[AMDGPU] Make <2 x bfloat> fabs legal (#142908)

Author: rampitec

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -754,7 +754,7 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::FNEG, {MVT::v2f16, MVT::v2bf16}, Legal);
     // This isn't really legal, but this avoids the legalizer unrolling it (and
     // allows matching fneg (fabs x) patterns)
-    setOperationAction(ISD::FABS, MVT::v2f16, Legal);
+    setOperationAction(ISD::FABS, {MVT::v2f16, MVT::v2bf16}, Legal);
 
     // Can do this in one BFI plus a constant materialize.
     setOperationAction(ISD::FCOPYSIGN,

llvm/lib/Target/AMDGPU/SIInstructions.td

@@ -1835,12 +1835,12 @@ def : GCNPat <
   (UniformUnaryFrag<fneg> (v2fp16vt SReg_32:$src)),
   (S_XOR_B32 SReg_32:$src, (S_MOV_B32 (i32 0x80008000)))
 >;
-}
 
 def : GCNPat <
-  (UniformUnaryFrag<fabs> (v2f16 SReg_32:$src)),
+  (UniformUnaryFrag<fabs> (v2fp16vt SReg_32:$src)),
   (S_AND_B32 SReg_32:$src, (S_MOV_B32 (i32 0x7fff7fff)))
 >;
+}
 
 // This is really (fneg (fabs v2f16:$src))
 //
@@ -1981,12 +1981,12 @@ def : GCNPat <
   (fneg (v2fp16vt VGPR_32:$src)),
   (V_XOR_B32_e64 (S_MOV_B32 (i32 0x80008000)), VGPR_32:$src)
 >;
-}
 
 def : GCNPat <
-  (fabs (v2f16 VGPR_32:$src)),
+  (fabs (v2fp16vt VGPR_32:$src)),
   (V_AND_B32_e64 (S_MOV_B32 (i32 0x7fff7fff)), VGPR_32:$src)
 >;
+}
 
