
; X64-NEXT:    vsubbf16 %ymm1, %ymm0, %ymm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x5c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_sub_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    vsubbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5c,0xc2]
; X86-NEXT:    vsubbf16 (%eax), %ymm1, %ymm1 # encoding: [0x62,0xf5,0x75,0x28,0x5c,0x08]
; X86-NEXT:    vsubbf16 %ymm1, %ymm0, %ymm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x5c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %val = load <16 x bfloat>, ptr %ptr
  %res0 = fsub <16 x bfloat> %x1, %x2
  %res1 = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> zeroinitializer
  %t2 = fsub <16 x bfloat> %x1, %val
  %res2 = select <16 x i1> %mask, <16 x bfloat> %t2, <16 x bfloat> zeroinitializer
  %res3  = fsub <16 x bfloat> %res1, %res2
  ret <16 x bfloat> %res3
}

define <8 x bfloat> @test_int_x86_avx10_sub_bf16_128(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_sub_bf16_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsubbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x5c,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fsub <8 x bfloat> %x1, %x2
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_mask_sub_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_sub_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsubbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x5c,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_sub_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsubbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x5c,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %res0 = fsub <8 x bfloat> %x1, %x2
  %res = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> %src
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_maskz_sub_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_sub_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsubbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5c,0xc2]
; X64-NEXT:    vsubbf16 (%rsi), %xmm1, %xmm1 # encoding: [0x62,0xf5,0x75,0x08,0x5c,0x0e]
; X64-NEXT:    vsubbf16 %xmm1, %xmm0, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x5c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_sub_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    vsubbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5c,0xc2]
; X86-NEXT:    vsubbf16 (%eax), %xmm1, %xmm1 # encoding: [0x62,0xf5,0x75,0x08,0x5c,0x08]
; X86-NEXT:    vsubbf16 %xmm1, %xmm0, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x5c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %val = load <8 x bfloat>, ptr %ptr
  %res0 = fsub <8 x bfloat> %x1, %x2
  %res1 = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> zeroinitializer
  %t2 = fsub <8 x bfloat> %x1, %val
  %res2 = select <8 x i1> %mask, <8 x bfloat> %t2, <8 x bfloat> zeroinitializer
  %res3  = fsub <8 x bfloat> %res1, %res2
  ret <8 x bfloat> %res3
}

define <16 x bfloat> @test_int_x86_avx10_mul_bf16_256(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_mul_bf16_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmulbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x59,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fmul <16 x bfloat> %x1, %x2
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_mask_mul_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_mul_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vmulbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x59,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_mul_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vmulbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x59,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %res0 = fmul <16 x bfloat> %x1, %x2
  %res = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> %src
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_maskz_mul_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_mul_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vmulbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x59,0xc2]
; X64-NEXT:    vmulbf16 (%rsi), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x59,0x0e]
; X64-NEXT:    vmulbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x59,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_mul_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vmulbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x59,0xc2]
; X86-NEXT:    vmulbf16 (%eax), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x59,0x08]
; X86-NEXT:    vmulbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x59,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %val = load <16 x bfloat>, ptr %ptr
  %res0 = fmul <16 x bfloat> %x1, %x2
  %res1 = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> zeroinitializer
  %t2 = fmul <16 x bfloat> %x1, %val
  %res2 = select <16 x i1> %mask, <16 x bfloat> %t2, <16 x bfloat> zeroinitializer
  %res3  = fmul <16 x bfloat> %res1, %res2
  ret <16 x bfloat> %res3
}

define <8 x bfloat> @test_int_x86_avx10_mul_bf16_128(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_mul_bf16_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmulbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x59,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fmul <8 x bfloat> %x1, %x2
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_mask_mul_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_mul_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vmulbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x59,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_mul_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vmulbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x59,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %res0 = fmul <8 x bfloat> %x1, %x2
  %res = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> %src
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_maskz_mul_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_mul_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vmulbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x59,0xc2]
; X64-NEXT:    vmulbf16 (%rsi), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x59,0x0e]
; X64-NEXT:    vmulbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x59,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_mul_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vmulbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x59,0xc2]
; X86-NEXT:    vmulbf16 (%eax), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x59,0x08]
; X86-NEXT:    vmulbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x59,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %val = load <8 x bfloat>, ptr %ptr
  %res0 = fmul <8 x bfloat> %x1, %x2
  %res1 = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> zeroinitializer
  %t2 = fmul <8 x bfloat> %x1, %val
  %res2 = select <8 x i1> %mask, <8 x bfloat> %t2, <8 x bfloat> zeroinitializer
  %res3  = fmul <8 x bfloat> %res1, %res2
  ret <8 x bfloat> %res3
}

define <16 x bfloat> @test_int_x86_avx10_div_bf16_256(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_div_bf16_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vdivbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x5e,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fdiv <16 x bfloat> %x1, %x2
  ret <16 x bfloat> %res
}

define <16 x bfloat> @test_int_x86_avx10_mask_div_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_div_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vdivbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x5e,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_div_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vdivbf16 %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x75,0x29,0x5e,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %res0 = fdiv <16 x bfloat> %x1, %x2
  %res = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> %src
  ret <16 x bfloat> %res
}

