[llvm] [LLVM][CodeGen][SME] Make FMA decision independent of streaming mode. (PR #222598)

Paul Walker via llvm-commits llvm-commits at lists.llvm.org
Thu Sep 10 03:40:39 PDT 2026


https://github.com/paulwalker-arm created https://github.com/llvm/llvm-project/pull/222598

The relevant bfloat instructions are available in streaming mode and thus we can benefit from the same BFMLAL based lowering.

>From d1b1f7092a3e85a072c3aefae1eae498dae80f75 Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker at arm.com>
Date: Wed, 9 Sep 2026 14:31:27 +0000
Subject: [PATCH 1/2] Update tests.

---
 .../test/CodeGen/AArch64/sve-bf16-combines.ll | 1221 +++++++++++------
 1 file changed, 834 insertions(+), 387 deletions(-)

diff --git a/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll b/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
index 9a45f67aad946..61e56f82a1f6f 100644
--- a/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
+++ b/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
@@ -1,172 +1,289 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mattr=+sve,+bf16             < %s | FileCheck %s --check-prefixes=SVE
-; RUN: llc -mattr=+sve,+bf16,+sve-b16b16 < %s | FileCheck %s --check-prefixes=SVE-B16B16
+; RUN: llc -mattr=+sve,+bf16             < %s | FileCheck %s --check-prefixes=SVE,SVE-NONSTREAMING
+; RUN: llc -mattr=+sve,+bf16,+sve-b16b16 < %s | FileCheck %s --check-prefixes=SVE-B16B16,SVE-B16B16-NONSTREAMING
+
+; RUN: llc -mattr=+sme --force-streaming              < %s | FileCheck %s --check-prefixes=SVE,SVE-STREAMING
+; RUN: llc -mattr=+sme2,+sve-b16b16 --force-streaming < %s | FileCheck %s --check-prefixes=SVE-B16B16,B16B16-STREAMING
 
 target triple = "aarch64-unknown-linux-gnu"
 
 define <vscale x 8 x bfloat> @fmla_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale x 8 x bfloat> %m1, <vscale x 8 x bfloat> %m2) {
-; SVE-LABEL: fmla_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    uunpkhi z4.s, z2.h
-; SVE-NEXT:    uunpkhi z5.s, z1.h
-; SVE-NEXT:    uunpklo z2.s, z2.h
-; SVE-NEXT:    uunpklo z1.s, z1.h
-; SVE-NEXT:    ptrue p0.s
-; SVE-NEXT:    zip2 z6.h, z3.h, z0.h
-; SVE-NEXT:    zip1 z0.h, z3.h, z0.h
-; SVE-NEXT:    bfmlalb z6.s, z5.h, z4.h
-; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z1.h, p0/m, z6.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    uzp1 z0.h, z0.h, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    uunpkhi z4.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpkhi z5.s, z1.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-NONSTREAMING-NEXT:    ptrue p0.s
+; SVE-NONSTREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z6.s, z5.h, z4.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p0/m, z6.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.h
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p0.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p0/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    fadd z2.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
   %res = fadd contract <vscale x 8 x bfloat> %acc, %mul
   ret <vscale x 8 x bfloat> %res
 }
 
 define <vscale x 4 x bfloat> @fmla_nxv4bf16(<vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-LABEL: fmla_nxv4bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NEXT:    ptrue p0.s
-; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_nxv4bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    ptrue p0.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.s
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_nxv4bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p0.s
+; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fadd z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %res = fadd contract <vscale x 4 x bfloat> %acc, %mul
   ret <vscale x 4 x bfloat> %res
 }
 
 define <vscale x 2 x bfloat> @fmla_nxv2bf16(<vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-LABEL: fmla_nxv2bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NEXT:    ptrue p0.d
-; SVE-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_nxv2bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    ptrue p0.d
+; SVE-NONSTREAMING-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.d
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_nxv2bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    ptrue p0.d
+; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-STREAMING-NEXT:    fmul z1.s, p0/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fadd z0.s, p0/m, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %res = fadd contract <vscale x 2 x bfloat> %acc, %mul
   ret <vscale x 2 x bfloat> %res
 }
 
 define <vscale x 8 x bfloat> @fmls_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale x 8 x bfloat> %m1, <vscale x 8 x bfloat> %m2) {
-; SVE-LABEL: fmls_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p0.h
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    uunpkhi z5.s, z2.h
-; SVE-NEXT:    uunpklo z2.s, z2.h
-; SVE-NEXT:    fneg z1.h, p0/m, z1.h
-; SVE-NEXT:    ptrue p0.s
-; SVE-NEXT:    zip2 z6.h, z3.h, z0.h
-; SVE-NEXT:    zip1 z0.h, z3.h, z0.h
-; SVE-NEXT:    uunpkhi z4.s, z1.h
-; SVE-NEXT:    uunpklo z1.s, z1.h
-; SVE-NEXT:    bfmlalb z6.s, z4.h, z5.h
-; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z1.h, p0/m, z6.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    uzp1 z0.h, z0.h, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p0.h
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    uunpkhi z5.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ptrue p0.s
+; SVE-NONSTREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    uunpkhi z4.s, z1.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z6.s, z4.h, z5.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p0/m, z6.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.h
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p0.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p0/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    fsub z2.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
   %res = fsub contract <vscale x 8 x bfloat> %acc, %mul
   ret <vscale x 8 x bfloat> %res
 }
 