 def : GCNPat <
   (fneg (v2f16 (fabs VGPR_32:$src))),

llvm/test/CodeGen/AMDGPU/fabs.bf16.ll

@@ -176,12 +176,7 @@ define amdgpu_kernel void @s_fabs_v2bf16(ptr addrspace(1) %out, <2 x bfloat> %in
 ; VI-NEXT:    s_mov_b32 flat_scratch_lo, s13
 ; VI-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_and_b32 s3, s2, 0x7fff
-; VI-NEXT:    s_lshr_b32 s2, s2, 16
-; VI-NEXT:    s_and_b32 s2, s2, 0x7fff
-; VI-NEXT:    s_and_b32 s3, 0xffff, s3
-; VI-NEXT:    s_lshl_b32 s2, s2, 16
-; VI-NEXT:    s_or_b32 s2, s3, s2
+; VI-NEXT:    s_and_b32 s2, s2, 0x7fff7fff
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    v_mov_b32_e32 v2, s2
@@ -194,44 +189,22 @@ define amdgpu_kernel void @s_fabs_v2bf16(ptr addrspace(1) %out, <2 x bfloat> %in
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_and_b32 s3, s2, 0x7fff
-; GFX9-NEXT:    s_lshr_b32 s2, s2, 16
-; GFX9-NEXT:    s_and_b32 s2, s2, 0x7fff
-; GFX9-NEXT:    s_pack_ll_b32_b16 s2, s3, s2
+; GFX9-NEXT:    s_and_b32 s2, s2, 0x7fff7fff
 ; GFX9-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
-; GFX11-TRUE16-LABEL: s_fabs_v2bf16:
-; GFX11-TRUE16:       ; %bb.0:
-; GFX11-TRUE16-NEXT:    s_clause 0x1
-; GFX11-TRUE16-NEXT:    s_load_b32 s2, s[4:5], 0x8
-; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
-; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    s_mov_b32 s3, s2
-; GFX11-TRUE16-NEXT:    s_lshr_b32 s2, s2, 16
-; GFX11-TRUE16-NEXT:    s_and_b32 s3, s3, 0x7fff
-; GFX11-TRUE16-NEXT:    s_and_b32 s2, s2, 0x7fff
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-TRUE16-NEXT:    s_pack_ll_b32_b16 s2, s3, s2
-; GFX11-TRUE16-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
-; GFX11-TRUE16-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-TRUE16-NEXT:    s_endpgm
-;
-; GFX11-FAKE16-LABEL: s_fabs_v2bf16:
-; GFX11-FAKE16:       ; %bb.0:
-; GFX11-FAKE16-NEXT:    s_clause 0x1
-; GFX11-FAKE16-NEXT:    s_load_b32 s2, s[4:5], 0x8
-; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
-; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-FAKE16-NEXT:    s_lshr_b32 s3, s2, 16
-; GFX11-FAKE16-NEXT:    s_and_b32 s2, s2, 0x7fff
-; GFX11-FAKE16-NEXT:    s_and_b32 s3, s3, 0x7fff
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
-; GFX11-FAKE16-NEXT:    s_pack_ll_b32_b16 s2, s2, s3
-; GFX11-FAKE16-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
-; GFX11-FAKE16-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-FAKE16-NEXT:    s_endpgm
+; GFX11-LABEL: s_fabs_v2bf16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    s_and_b32 s2, s2, 0x7fff7fff
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %fabs = call <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat> %in)
   store <2 x bfloat> %fabs, ptr addrspace(1) %out
   ret void
@@ -492,59 +465,34 @@ define amdgpu_kernel void @v_fabs_v2bf16(ptr addrspace(1) %out, ptr addrspace(1)
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v2, v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v3, 0x7fff
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_and_b32_e32 v4, 0x7fff, v2
-; VI-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_sdwa v2, v4, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_and_b32_e32 v2, 0x7fff7fff, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: v_fabs_v2bf16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x8
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-NEXT:    s_movk_i32 s2, 0x7fff
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v2, 0x7fff, v1
-; GFX9-NEXT:    v_and_b32_sdwa v1, s2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    s_mov_b32 s2, 0x5040100
-; GFX9-NEXT:    v_perm_b32 v1, v1, v2, s2
+; GFX9-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
 ; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
-; GFX11-TRUE16-LABEL: v_fabs_v2bf16:
-; GFX11-TRUE16:       ; %bb.0:
-; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x8
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    global_load_b32 v1, v0, s[0:1]
-; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_and_b16 v1.l, 0x7fff, v1.l
-; GFX11-TRUE16-NEXT:    v_and_b16 v1.h, 0x7fff, v1.h
-; GFX11-TRUE16-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-TRUE16-NEXT:    s_endpgm
-;
-; GFX11-FAKE16-LABEL: v_fabs_v2bf16:
-; GFX11-FAKE16:       ; %bb.0:
-; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x8
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-FAKE16-NEXT:    global_load_b32 v1, v0, s[0:1]
-; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v1
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v1, 0x7fff, v1
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v2, 0x7fff, v2