; FIXME: assembly order is different from fp16 ones
define <16 x bfloat> @test_int_x86_avx10_maskz_div_bf16_256(<16 x bfloat> %src, <16 x bfloat> %x1, <16 x bfloat> %x2, i16 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_div_bf16_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vdivbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5e,0xc2]
; X64-NEXT:    vdivbf16 (%rsi), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5e,0x0e]
; X64-NEXT:    vdivbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x5e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_div_bf16_256:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vdivbf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5e,0xc2]
; X86-NEXT:    vdivbf16 (%eax), %ymm1, %ymm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0xa9,0x5e,0x08]
; X86-NEXT:    vdivbf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x5e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i16 %msk to <16 x i1>
  %val = load <16 x bfloat>, ptr %ptr
  %res0 = fdiv <16 x bfloat> %x1, %x2
  %res1 = select <16 x i1> %mask, <16 x bfloat> %res0, <16 x bfloat> zeroinitializer
  %t2 = fdiv <16 x bfloat> %x1, %val
  %res2 = select <16 x i1> %mask, <16 x bfloat> %t2, <16 x bfloat> zeroinitializer
  %res3  = fdiv <16 x bfloat> %res1, %res2
  ret <16 x bfloat> %res3
}

define <8 x bfloat> @test_int_x86_avx10_div_bf16_128(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_div_bf16_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vdivbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x5e,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = fdiv <8 x bfloat> %x1, %x2
  ret <8 x bfloat> %res
}

define <8 x bfloat> @test_int_x86_avx10_mask_div_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_mask_div_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vdivbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x5e,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_mask_div_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vdivbf16 %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x75,0x09,0x5e,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %res0 = fdiv <8 x bfloat> %x1, %x2
  %res = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> %src
  ret <8 x bfloat> %res
}

; FIXME: assembly order is different from fp16 ones
define <8 x bfloat> @test_int_x86_avx10_maskz_div_bf16_128(<8 x bfloat> %src, <8 x bfloat> %x1, <8 x bfloat> %x2, i8 %msk, ptr %ptr) {
; X64-LABEL: test_int_x86_avx10_maskz_div_bf16_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vdivbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5e,0xc2]
; X64-NEXT:    vdivbf16 (%rsi), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5e,0x0e]
; X64-NEXT:    vdivbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x5e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_int_x86_avx10_maskz_div_bf16_128:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vdivbf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5e,0xc2]
; X86-NEXT:    vdivbf16 (%eax), %xmm1, %xmm1 {%k1} {z} # encoding: [0x62,0xf5,0x75,0x89,0x5e,0x08]
; X86-NEXT:    vdivbf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x5e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
  %mask = bitcast i8 %msk to <8 x i1>
  %val = load <8 x bfloat>, ptr %ptr
  %res0 = fdiv <8 x bfloat> %x1, %x2
  %res1 = select <8 x i1> %mask, <8 x bfloat> %res0, <8 x bfloat> zeroinitializer
  %t2 = fdiv <8 x bfloat> %x1, %val
  %res2 = select <8 x i1> %mask, <8 x bfloat> %t2, <8 x bfloat> zeroinitializer
  %res3  = fdiv <8 x bfloat> %res1, %res2
  ret <8 x bfloat> %res3
}

define i16 @test_int_x86_avx10_vcmpbf16256(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_vcmpbf16256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vcmpunordbf16 %ymm1, %ymm0, %k0 # encoding: [0x62,0xf3,0x7f,0x28,0xc2,0xc1,0x03]
; CHECK-NEXT:    kmovd %k0, %eax # encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fcmp uno <16 x bfloat> %x1, %x2
  %res = bitcast <16 x i1> %1 to i16
  ret i16 %res
}