 define <vscale x 4 x bfloat> @fmls_nxv4bf16(<vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-LABEL: fmls_nxv4bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p0.s
-; SVE-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NEXT:    fneg z1.h, p0/m, z1.h
-; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_nxv4bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p0.s
+; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.s
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_nxv4bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p0.s
+; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fsub z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %res = fsub contract <vscale x 4 x bfloat> %acc, %mul
   ret <vscale x 4 x bfloat> %res
 }
 
 define <vscale x 2 x bfloat> @fmls_nxv2bf16(<vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-LABEL: fmls_nxv2bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p0.d
-; SVE-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NEXT:    fneg z1.h, p0/m, z1.h
-; SVE-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_nxv2bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p0.d
+; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NONSTREAMING-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.d
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_nxv2bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    ptrue p0.d
+; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
+; SVE-STREAMING-NEXT:    fmul z1.s, p0/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fsub z0.s, p0/m, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %res = fsub contract <vscale x 2 x bfloat> %acc, %mul
   ret <vscale x 2 x bfloat> %res
 }
 
 define <vscale x 8 x bfloat> @fmla_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale x 8 x bfloat> %acc, <vscale x 8 x bfloat> %m1, <vscale x 8 x bfloat> %m2) {
-; SVE-LABEL: fmla_sel_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    uunpkhi z4.s, z2.h
-; SVE-NEXT:    uunpkhi z5.s, z1.h
-; SVE-NEXT:    uunpklo z2.s, z2.h
-; SVE-NEXT:    uunpklo z1.s, z1.h
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z6.h, z3.h, z0.h
-; SVE-NEXT:    zip1 z3.h, z3.h, z0.h
-; SVE-NEXT:    bfmlalb z6.s, z5.h, z4.h
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z1.h, p1/m, z6.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    uzp1 z1.h, z2.h, z1.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_sel_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    uunpkhi z4.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpkhi z5.s, z1.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z3.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z6.s, z5.h, z4.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z6.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_sel_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_sel_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
   %add = fadd contract <vscale x 8 x bfloat> %acc, %mul
   %res = select <vscale x 8 x i1> %pred, <vscale x 8 x bfloat> %add, <vscale x 8 x bfloat> %acc
@@ -174,17 +291,29 @@ define <vscale x 8 x bfloat> @fmla_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmla_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-LABEL: fmla_sel_nxv4bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z0.h, p0/m, z3.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_sel_nxv4bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z3.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_sel_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_sel_nxv4bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %add = fadd contract <vscale x 4 x bfloat> %acc, %mul
   %res = select <vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %add, <vscale x 4 x bfloat> %acc
@@ -192,20 +321,33 @@ define <vscale x 4 x bfloat> @fmla_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale
 }
 
 define <vscale x 2 x bfloat> @fmla_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-LABEL: fmla_sel_nxv2bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NEXT:    ptrue p1.d
-; SVE-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmla_sel_nxv2bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    ptrue p1.d
+; SVE-NONSTREAMING-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_sel_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmla_sel_nxv2bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    ptrue p1.d
+; SVE-STREAMING-NEXT:    fmul z1.s, p1/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fadd z1.s, p1/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %add = fadd contract <vscale x 2 x bfloat> %acc, %mul
   %res = select <vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %add, <vscale x 2 x bfloat> %acc
@@ -213,30 +355,52 @@ define <vscale x 2 x bfloat> @fmla_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale
 }
 
 define <vscale x 8 x bfloat> @fmls_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale x 8 x bfloat> %acc, <vscale x 8 x bfloat> %m1, <vscale x 8 x bfloat> %m2) {
-; SVE-LABEL: fmls_sel_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p1.h
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    uunpkhi z5.s, z2.h
-; SVE-NEXT:    uunpklo z2.s, z2.h
-; SVE-NEXT:    fneg z1.h, p1/m, z1.h
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z6.h, z3.h, z0.h
-; SVE-NEXT:    zip1 z3.h, z3.h, z0.h
-; SVE-NEXT:    uunpkhi z4.s, z1.h
-; SVE-NEXT:    uunpklo z1.s, z1.h
-; SVE-NEXT:    bfmlalb z6.s, z4.h, z5.h
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z1.h, p1/m, z6.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    uzp1 z1.h, z2.h, z1.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_sel_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p1.h
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    uunpkhi z5.s, z2.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p1/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z3.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    uunpkhi z4.s, z1.h
+; SVE-NONSTREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z6.s, z4.h, z5.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z6.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_sel_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_sel_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
   %sub = fsub contract <vscale x 8 x bfloat> %acc, %mul
   %res = select <vscale x 8 x i1> %pred, <vscale x 8 x bfloat> %sub, <vscale x 8 x bfloat> %acc
@@ -244,19 +408,31 @@ define <vscale x 8 x bfloat> @fmls_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmls_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-LABEL: fmls_sel_nxv4bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NEXT:    fneg z1.h, p1/m, z1.h
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfcvt z0.h, p0/m, z3.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_sel_nxv4bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p1/m, z1.h
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z3.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_sel_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_sel_nxv4bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %sub = fsub contract <vscale x 4 x bfloat> %acc, %mul
   %res = select <vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %sub, <vscale x 4 x bfloat> %acc
@@ -264,21 +440,34 @@ define <vscale x 4 x bfloat> @fmls_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale
 }
 