-; GFX11-FAKE16-NEXT:    v_perm_b32 v1, v2, v1, 0x5040100
-; GFX11-FAKE16-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-FAKE16-NEXT:    s_endpgm
+; GFX11-LABEL: v_fabs_v2bf16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x8
+; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.in = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i32 %tid
   %gep.out = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i32 %tid
@@ -661,12 +609,12 @@ define amdgpu_kernel void @v_fabs_fold_self_v2bf16(ptr addrspace(1) %out, ptr ad
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_and_b32_sdwa v4, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_and_b32_sdwa v3, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_lshlrev_b32_e32 v5, 16, v2
-; VI-NEXT:    v_and_b32_e32 v2, 0xffff0000, v2
-; VI-NEXT:    v_mul_f32_e32 v3, v3, v5
-; VI-NEXT:    v_mul_f32_e32 v2, v4, v2
+; VI-NEXT:    v_lshlrev_b32_e32 v4, 16, v2
+; VI-NEXT:    v_and_b32_sdwa v5, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_and_b32_e32 v6, 0xffff0000, v2
+; VI-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_mul_f32_e32 v3, v5, v4
+; VI-NEXT:    v_mul_f32_e32 v2, v2, v6
 ; VI-NEXT:    v_bfe_u32 v4, v3, 16, 1
 ; VI-NEXT:    v_bfe_u32 v6, v2, 16, 1
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, v4, v3
@@ -693,20 +641,20 @@ define amdgpu_kernel void @v_fabs_fold_self_v2bf16(ptr addrspace(1) %out, ptr ad
 ; GFX9-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-NEXT:    s_movk_i32 s2, 0x7fff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    v_lshlrev_b32_e32 v2, 16, v0
 ; GFX9-NEXT:    v_and_b32_sdwa v3, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 16, v0
-; GFX9-NEXT:    v_and_b32_sdwa v2, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff0000, v0
-; GFX9-NEXT:    v_mul_f32_e32 v3, v3, v4
-; GFX9-NEXT:    v_mul_f32_e32 v0, v2, v0
-; GFX9-NEXT:    v_bfe_u32 v2, v3, 16, 1
-; GFX9-NEXT:    v_or_b32_e32 v4, 0x400000, v3
+; GFX9-NEXT:    v_and_b32_e32 v4, 0xffff0000, v0
+; GFX9-NEXT:    v_and_b32_sdwa v0, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_mul_f32_e32 v2, v3, v2
+; GFX9-NEXT:    v_mul_f32_e32 v0, v0, v4
+; GFX9-NEXT:    v_bfe_u32 v3, v2, 16, 1
+; GFX9-NEXT:    v_or_b32_e32 v4, 0x400000, v2
 ; GFX9-NEXT:    v_bfe_u32 v5, v0, 16, 1
-; GFX9-NEXT:    v_add3_u32 v2, v2, v3, s2
-; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v3, v3
+; GFX9-NEXT:    v_add3_u32 v3, v3, v2, s2
+; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
 ; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v0
 ; GFX9-NEXT:    v_add3_u32 v5, v5, v0, s2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v4, vcc
 ; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v5, v6, vcc
 ; GFX9-NEXT:    s_mov_b32 s2, 0x7060302
@@ -846,24 +794,24 @@ define amdgpu_kernel void @v_fabs_fold_v2bf16(ptr addrspace(1) %out, ptr addrspa
 ; VI-NEXT:    s_and_b32 s1, s4, 0xffff0000
 ; VI-NEXT:    s_movk_i32 s2, 0x7fff
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_and_b32_sdwa v4, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_mul_f32_e32 v2, s0, v2
-; VI-NEXT:    v_mul_f32_e32 v3, s1, v4
-; VI-NEXT:    v_bfe_u32 v4, v2, 16, 1
-; VI-NEXT:    v_bfe_u32 v6, v3, 16, 1
-; VI-NEXT:    v_add_u32_e32 v4, vcc, v4, v2
-; VI-NEXT:    v_add_u32_e32 v6, vcc, v6, v3
+; VI-NEXT:    v_and_b32_sdwa v4, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_and_b32_sdwa v2, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-NEXT:    v_mul_f32_e32 v3, s0, v4
+; VI-NEXT:    v_mul_f32_e32 v2, s1, v2
+; VI-NEXT:    v_bfe_u32 v4, v3, 16, 1
+; VI-NEXT:    v_bfe_u32 v6, v2, 16, 1
+; VI-NEXT:    v_add_u32_e32 v4, vcc, v4, v3
+; VI-NEXT:    v_add_u32_e32 v6, vcc, v6, v2
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, s2, v4
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, 0x7fff, v6
-; VI-NEXT:    v_or_b32_e32 v5, 0x400000, v2
-; VI-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
-; VI-NEXT:    v_or_b32_e32 v7, 0x400000, v3
-; VI-NEXT:    v_cndmask_b32_e32 v2, v4, v5, vcc
+; VI-NEXT:    v_or_b32_e32 v5, 0x400000, v3
 ; VI-NEXT:    v_cmp_u_f32_e32 vcc, v3, v3
-; VI-NEXT:    v_cndmask_b32_e32 v3, v6, v7, vcc
-; VI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
-; VI-NEXT:    v_alignbit_b32 v2, v3, v2, 16
+; VI-NEXT:    v_or_b32_e32 v7, 0x400000, v2
+; VI-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
+; VI-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, v6, v7, vcc
+; VI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
+; VI-NEXT:    v_alignbit_b32 v2, v2, v3, 16
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
@@ -879,22 +827,22 @@ define amdgpu_kernel void @v_fabs_fold_v2bf16(ptr addrspace(1) %out, ptr addrspa