define i16 @test_int_x86_avx10_vcmpbf16256_mask2(<16 x bfloat> %x1, <16 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_vcmpbf16256_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vcmpeqbf16 %ymm1, %ymm0, %k0 # encoding: [0x62,0xf3,0x7f,0x28,0xc2,0xc1,0x00]
; CHECK-NEXT:    kmovd %k0, %eax # encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT:    andl $3, %eax # encoding: [0x83,0xe0,0x03]
; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fcmp oeq <16 x bfloat> %x1, %x2
  %2 = and <16 x i1> %1, <i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>
  %3 = bitcast <16 x i1> %2 to i16
  ret i16 %3
}

define i8 @test_int_x86_avx10_vcmpbf16128(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_vcmpbf16128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vcmpunordbf16 %xmm1, %xmm0, %k0 # encoding: [0x62,0xf3,0x7f,0x08,0xc2,0xc1,0x03]
; CHECK-NEXT:    kmovd %k0, %eax # encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT:    # kill: def $al killed $al killed $eax
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fcmp uno <8 x bfloat> %x1, %x2
  %res = bitcast <8 x i1> %1 to i8
  ret i8 %res
}

define i8 @test_int_x86_avx10_vcmpbf16128_mask2(<8 x bfloat> %x1, <8 x bfloat> %x2) {
; CHECK-LABEL: test_int_x86_avx10_vcmpbf16128_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vcmpeqbf16 %xmm1, %xmm0, %k0 # encoding: [0x62,0xf3,0x7f,0x08,0xc2,0xc1,0x00]
; CHECK-NEXT:    kmovd %k0, %eax # encoding: [0xc5,0xfb,0x93,0xc0]
; CHECK-NEXT:    andb $3, %al # encoding: [0x24,0x03]
; CHECK-NEXT:    # kill: def $al killed $al killed $eax
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fcmp oeq <8 x bfloat> %x1, %x2
  %2 = and <8 x i1> %1, <i1 true, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>
  %3 = bitcast <8 x i1> %2 to i8
  ret i8 %3
}

define <16 x bfloat> @test_sqrt_bf16_256(<16 x bfloat> %a0) {
; CHECK-LABEL: test_sqrt_bf16_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsqrtbf16 %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x51,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = tail call <16 x bfloat> @llvm.sqrt.v16bf16(<16 x bfloat> %a0)
  ret <16 x bfloat> %1
}

define <16 x bfloat> @test_mm256_mask_sqrt_pbh(<16 x bfloat> %__W, i16 %__U, <16 x bfloat> %__A) {
; X64-LABEL: test_mm256_mask_sqrt_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsqrtbf16 %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x51,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask_sqrt_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsqrtbf16 %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x29,0x51,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.sqrt.v16bf16(<16 x bfloat> %__A)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__W
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_maskz_sqrt_pbh(i16 %__U, <16 x bfloat>%__A) {
; X64-LABEL: test_mm256_maskz_sqrt_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsqrtbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0xa9,0x51,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_maskz_sqrt_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsqrtbf16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0xa9,0x51,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.sqrt.v16bf16(<16 x bfloat> %__A)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer
  ret <16 x bfloat> %2
}

define <8 x bfloat> @test_sqrt_bf16_128(<8 x bfloat> %a0) {
; CHECK-LABEL: test_sqrt_bf16_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vsqrtbf16 %xmm0, %xmm0 # encoding: [0x62,0xf5,0x7d,0x08,0x51,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = tail call <8 x bfloat> @llvm.sqrt.v8bf16(<8 x bfloat> %a0)
  ret <8 x bfloat> %1
}

define <8 x bfloat> @test_mm_mask_sqrt_pbh(<8 x bfloat> %__W, i8 %__U, <8 x bfloat> %__A) {
; X64-LABEL: test_mm_mask_sqrt_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsqrtbf16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x51,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask_sqrt_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsqrtbf16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7d,0x09,0x51,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.sqrt.v8bf16(<8 x bfloat> %__A)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__W
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_maskz_sqrt_pbh(i8 %__U, <8 x bfloat>%__A) {
; X64-LABEL: test_mm_maskz_sqrt_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vsqrtbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0x89,0x51,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_maskz_sqrt_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vsqrtbf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf5,0x7d,0x89,0x51,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.sqrt.v8bf16(<8 x bfloat> %__A)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> zeroinitializer
  ret <8 x bfloat> %2
}

define <16 x bfloat> @test_mm256_fmaddne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; CHECK-LABEL: test_mm256_fmaddne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfmadd213bf16 %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x74,0x28,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C)
  ret <16 x bfloat> %0
}