 define <vscale x 2 x bfloat> @fmls_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-LABEL: fmls_sel_nxv2bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    ptrue p1.d
-; SVE-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NEXT:    fneg z1.h, p1/m, z1.h
-; SVE-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
-; SVE-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fmls_sel_nxv2bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    ptrue p1.d
+; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NONSTREAMING-NEXT:    fneg z1.h, p1/m, z1.h
+; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NONSTREAMING-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_sel_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fmls_sel_nxv2bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    ptrue p1.d
+; SVE-STREAMING-NEXT:    fmul z1.s, p1/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
+; SVE-STREAMING-NEXT:    fsubr z1.s, p1/m, z1.s, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %sub = fsub contract <vscale x 2 x bfloat> %acc, %mul
   %res = select <vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %sub, <vscale x 2 x bfloat> %acc
@@ -286,78 +475,126 @@ define <vscale x 2 x bfloat> @fmls_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fadd_sel_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfadd z0.h, p0/m, z0.h, z1.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> zeroinitializer
   %fadd = fadd nsz <vscale x 8 x bfloat> %a, %sel
   ret <vscale x 8 x bfloat> %fadd
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fsub_sel_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfsub z0.h, p0/m, z0.h, z1.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> zeroinitializer
   %fsub = fsub <vscale x 8 x bfloat> %a, %sel
   ret <vscale x 8 x bfloat> %fsub
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_negzero_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_negzero_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_negzero_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fadd_sel_negzero_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfadd z0.h, p0/m, z0.h, z1.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_negzero_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %nz
   %fadd = fadd <vscale x 8 x bfloat> %a, %sel
@@ -365,26 +602,42 @@ define <vscale x 8 x bfloat> @fadd_sel_negzero_nxv8bf16(<vscale x 8 x bfloat> %a
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_negzero_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_negzero_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_negzero_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fsub_sel_negzero_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfsub z0.h, p0/m, z0.h, z1.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_negzero_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %nz
   %fsub = fsub nsz <vscale x 8 x bfloat> %a, %sel
@@ -392,35 +645,65 @@ define <vscale x 8 x bfloat> @fsub_sel_negzero_nxv8bf16(<vscale x 8 x bfloat> %a
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_fmul_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_fmul_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    mov z3.s, #0x80000000
-; SVE-NEXT:    mov z4.s, #0x80000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip1 z0.h, z2.h, z0.h
-; SVE-NEXT:    sel z1.h, p0, z1.h, z2.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    fadd z2.s, z4.s, z3.s
-; SVE-NEXT:    fadd z0.s, z0.s, z1.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z2.s
-; SVE-NEXT:    bfcvt z0.h, p1/m, z0.s
-; SVE-NEXT:    uzp1 z0.h, z0.h, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_fmul_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    sel z1.h, p0, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    fadd z2.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z0.s, z0.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z2.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p1/m, z0.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
-; SVE-B16B16-LABEL: fadd_sel_fmul_nxv8bf16:
-; SVE-B16B16:       // %bb.0:
-; SVE-B16B16-NEXT:    movi v3.2d, #0000000000000000
-; SVE-B16B16-NEXT:    bfmul z1.h, z1.h, z2.h
-; SVE-B16B16-NEXT:    sel z1.h, p0, z1.h, z3.h
-; SVE-B16B16-NEXT:    bfadd z0.h, z0.h, z1.h
-; SVE-B16B16-NEXT:    ret
+; SVE-B16B16-NONSTREAMING-LABEL: fadd_sel_fmul_nxv8bf16:
+; SVE-B16B16-NONSTREAMING:       // %bb.0:
+; SVE-B16B16-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-B16B16-NONSTREAMING-NEXT:    bfmul z1.h, z1.h, z2.h
+; SVE-B16B16-NONSTREAMING-NEXT:    sel z1.h, p0, z1.h, z3.h
+; SVE-B16B16-NONSTREAMING-NEXT:    bfadd z0.h, z0.h, z1.h
+; SVE-B16B16-NONSTREAMING-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_fmul_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    sel z1.h, p0, z1.h, z2.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    fadd z2.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p1/m, z0.s
+; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-STREAMING-NEXT:    ret
+;
+; B16B16-STREAMING-LABEL: fadd_sel_fmul_nxv8bf16:
+; B16B16-STREAMING:       // %bb.0:
+; B16B16-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; B16B16-STREAMING-NEXT:    bfmul z1.h, z1.h, z2.h
+; B16B16-STREAMING-NEXT:    sel z1.h, p0, z1.h, z3.h
+; B16B16-STREAMING-NEXT:    bfadd z0.h, z0.h, z1.h
+; B16B16-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> zeroinitializer
   %fadd = fadd contract <vscale x 8 x bfloat> %a, %sel
@@ -428,32 +711,54 @@ define <vscale x 8 x bfloat> @fadd_sel_fmul_nxv8bf16(<vscale x 8 x bfloat> %a, <
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_fmul_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_fmul_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    mov z3.s, #0x80000000