 ; GFX9-NEXT:    s_lshl_b32 s3, s4, 16
 ; GFX9-NEXT:    s_and_b32 s4, s4, 0xffff0000
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v2, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_and_b32_sdwa v0, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT:    v_mul_f32_e32 v0, s3, v0
-; GFX9-NEXT:    v_mul_f32_e32 v2, s4, v2
-; GFX9-NEXT:    v_bfe_u32 v3, v0, 16, 1
-; GFX9-NEXT:    v_or_b32_e32 v4, 0x400000, v0
-; GFX9-NEXT:    v_bfe_u32 v5, v2, 16, 1
-; GFX9-NEXT:    v_add3_u32 v3, v3, v0, s2
-; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
-; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v2
-; GFX9-NEXT:    v_add3_u32 v5, v5, v2, s2
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v4, vcc
+; GFX9-NEXT:    v_and_b32_sdwa v2, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v0, s2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX9-NEXT:    v_mul_f32_e32 v2, s3, v2
+; GFX9-NEXT:    v_mul_f32_e32 v0, s4, v0
+; GFX9-NEXT:    v_bfe_u32 v3, v2, 16, 1
+; GFX9-NEXT:    v_or_b32_e32 v4, 0x400000, v2
+; GFX9-NEXT:    v_bfe_u32 v5, v0, 16, 1
+; GFX9-NEXT:    v_add3_u32 v3, v3, v2, s2
 ; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v5, v6, vcc
+; GFX9-NEXT:    v_or_b32_e32 v6, 0x400000, v0
+; GFX9-NEXT:    v_add3_u32 v5, v5, v0, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v4, vcc
+; GFX9-NEXT:    v_cmp_u_f32_e32 vcc, v0, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v5, v6, vcc
 ; GFX9-NEXT:    s_mov_b32 s2, 0x7060302
-; GFX9-NEXT:    v_perm_b32 v0, v2, v0, s2
+; GFX9-NEXT:    v_perm_b32 v0, v0, v2, s2
 ; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
 ;
@@ -1194,10 +1142,9 @@ define amdgpu_kernel void @v_extract_fabs_no_fold_v2bf16(ptr addrspace(1) %in) #
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v1, 0x7fff
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_and_b32_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; VI-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
+; VI-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
+; VI-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; VI-NEXT:    flat_store_short v[0:1], v0
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    flat_store_short v[0:1], v1
@@ -1210,51 +1157,29 @@ define amdgpu_kernel void @v_extract_fabs_no_fold_v2bf16(ptr addrspace(1) %in) #
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_load_dword v0, v0, s[0:1]
-; GFX9-NEXT:    s_movk_i32 s0, 0x7fff
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_and_b32_sdwa v1, s0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
+; GFX9-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
 ; GFX9-NEXT:    global_store_short v[0:1], v0, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    global_store_short v[0:1], v1, off
+; GFX9-NEXT:    global_store_short_d16_hi v[0:1], v0, off
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-NEXT:    s_endpgm
 ;
-; GFX11-TRUE16-LABEL: v_extract_fabs_no_fold_v2bf16:
-; GFX11-TRUE16:       ; %bb.0:
-; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-TRUE16-NEXT:    global_load_b32 v0, v0, s[0:1]
-; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_and_b16 v0.l, 0x7fff, v0.l
-; GFX11-TRUE16-NEXT:    v_and_b16 v0.h, 0x7fff, v0.h
-; GFX11-TRUE16-NEXT:    global_store_b16 v[0:1], v0, off dlc
-; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-TRUE16-NEXT:    global_store_d16_hi_b16 v[0:1], v0, off dlc
-; GFX11-TRUE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-TRUE16-NEXT:    s_endpgm
-;
-; GFX11-FAKE16-LABEL: v_extract_fabs_no_fold_v2bf16:
-; GFX11-FAKE16:       ; %bb.0:
-; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-FAKE16-NEXT:    global_load_b32 v0, v0, s[0:1]
-; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
-; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-FAKE16-NEXT:    v_and_b32_e32 v1, 0x7fff, v1
-; GFX11-FAKE16-NEXT:    global_store_b16 v[0:1], v0, off dlc
-; GFX11-FAKE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-FAKE16-NEXT:    global_store_b16 v[0:1], v1, off dlc
-; GFX11-FAKE16-NEXT:    s_waitcnt_vscnt null, 0x0
-; GFX11-FAKE16-NEXT:    s_endpgm
+; GFX11-LABEL: v_extract_fabs_no_fold_v2bf16:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_load_b32 v0, v0, s[0:1]
+; GFX11-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
+; GFX11-NEXT:    global_store_b16 v[0:1], v0, off dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_store_d16_hi_b16 v[0:1], v0, off dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.in = getelementptr inbounds <2 x bfloat>, ptr addrspace(1) %in, i32 %tid
   %val = load <2 x bfloat>, ptr addrspace(1) %gep.in