define <16 x bfloat> @test_mm256_mask_fmaddne_pbh(<16 x bfloat> %__A, i16 zeroext %__U, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_mask_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__A
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_mask3_fmaddne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C, i16 zeroext %__U) {
; X64-LABEL: test_mm256_mask3_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xb8,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask3_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xb8,0xd1]
; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__C
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_maskz_fmaddne_pbh(i16 zeroext %__U, <16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_maskz_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xa8,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_maskz_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xa8,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_fmsubne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; CHECK-LABEL: test_mm256_fmsubne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfmsub213bf16 %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x74,0x28,0xaa,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %fneg.i)
  ret <16 x bfloat> %0
}

define <16 x bfloat> @test_mm256_mask_fmsubne_pbh(<16 x bfloat> %__A, i16 zeroext %__U, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_mask_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9a,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9a,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %fneg.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__A
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_mask3_fmsubne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C, i16 zeroext %__U) {
; X64-LABEL: test_mm256_mask3_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xba,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask3_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xba,0xd1]
; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %fneg.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__C
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_maskz_fmsubne_pbh(i16 zeroext %__U, <16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_maskz_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xaa,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_maskz_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xaa,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %fneg.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_fnmaddne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; CHECK-LABEL: test_mm256_fnmaddne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfnmadd213bf16 %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x74,0x28,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <16 x bfloat> %__B
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i, <16 x bfloat> %__C)
  ret <16 x bfloat> %0
}

define <16 x bfloat> @test_mm256_mask_fnmaddne_pbh(<16 x bfloat> %__A, i16 zeroext %__U, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_mask_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__A
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_mask3_fnmaddne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C, i16 zeroext %__U) {
; X64-LABEL: test_mm256_mask3_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xbc,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask3_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xbc,0xd1]
; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__C
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_maskz_fnmaddne_pbh(i16 zeroext %__U, <16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_maskz_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xac,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_maskz_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xac,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %__C)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_fnmsubne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; CHECK-LABEL: test_mm256_fnmsubne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfnmsub213bf16 %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x74,0x28,0xae,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <16 x bfloat> %__B
  %fneg1.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i, <16 x bfloat> %fneg1.i)
  ret <16 x bfloat> %0
}

define <16 x bfloat> @test_mm256_mask_fnmsubne_pbh(<16 x bfloat> %__A, i16 zeroext %__U, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_mask_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132bf16 %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x29,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %fneg1.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %fneg1.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__A
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_mask3_fnmsubne_pbh(<16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C, i16 zeroext %__U) {
; X64-LABEL: test_mm256_mask3_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xbe,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_mask3_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub231bf16 %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x29,0xbe,0xd1]
; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %fneg1.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %fneg1.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> %__C
  ret <16 x bfloat> %2
}

define <16 x bfloat> @test_mm256_maskz_fnmsubne_pbh(i16 zeroext %__U, <16 x bfloat> %__A, <16 x bfloat> %__B, <16 x bfloat> %__C) {
; X64-LABEL: test_mm256_maskz_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xae,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm256_maskz_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub213bf16 %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0xa9,0xae,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <16 x bfloat> %__B
  %fneg1.i.i = fneg <16 x bfloat> %__C
  %0 = tail call <16 x bfloat> @llvm.fma.v16bf16(<16 x bfloat> %__A, <16 x bfloat> %fneg.i.i, <16 x bfloat> %fneg1.i.i)
  %1 = bitcast i16 %__U to <16 x i1>
  %2 = select <16 x i1> %1, <16 x bfloat> %0, <16 x bfloat> zeroinitializer
  ret <16 x bfloat> %2
}

define <8 x bfloat> @test_mm_fmaddne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; CHECK-LABEL: test_mm_fmaddne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfmadd213bf16 %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x74,0x08,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C)
  ret <8 x bfloat> %0
}

define <8 x bfloat> @test_mm_mask_fmaddne_pbh(<8 x bfloat> %__A, i8 zeroext %__U, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_mask_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__A
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_mask3_fmaddne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C, i8 zeroext %__U) {
; X64-LABEL: test_mm_mask3_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xb8,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask3_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xb8,0xd1]
; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__C
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_maskz_fmaddne_pbh(i8 zeroext %__U, <8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_maskz_fmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xa8,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_maskz_fmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xa8,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> zeroinitializer
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_fmsubne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; CHECK-LABEL: test_mm_fmsubne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfmsub213bf16 %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x74,0x08,0xaa,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %fneg.i)
  ret <8 x bfloat> %0
}