-; SVE-NEXT:    mov z4.s, #0x80000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_fmul_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fsub_sel_fmul_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_fmul_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> zeroinitializer
   %fsub = fsub contract <vscale x 8 x bfloat> %a, %sel
@@ -461,32 +766,54 @@ define <vscale x 8 x bfloat> @fsub_sel_fmul_nxv8bf16(<vscale x 8 x bfloat> %a, <
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_fmul_nsz_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_fmul_nsz_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    movi v4.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_fmul_nsz_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    movi v4.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fadd_sel_fmul_nsz_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_fmul_nsz_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    mov z4.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> zeroinitializer
   %fadd = fadd nsz contract <vscale x 8 x bfloat> %a, %sel
@@ -494,32 +821,54 @@ define <vscale x 8 x bfloat> @fadd_sel_fmul_nsz_nxv8bf16(<vscale x 8 x bfloat> %
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_fmul_nsz_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_fmul_nsz_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    movi v4.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_fmul_nsz_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    movi v4.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fsub_sel_fmul_nsz_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_fmul_nsz_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    mov z4.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> zeroinitializer
   %fsub = fsub nsz contract <vscale x 8 x bfloat> %a, %sel
@@ -527,32 +876,54 @@ define <vscale x 8 x bfloat> @fsub_sel_fmul_nsz_nxv8bf16(<vscale x 8 x bfloat> %
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_fmul_negzero_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_fmul_negzero_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    mov z3.s, #0x80000000
-; SVE-NEXT:    mov z4.s, #0x80000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_fmul_negzero_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fadd_sel_fmul_negzero_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_fmul_negzero_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> %nz
@@ -561,36 +932,71 @@ define <vscale x 8 x bfloat> @fadd_sel_fmul_negzero_nxv8bf16(<vscale x 8 x bfloa
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_fmul_negzero_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    mov z3.s, #0x80000000
-; SVE-NEXT:    mov z4.s, #0x80000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    dupm z2.h, #0x8000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z3.h, z0.h
-; SVE-NEXT:    zip1 z0.h, z3.h, z0.h
-; SVE-NEXT:    sel z1.h, p0, z1.h, z2.h
-; SVE-NEXT:    zip2 z2.h, z3.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z3.h, z1.h
-; SVE-NEXT:    fsub z2.s, z4.s, z2.s
-; SVE-NEXT:    fsub z0.s, z0.s, z1.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z2.s
-; SVE-NEXT:    bfcvt z0.h, p1/m, z0.s
-; SVE-NEXT:    uzp1 z0.h, z0.h, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    dupm z2.h, #0x8000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-NONSTREAMING-NEXT:    sel z1.h, p0, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    zip2 z2.h, z3.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z3.h, z1.h
+; SVE-NONSTREAMING-NEXT:    fsub z2.s, z4.s, z2.s
+; SVE-NONSTREAMING-NEXT:    fsub z0.s, z0.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z2.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p1/m, z0.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
-; SVE-B16B16-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
-; SVE-B16B16:       // %bb.0:
-; SVE-B16B16-NEXT:    dupm z3.h, #0x8000
-; SVE-B16B16-NEXT:    bfmul z1.h, z1.h, z2.h
-; SVE-B16B16-NEXT:    sel z1.h, p0, z1.h, z3.h
-; SVE-B16B16-NEXT:    bfsub z0.h, z0.h, z1.h
-; SVE-B16B16-NEXT:    ret
+; SVE-B16B16-NONSTREAMING-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
+; SVE-B16B16-NONSTREAMING:       // %bb.0:
+; SVE-B16B16-NONSTREAMING-NEXT:    dupm z3.h, #0x8000
+; SVE-B16B16-NONSTREAMING-NEXT:    bfmul z1.h, z1.h, z2.h
+; SVE-B16B16-NONSTREAMING-NEXT:    sel z1.h, p0, z1.h, z3.h
+; SVE-B16B16-NONSTREAMING-NEXT:    bfsub z0.h, z0.h, z1.h
+; SVE-B16B16-NONSTREAMING-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    adrp x8, .LCPI21_0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    ldr h2, [x8, :lo12:.LCPI21_0]
+; SVE-STREAMING-NEXT:    mov z2.h, h2
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    sel z1.h, p0, z1.h, z2.h
+; SVE-STREAMING-NEXT:    zip2 z2.h, z3.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z3.h, z1.h
+; SVE-STREAMING-NEXT:    fsub z2.s, z4.s, z2.s
+; SVE-STREAMING-NEXT:    fsub z0.s, z0.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z2.s
+; SVE-STREAMING-NEXT:    bfcvt z0.h, p1/m, z0.s
+; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
+; SVE-STREAMING-NEXT:    ret
+;
+; B16B16-STREAMING-LABEL: fsub_sel_fmul_negzero_nxv8bf16:
+; B16B16-STREAMING:       // %bb.0:
+; B16B16-STREAMING-NEXT:    adrp x8, .LCPI21_0
+; B16B16-STREAMING-NEXT:    bfmul z1.h, z1.h, z2.h
+; B16B16-STREAMING-NEXT:    ldr h3, [x8, :lo12:.LCPI21_0]
+; B16B16-STREAMING-NEXT:    mov z3.h, h3
+; B16B16-STREAMING-NEXT:    sel z1.h, p0, z1.h, z3.h
+; B16B16-STREAMING-NEXT:    bfsub z0.h, z0.h, z1.h
+; B16B16-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> %nz
@@ -599,32 +1005,54 @@ define <vscale x 8 x bfloat> @fsub_sel_fmul_negzero_nxv8bf16(<vscale x 8 x bfloa
 }
 
