[llvm] [LLVM] Combine v_cvt_f32_f16 and v_add_f32/v_mul_f32 into v_fma_mix_f32 (PR #160151)
Acim Maravic via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 22 09:50:47 PDT 2025
https://github.com/Acim-Maravic created https://github.com/llvm/llvm-project/pull/160151
None
>From d8873c0d7156731d83f341875228a1aeae641f41 Mon Sep 17 00:00:00 2001
From: Acim Maravic <Acim.Maravic at amd.com>
Date: Mon, 22 Sep 2025 18:11:48 +0200
Subject: [PATCH] [LLVM] Combine v_cvt_f32_f16 and v_add_f32/v_mul_f32 into
v_fma_mix_f32
---
llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp | 18 +
llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h | 6 +
llvm/lib/Target/AMDGPU/SIInstrInfo.td | 2 +
llvm/lib/Target/AMDGPU/VOP3PInstructions.td | 25 +
.../GlobalISel/combine-fma-add-ext-fma.ll | 130 +-
.../GlobalISel/combine-fma-add-ext-mul.ll | 53 +-
.../CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll | 1208 ++++---
.../AMDGPU/agpr-copy-no-free-registers.ll | 32 +-
llvm/test/CodeGen/AMDGPU/fdiv.f16.ll | 37 +-
llvm/test/CodeGen/AMDGPU/fpext-free.ll | 168 +-
llvm/test/CodeGen/AMDGPU/frem.ll | 346 +-
.../CodeGen/AMDGPU/global-atomicrmw-fadd.ll | 326 +-
llvm/test/CodeGen/AMDGPU/mad-mix-bf16.ll | 650 +++-
llvm/test/CodeGen/AMDGPU/mad-mix.ll | 2995 ++++++++++++++++-
14 files changed, 4671 insertions(+), 1325 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index d4210b8bc9a87..9654a6e1fbd5f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -4127,6 +4127,24 @@ bool AMDGPUDAGToDAGISel::SelectVOP3PMadMixBF16Mods(SDValue In, SDValue &Src,
return true;
}
+bool AMDGPUDAGToDAGISel::SelectVOP3PMadMixModsNeg(SDValue In, SDValue &Src,
+ SDValue &SrcMods) const {
+ unsigned Mods = 0;
+ SelectVOP3PMadMixModsImpl(In, Src, Mods, MVT::f16);
+ Mods ^= SISrcMods::NEG;
+ SrcMods = CurDAG->getTargetConstant(Mods, SDLoc(In), MVT::i32);
+ return true;
+}
+
+bool AMDGPUDAGToDAGISel::SelectVOP3PMadMixBF16ModsNeg(SDValue In, SDValue &Src,
+ SDValue &SrcMods) const {
+ unsigned Mods = 0;
+ SelectVOP3PMadMixModsImpl(In, Src, Mods, MVT::bf16);
+ Mods ^= SISrcMods::NEG;
+ SrcMods = CurDAG->getTargetConstant(Mods, SDLoc(In), MVT::i32);
+ return true;
+}
+
// Match BITOP3 operation and return a number of matched instructions plus
// truth table.
static std::pair<unsigned, uint8_t> BitOp3_Op(SDValue In,
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
index 4fa0d3f72e1c7..b122b5cd310b3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h
@@ -272,6 +272,12 @@ class AMDGPUDAGToDAGISel : public SelectionDAGISel {
bool SelectVOP3PMadMixBF16Mods(SDValue In, SDValue &Src,
SDValue &SrcMods) const;
+ bool SelectVOP3PMadMixModsNeg(SDValue In, SDValue &Src,
+ SDValue &SrcMods) const;
+
+ bool SelectVOP3PMadMixBF16ModsNeg(SDValue In, SDValue &Src,
+ SDValue &SrcMods) const;
+
bool SelectBITOP3(SDValue In, SDValue &Src0, SDValue &Src1, SDValue &Src2,
SDValue &Tbl) const;
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index fb2cd04b364d7..af6d4ff319fd9 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -1704,6 +1704,8 @@ def VOP3PMadMixModsExt : ComplexPattern<untyped, 2, "SelectVOP3PMadMixModsExt">;
def VOP3PMadMixMods : ComplexPattern<untyped, 2, "SelectVOP3PMadMixMods">;
def VOP3PMadMixBF16ModsExt : ComplexPattern<untyped, 2, "SelectVOP3PMadMixBF16ModsExt">;
def VOP3PMadMixBF16Mods : ComplexPattern<untyped, 2, "SelectVOP3PMadMixBF16Mods">;
+def VOP3PMadMixModsNeg : ComplexPattern<untyped, 2, "SelectVOP3PMadMixModsNeg">;
+def VOP3PMadMixBF16ModsNeg : ComplexPattern<untyped, 2, "SelectVOP3PMadMixBF16ModsNeg">;
def VINTERPMods : ComplexPattern<untyped, 2, "SelectVINTERPMods">;
def VINTERPModsHi : ComplexPattern<untyped, 2, "SelectVINTERPModsHi">;
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index f7279b664ed27..5ea8dbe9a1b7f 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -168,6 +168,7 @@ multiclass MadFmaMixPats<SDPatternOperator fma_like,
ValueType vecVT = v2f16> {
defvar VOP3PMadMixModsPat = !if (!eq(VT, bf16), VOP3PMadMixBF16Mods, VOP3PMadMixMods);
defvar VOP3PMadMixModsExtPat = !if (!eq(VT, bf16), VOP3PMadMixBF16ModsExt, VOP3PMadMixModsExt);
+ defvar VOP3PMadMixModsNegPat = !if (!eq(VT, bf16), VOP3PMadMixBF16ModsNeg, VOP3PMadMixModsNeg);
// At least one of the operands needs to be an fpextend of an f16
// for this to be worthwhile, so we need three patterns here.