define <8 x bfloat> @test_mm_mask_fmsubne_pbh(<8 x bfloat> %__A, i8 zeroext %__U, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_mask_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9a,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9a,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %fneg.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__A
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_mask3_fmsubne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C, i8 zeroext %__U) {
; X64-LABEL: test_mm_mask3_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xba,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask3_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xba,0xd1]
; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %fneg.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__C
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_maskz_fmsubne_pbh(i8 zeroext %__U, <8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_maskz_fmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xaa,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_maskz_fmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xaa,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %fneg.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> zeroinitializer
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_fnmaddne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; CHECK-LABEL: test_mm_fnmaddne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfnmadd213bf16 %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x74,0x08,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <8 x bfloat> %__B
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i, <8 x bfloat> %__C)
  ret <8 x bfloat> %0
}

define <8 x bfloat> @test_mm_mask_fnmaddne_pbh(<8 x bfloat> %__A, i8 zeroext %__U, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_mask_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__A
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_mask3_fnmaddne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C, i8 zeroext %__U) {
; X64-LABEL: test_mm_mask3_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xbc,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask3_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xbc,0xd1]
; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__C
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_maskz_fnmaddne_pbh(i8 zeroext %__U, <8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_maskz_fnmaddne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xac,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_maskz_fnmaddne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xac,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %__C)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> zeroinitializer
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_fnmsubne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; CHECK-LABEL: test_mm_fnmsubne_pbh:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vfnmsub213bf16 %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x74,0x08,0xae,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %fneg.i = fneg <8 x bfloat> %__B
  %fneg1.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i, <8 x bfloat> %fneg1.i)
  ret <8 x bfloat> %0
}

define <8 x bfloat> @test_mm_mask_fnmsubne_pbh(<8 x bfloat> %__A, i8 zeroext %__U, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_mask_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132bf16 %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6c,0x09,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %fneg1.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %fneg1.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__A
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_mask3_fnmsubne_pbh(<8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C, i8 zeroext %__U) {
; X64-LABEL: test_mm_mask3_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xbe,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_mask3_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub231bf16 %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7c,0x09,0xbe,0xd1]
; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %fneg1.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %fneg1.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> %__C
  ret <8 x bfloat> %2
}

define <8 x bfloat> @test_mm_maskz_fnmsubne_pbh(i8 zeroext %__U, <8 x bfloat> %__A, <8 x bfloat> %__B, <8 x bfloat> %__C) {
; X64-LABEL: test_mm_maskz_fnmsubne_pbh:
; X64:       # %bb.0: # %entry
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xae,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: test_mm_maskz_fnmsubne_pbh:
; X86:       # %bb.0: # %entry
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub213bf16 %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x74,0x89,0xae,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
entry:
  %fneg.i.i = fneg <8 x bfloat> %__B
  %fneg1.i.i = fneg <8 x bfloat> %__C
  %0 = tail call <8 x bfloat> @llvm.fma.v8bf16(<8 x bfloat> %__A, <8 x bfloat> %fneg.i.i, <8 x bfloat> %fneg1.i.i)
  %1 = bitcast i8 %__U to <8 x i1>
  %2 = select <8 x i1> %1, <8 x bfloat> %0, <8 x bfloat> zeroinitializer
  ret <8 x bfloat> %2
}

define <32 x bfloat> @addv(<32 x bfloat> %a, <32 x bfloat> %b) nounwind {
; X64-LABEL: addv:
; X64:       # %bb.0:
; X64-NEXT:    vaddbf16 %ymm2, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x58,0xc2]
; X64-NEXT:    vaddbf16 %ymm3, %ymm1, %ymm1 # encoding: [0x62,0xf5,0x75,0x28,0x58,0xcb]
; X64-NEXT:    retq # encoding: [0xc3]
;
; X86-LABEL: addv:
; X86:       # %bb.0:
; X86-NEXT:    pushl %ebp # encoding: [0x55]
; X86-NEXT:    movl %esp, %ebp # encoding: [0x89,0xe5]
; X86-NEXT:    andl $-32, %esp # encoding: [0x83,0xe4,0xe0]
; X86-NEXT:    subl $32, %esp # encoding: [0x83,0xec,0x20]
; X86-NEXT:    vaddbf16 %ymm2, %ymm0, %ymm0 # encoding: [0x62,0xf5,0x7d,0x28,0x58,0xc2]
; X86-NEXT:    vaddbf16 8(%ebp), %ymm1, %ymm1 # encoding: [0x62,0xf5,0x75,0x28,0x58,0x8d,0x08,0x00,0x00,0x00]
; X86-NEXT:    movl %ebp, %esp # encoding: [0x89,0xec]
; X86-NEXT:    popl %ebp # encoding: [0x5d]
; X86-NEXT:    retl # encoding: [0xc3]
  %add = fadd <32 x bfloat> %a, %b
  ret <32 x bfloat> %add
}