 define <vscale x 8 x bfloat> @fadd_sel_fmul_negzero_nsz_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fadd_sel_fmul_negzero_nsz_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    movi v4.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fadd_sel_fmul_negzero_nsz_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    movi v4.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fadd_sel_fmul_negzero_nsz_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fadd_sel_fmul_negzero_nsz_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    mov z4.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> %nz
@@ -633,32 +1061,54 @@ define <vscale x 8 x bfloat> @fadd_sel_fmul_negzero_nsz_nxv8bf16(<vscale x 8 x b
 }
 
 define <vscale x 8 x bfloat> @fsub_sel_fmul_negzero_nsz_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, <vscale x 8 x i1> %mask) {
-; SVE-LABEL: fsub_sel_fmul_negzero_nsz_nxv8bf16:
-; SVE:       // %bb.0:
-; SVE-NEXT:    movi v3.2d, #0000000000000000
-; SVE-NEXT:    movi v4.2d, #0000000000000000
-; SVE-NEXT:    ptrue p1.s
-; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-NEXT:    movi v2.2d, #0000000000000000
-; SVE-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-NEXT:    uzp1 z1.h, z1.h, z2.h
-; SVE-NEXT:    mov z0.h, p0/m, z1.h
-; SVE-NEXT:    ret
+; SVE-NONSTREAMING-LABEL: fsub_sel_fmul_negzero_nsz_nxv8bf16:
+; SVE-NONSTREAMING:       // %bb.0:
+; SVE-NONSTREAMING-NEXT:    movi v3.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    movi v4.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    ptrue p1.s
+; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    movi v2.2d, #0000000000000000
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-NONSTREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-NONSTREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-NONSTREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-NONSTREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-NONSTREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-NONSTREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-NONSTREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-NONSTREAMING-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fsub_sel_fmul_negzero_nsz_nxv8bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
+;
+; SVE-STREAMING-LABEL: fsub_sel_fmul_negzero_nsz_nxv8bf16:
+; SVE-STREAMING:       // %bb.0:
+; SVE-STREAMING-NEXT:    mov z3.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    mov z4.s, #0 // =0x0
+; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
+; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
+; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
+; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
+; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
+; SVE-STREAMING-NEXT:    ret
   %fmul = fmul <vscale x 8 x bfloat> %b, %c
   %nz = fneg <vscale x 8 x bfloat> zeroinitializer
   %sel = select <vscale x 8 x i1> %mask, <vscale x 8 x bfloat> %fmul, <vscale x 8 x bfloat> %nz
@@ -668,22 +1118,19 @@ define <vscale x 8 x bfloat> @fsub_sel_fmul_negzero_nsz_nxv8bf16(<vscale x 8 x b
 
 define <vscale x 4 x float> @partial_reduce_to_nxv4f32(<vscale x 4 x float> %acc, <vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b) {
 ; SVE-LABEL: partial_reduce_to_nxv4f32:
-; SVE:       // %bb.0: // %entry
+; SVE:       // %bb.0:
 ; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
 ; SVE-NEXT:    bfmlalt z0.s, z1.h, z2.h
 ; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: partial_reduce_to_nxv4f32:
-; SVE-B16B16:       // %bb.0: // %entry
+; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmlalb z0.s, z1.h, z2.h
 ; SVE-B16B16-NEXT:    bfmlalt z0.s, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-entry:
   %a.wide = fpext <vscale x 8 x bfloat> %a to <vscale x 8 x float>
   %b.wide = fpext <vscale x 8 x bfloat> %b to <vscale x 8 x float>
   %mult = fmul <vscale x 8 x float> %a.wide, %b.wide
   %partial.reduce = call <vscale x 4 x float> @llvm.vector.partial.reduce.fadd(<vscale x 4 x float> %acc, <vscale x 8 x float> %mult)
   ret <vscale x 4 x float> %partial.reduce
 }
-
-declare <vscale x 8 x bfloat> @llvm.fma.nxv8bf16(<vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>)

