[llvm] [X86][BF16] Add X86SubVBroadcastld patterns (PR #76479)

[Phoebe Wang via llvm-commits]

https://github.com/phoebewang created https://github.com/llvm/llvm-project/pull/76479

None

>From a15f594165341ab7550d950bb2f9eb366695ea2d Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang at intel.com>
Date: Wed, 27 Dec 2023 21:47:51 +0800
Subject: [PATCH] [X86][BF16] Add X86SubVBroadcastld patterns

---
 llvm/lib/Target/X86/X86InstrAVX512.td | 11 +++++++++++
 llvm/lib/Target/X86/X86InstrSSE.td    |  4 ++++
 llvm/test/CodeGen/X86/bfloat.ll       | 24 ++++++++++++++++++++++++
 3 files changed, 39 insertions(+)

diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td
index 7c3c1d5fe42b3c..c3a673f97d3474 100644
--- a/llvm/lib/Target/X86/X86InstrAVX512.td
+++ b/llvm/lib/Target/X86/X86InstrAVX512.td
@@ -1447,6 +1447,17 @@ def : Pat<(vselect_mask VK8WM:$mask,
           (VBROADCASTI32X4Z256rmk VR256X:$src0, VK8WM:$mask, addr:$src)>;
 }
 
+let Predicates = [HasBF16] in {
+  def : Pat<(v32bf16 (X86SubVBroadcastld256 addr:$src)),
+            (VBROADCASTF64X4rm addr:$src)>;
+  def : Pat<(v32bf16 (X86SubVBroadcastld128 addr:$src)),
+            (VBROADCASTF32X4rm addr:$src)>;
+}
+
+let Predicates = [HasBF16, HasVLX] in
+  def : Pat<(v16bf16 (X86SubVBroadcastld128 addr:$src)),
+            (VBROADCASTF32X4Z256rm addr:$src)>;
+
 let Predicates = [HasVLX, HasDQI] in {
 defm VBROADCASTI64X2Z128 : avx512_subvec_broadcast_rm_dq<0x5a, "vbroadcasti64x2",
                            X86SubVBroadcastld128, v4i64x_info, v2i64x_info>, VEX_W1X,
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index df1f0b5b4ca727..0e4e6937bf44cd 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -7160,6 +7160,10 @@ def : Pat<(v32i8 (X86SubVBroadcastld128 addr:$src)),
           (VBROADCASTF128rm addr:$src)>;
 }
 
+let Predicates = [HasAVXNECONVERT, NoVLX] in
+  def : Pat<(v16bf16 (X86SubVBroadcastld128 addr:$src)),
+            (VBROADCASTF128rm addr:$src)>;
+
 //===----------------------------------------------------------------------===//
 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
 //
diff --git a/llvm/test/CodeGen/X86/bfloat.ll b/llvm/test/CodeGen/X86/bfloat.ll
index 4792e8343d7589..72f3eacf87594c 100644
--- a/llvm/test/CodeGen/X86/bfloat.ll
+++ b/llvm/test/CodeGen/X86/bfloat.ll
@@ -2461,3 +2461,27 @@ define <8 x bfloat> @fptrunc_v8f64(<8 x double> %a) nounwind {
   %b = fptrunc <8 x double> %a to <8 x bfloat>
   ret <8 x bfloat> %b
 }
+
+define <32 x bfloat> @test_v8bf16_v32bf16(ptr %0) {
+; SSE2-LABEL: test_v8bf16_v32bf16:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movaps (%rdi), %xmm0
+; SSE2-NEXT:    movaps %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm0, %xmm2
+; SSE2-NEXT:    movaps %xmm0, %xmm3
+; SSE2-NEXT:    retq
+;
+; F16-LABEL: test_v8bf16_v32bf16:
+; F16:       # %bb.0:
+; F16-NEXT:    vbroadcastf32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
+; F16-NEXT:    retq
+;
+; AVXNC-LABEL: test_v8bf16_v32bf16:
+; AVXNC:       # %bb.0:
+; AVXNC-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
+; AVXNC-NEXT:    vmovaps %ymm0, %ymm1
+; AVXNC-NEXT:    retq
+  %2 = load <8 x bfloat>, ptr %0, align 16
+  %3 = shufflevector <8 x bfloat> %2, <8 x bfloat> %2, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  ret <32 x bfloat> %3
+}