>From f4806bf89de5af74988c1f56cca5e500ea717db9 Mon Sep 17 00:00:00 2001
From: Paul Walker <paul.walker at arm.com>
Date: Wed, 9 Sep 2026 14:35:53 +0000
Subject: [PATCH 2/2] [LLVM][CodeGen][SME] Make FMA decision independent of
 streaming mode.

The relevant bfloat instructions are available in streaming mode and
thus we can benefit from the same BFMLAL based lowering.
---
 .../Target/AArch64/AArch64ISelLowering.cpp    |   3 +-
 .../test/CodeGen/AArch64/sve-bf16-combines.ll | 334 ++++++------------
 2 files changed, 112 insertions(+), 225 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index a3823f8552a3a..ffd90e68f070c 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -20369,8 +20369,7 @@ bool AArch64TargetLowering::isFMAFasterThanFMulAndFAdd(
   case MVT::f64:
     return Subtarget->hasFPARMv8();
   case MVT::bf16:
-    return VT.isScalableVector() && Subtarget->hasBF16() &&
-           Subtarget->isNonStreamingSVEorSME2Available();
+    return VT.isScalableVector() && Subtarget->hasBF16();
   default:
     break;
   }
diff --git a/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll b/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
index 61e56f82a1f6f..eab91bde321b0 100644
--- a/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
+++ b/llvm/test/CodeGen/AArch64/sve-bf16-combines.ll
@@ -33,21 +33,17 @@ define <vscale x 8 x bfloat> @fmla_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale
 ;
 ; SVE-STREAMING-LABEL: fmla_nxv8bf16:
 ; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    uunpkhi z4.s, z2.h
+; SVE-STREAMING-NEXT:    uunpkhi z5.s, z1.h
+; SVE-STREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-STREAMING-NEXT:    uunpklo z1.s, z1.h
 ; SVE-STREAMING-NEXT:    ptrue p0.s
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p0/m, z4.s
-; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    fadd z2.s, z4.s, z3.s
-; SVE-STREAMING-NEXT:    fadd z0.s, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z2.s
+; SVE-STREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    bfmlalb z6.s, z5.h, z4.h
+; SVE-STREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z6.s
 ; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
 ; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
 ; SVE-STREAMING-NEXT:    ret
@@ -57,65 +53,40 @@ define <vscale x 8 x bfloat> @fmla_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmla_nxv4bf16(<vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmla_nxv4bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    ptrue p0.s
-; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmla_nxv4bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NEXT:    ptrue p0.s
+; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.s
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmla_nxv4bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    ptrue p0.s
-; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fadd z0.s, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %res = fadd contract <vscale x 4 x bfloat> %acc, %mul
   ret <vscale x 4 x bfloat> %res
 }
 
 define <vscale x 2 x bfloat> @fmla_nxv2bf16(<vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmla_nxv2bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    ptrue p0.d
-; SVE-NONSTREAMING-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmla_nxv2bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NEXT:    ptrue p0.d
+; SVE-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
+; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.d
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmla_nxv2bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    ptrue p0.d
-; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-STREAMING-NEXT:    fmul z1.s, p0/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fadd z0.s, p0/m, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %res = fadd contract <vscale x 2 x bfloat> %acc, %mul
   ret <vscale x 2 x bfloat> %res
@@ -149,21 +120,19 @@ define <vscale x 8 x bfloat> @fmls_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale
 ;
 ; SVE-STREAMING-LABEL: fmls_nxv8bf16:
 ; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p0.h
+; SVE-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    uunpkhi z5.s, z2.h
+; SVE-STREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-STREAMING-NEXT:    fneg z1.h, p0/m, z1.h
 ; SVE-STREAMING-NEXT:    ptrue p0.s
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p0/m, z4.s
-; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip1 z0.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    fsub z2.s, z4.s, z3.s
-; SVE-STREAMING-NEXT:    fsub z0.s, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z2.s
+; SVE-STREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z0.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    uunpkhi z4.s, z1.h
+; SVE-STREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-STREAMING-NEXT:    bfmlalb z6.s, z4.h, z5.h
+; SVE-STREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z6.s
 ; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
 ; SVE-STREAMING-NEXT:    uzp1 z0.h, z0.h, z1.h
 ; SVE-STREAMING-NEXT:    ret
@@ -173,67 +142,42 @@ define <vscale x 8 x bfloat> @fmls_nxv8bf16(<vscale x 8 x bfloat> %acc, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmls_nxv4bf16(<vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmls_nxv4bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    ptrue p0.s
-; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    fneg z1.h, p0/m, z1.h
-; SVE-NONSTREAMING-NEXT:    bfmlalb z0.s, z1.h, z2.h
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmls_nxv4bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    ptrue p0.s
+; SVE-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NEXT:    fneg z1.h, p0/m, z1.h
+; SVE-NEXT:    bfmlalb z0.s, z1.h, z2.h
+; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.s
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmls_nxv4bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    ptrue p0.s
-; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z3.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fsub z0.s, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %res = fsub contract <vscale x 4 x bfloat> %acc, %mul
   ret <vscale x 4 x bfloat> %res
 }
 
 define <vscale x 2 x bfloat> @fmls_nxv2bf16(<vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmls_nxv2bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    ptrue p0.d
-; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    fneg z1.h, p0/m, z1.h
-; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NONSTREAMING-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmls_nxv2bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    ptrue p0.d
+; SVE-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NEXT:    lsl z0.s, z0.s, #16
+; SVE-NEXT:    fneg z1.h, p0/m, z1.h
+; SVE-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NEXT:    fmla z0.s, p0/m, z1.s, z2.s
+; SVE-NEXT:    bfcvt z0.h, p0/m, z0.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    ptrue p0.d
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmls_nxv2bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    ptrue p0.d
-; SVE-STREAMING-NEXT:    lsl z0.s, z0.s, #16
-; SVE-STREAMING-NEXT:    fmul z1.s, p0/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fsub z0.s, p0/m, z0.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z0.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %res = fsub contract <vscale x 2 x bfloat> %acc, %mul
   ret <vscale x 2 x bfloat> %res
@@ -265,23 +209,19 @@ define <vscale x 8 x bfloat> @fmla_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 ;
 ; SVE-STREAMING-LABEL: fmla_sel_nxv8bf16:
 ; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    uunpkhi z4.s, z2.h
+; SVE-STREAMING-NEXT:    uunpkhi z5.s, z1.h
+; SVE-STREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-STREAMING-NEXT:    uunpklo z1.s, z1.h
 ; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z3.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    bfmlalb z6.s, z5.h, z4.h
 ; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    fadd z3.s, z4.s, z3.s
-; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z6.s
 ; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z2.h, z1.h
 ; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
 ; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
@@ -291,29 +231,17 @@ define <vscale x 8 x bfloat> @fmla_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmla_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmla_sel_nxv4bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z3.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmla_sel_nxv4bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NEXT:    bfcvt z0.h, p0/m, z3.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_sel_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmla_sel_nxv4bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    ptrue p1.s
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fadd z1.s, z2.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %add = fadd contract <vscale x 4 x bfloat> %acc, %mul
   %res = select <vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %add, <vscale x 4 x bfloat> %acc
@@ -321,33 +249,20 @@ define <vscale x 4 x bfloat> @fmla_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale
 }
 
 define <vscale x 2 x bfloat> @fmla_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmla_sel_nxv2bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    ptrue p1.d
-; SVE-NONSTREAMING-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmla_sel_nxv2bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NEXT:    ptrue p1.d
+; SVE-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
+; SVE-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmla_sel_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmla z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmla_sel_nxv2bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    ptrue p1.d
-; SVE-STREAMING-NEXT:    fmul z1.s, p1/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fadd z1.s, p1/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %add = fadd contract <vscale x 2 x bfloat> %acc, %mul
   %res = select <vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %add, <vscale x 2 x bfloat> %acc
@@ -382,23 +297,21 @@ define <vscale x 8 x bfloat> @fmls_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 ;
 ; SVE-STREAMING-LABEL: fmls_sel_nxv8bf16:
 ; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    mov z4.s, #0x80000000
+; SVE-STREAMING-NEXT:    ptrue p1.h
+; SVE-STREAMING-NEXT:    mov z3.h, #0 // =0x0
+; SVE-STREAMING-NEXT:    uunpkhi z5.s, z2.h
+; SVE-STREAMING-NEXT:    uunpklo z2.s, z2.h
+; SVE-STREAMING-NEXT:    fneg z1.h, p1/m, z1.h
 ; SVE-STREAMING-NEXT:    ptrue p1.s
+; SVE-STREAMING-NEXT:    zip2 z6.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    zip1 z3.h, z3.h, z0.h
+; SVE-STREAMING-NEXT:    uunpkhi z4.s, z1.h
+; SVE-STREAMING-NEXT:    uunpklo z1.s, z1.h
+; SVE-STREAMING-NEXT:    bfmlalb z6.s, z4.h, z5.h
 ; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    bfmlalt z4.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    mov z2.h, #0 // =0x0
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvtnt z1.h, p1/m, z4.s
-; SVE-STREAMING-NEXT:    zip2 z4.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    zip2 z3.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z1.h, z2.h, z1.h
-; SVE-STREAMING-NEXT:    zip1 z2.h, z2.h, z0.h
-; SVE-STREAMING-NEXT:    fsub z3.s, z4.s, z3.s
-; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
+; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z6.s
 ; SVE-STREAMING-NEXT:    bfcvt z2.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-STREAMING-NEXT:    uzp1 z1.h, z1.h, z2.h
+; SVE-STREAMING-NEXT:    uzp1 z1.h, z2.h, z1.h
 ; SVE-STREAMING-NEXT:    mov z0.h, p0/m, z1.h
 ; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 8 x bfloat> %m1, %m2
@@ -408,31 +321,19 @@ define <vscale x 8 x bfloat> @fmls_sel_nxv8bf16(<vscale x 8 x i1> %pred, <vscale
 }
 
 define <vscale x 4 x bfloat> @fmls_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %acc, <vscale x 4 x bfloat> %m1, <vscale x 4 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmls_sel_nxv4bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    ptrue p1.s
-; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    fneg z1.h, p1/m, z1.h
-; SVE-NONSTREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z3.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmls_sel_nxv4bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    ptrue p1.s
+; SVE-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NEXT:    fneg z1.h, p1/m, z1.h
+; SVE-NEXT:    bfmlalb z3.s, z1.h, z2.h
+; SVE-NEXT:    bfcvt z0.h, p0/m, z3.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_sel_nxv4bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmls_sel_nxv4bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    mov z3.s, #0x80000000
-; SVE-STREAMING-NEXT:    ptrue p1.s
-; SVE-STREAMING-NEXT:    bfmlalb z3.s, z1.h, z2.h
-; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z3.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fsub z1.s, z2.s, z1.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 4 x bfloat> %m1, %m2
   %sub = fsub contract <vscale x 4 x bfloat> %acc, %mul
   %res = select <vscale x 4 x i1> %pred, <vscale x 4 x bfloat> %sub, <vscale x 4 x bfloat> %acc
@@ -440,34 +341,21 @@ define <vscale x 4 x bfloat> @fmls_sel_nxv4bf16(<vscale x 4 x i1> %pred, <vscale
 }
 
 define <vscale x 2 x bfloat> @fmls_sel_nxv2bf16(<vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %acc, <vscale x 2 x bfloat> %m1, <vscale x 2 x bfloat> %m2) {
-; SVE-NONSTREAMING-LABEL: fmls_sel_nxv2bf16:
-; SVE-NONSTREAMING:       // %bb.0:
-; SVE-NONSTREAMING-NEXT:    ptrue p1.d
-; SVE-NONSTREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-NONSTREAMING-NEXT:    lsl z3.s, z0.s, #16
-; SVE-NONSTREAMING-NEXT:    fneg z1.h, p1/m, z1.h
-; SVE-NONSTREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-NONSTREAMING-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
-; SVE-NONSTREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-NONSTREAMING-NEXT:    ret
+; SVE-LABEL: fmls_sel_nxv2bf16:
+; SVE:       // %bb.0:
+; SVE-NEXT:    ptrue p1.d
+; SVE-NEXT:    lsl z2.s, z2.s, #16
+; SVE-NEXT:    lsl z3.s, z0.s, #16
+; SVE-NEXT:    fneg z1.h, p1/m, z1.h
+; SVE-NEXT:    lsl z1.s, z1.s, #16
+; SVE-NEXT:    fmad z1.s, p1/m, z2.s, z3.s
+; SVE-NEXT:    bfcvt z0.h, p0/m, z1.s
+; SVE-NEXT:    ret
 ;
 ; SVE-B16B16-LABEL: fmls_sel_nxv2bf16:
 ; SVE-B16B16:       // %bb.0:
 ; SVE-B16B16-NEXT:    bfmls z0.h, p0/m, z1.h, z2.h
 ; SVE-B16B16-NEXT:    ret
-;
-; SVE-STREAMING-LABEL: fmls_sel_nxv2bf16:
-; SVE-STREAMING:       // %bb.0:
-; SVE-STREAMING-NEXT:    lsl z2.s, z2.s, #16
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    ptrue p1.d
-; SVE-STREAMING-NEXT:    fmul z1.s, p1/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    lsl z2.s, z0.s, #16
-; SVE-STREAMING-NEXT:    bfcvt z1.h, p1/m, z1.s
-; SVE-STREAMING-NEXT:    lsl z1.s, z1.s, #16
-; SVE-STREAMING-NEXT:    fsubr z1.s, p1/m, z1.s, z2.s
-; SVE-STREAMING-NEXT:    bfcvt z0.h, p0/m, z1.s
-; SVE-STREAMING-NEXT:    ret
   %mul = fmul contract <vscale x 2 x bfloat> %m1, %m2
   %sub = fsub contract <vscale x 2 x bfloat> %acc, %mul
   %res = select <vscale x 2 x i1> %pred, <vscale x 2 x bfloat> %sub, <vscale x 2 x bfloat> %acc



More information about the llvm-commits mailing list