// TODO: Could we use a predicate to inspect src1/2/3 instead?
@@ -190,6 +191,30 @@ multiclass MadFmaMixPats<SDPatternOperator fma_like,
(mix_inst $src0_mods, $src0, $src1_mods, $src1, $src2_mods, $src2,
DSTCLAMP.NONE)>;
+ def : GCNPat <
+ (f32 (fadd (f32 (VOP3PMadMixModsExtPat VT:$src0, i32:$src0_mods)),
+ (f32 (VOP3PMadMixModsPat f32:$src1, i32:$src1_mods)))),
+ (mix_inst $src0_mods, $src0, (i32 0), (i32 1), $src1_mods, $src1,
+ DSTCLAMP.NONE)>;
+
+ def : GCNPat <
+ (f32 (fmul (f32 (VOP3PMadMixModsExtPat VT:$src0, i32:$src0_mods)),
+ (f32 (VOP3PMadMixModsPat f32:$src1, i32:$src1_mods)))),
+ (mix_inst $src0_mods, $src0, $src1_mods, $src1, (i32 0), (i32 0),
+ DSTCLAMP.NONE)>;
+
+ def : GCNPat <
+ (f32 (fsub (f32 (VOP3PMadMixModsExtPat VT:$src0, i32:$src0_mods)),
+ (f32 (VOP3PMadMixModsNegPat f32:$src1, i32:$src1_mods)))),
+ (mix_inst $src0_mods, $src0, (i32 0), (i32 1), $src1_mods, $src1,
+ DSTCLAMP.NONE)>;
+
+ def : GCNPat <
+ (f32 (fsub (f32 (VOP3PMadMixModsNegPat f32:$src0, i32:$src0_mods)),
+ (f32 (VOP3PMadMixModsExtPat VT:$src1, i32:$src1_mods)))),
+ (mix_inst $src0_mods, $src0, (i32 0), (i32 1), $src1_mods, $src1,
+ DSTCLAMP.NONE)>;
+
def : GCNPat <
(AMDGPUclamp (build_vector
(VT (fpround (fma_like (f32 (VOP3PMadMixModsPat VT:$lo_src0, i32:$lo_src0_modifiers)),
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-fma.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-fma.ll
index b2b433167fe4d..f9b63ef8e96e5 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-fma.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-fma.ll
@@ -53,16 +53,14 @@ define amdgpu_vs float @test_f16_f32_add_ext_fma_mul(half %x, half %y, float %z,
; GFX10: ; %bb.0: ; %.entry
; GFX10-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-NEXT: v_fmac_f16_e32 v3, v0, v1
-; GFX10-NEXT: v_cvt_f32_f16_e32 v0, v3
-; GFX10-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-NEXT: v_fma_mix_f32 v0, v3, 1, v2 op_sel_hi:[1,0,0]
; GFX10-NEXT: ; return to shader part epilog
;
; GFX10-CONTRACT-LABEL: test_f16_f32_add_ext_fma_mul:
; GFX10-CONTRACT: ; %bb.0: ; %.entry
; GFX10-CONTRACT-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-CONTRACT-NEXT: v_fmac_f16_e32 v3, v0, v1
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v0, v3
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v0, v3, 1, v2 op_sel_hi:[1,0,0]
; GFX10-CONTRACT-NEXT: ; return to shader part epilog
;
; GFX10-DENORM-LABEL: test_f16_f32_add_ext_fma_mul:
@@ -70,8 +68,7 @@ define amdgpu_vs float @test_f16_f32_add_ext_fma_mul(half %x, half %y, float %z,
; GFX10-DENORM-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-DENORM-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX10-DENORM-NEXT: v_add_f16_e32 v0, v0, v3
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX10-DENORM-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v0, v0, 1, v2 op_sel_hi:[1,0,0]
; GFX10-DENORM-NEXT: ; return to shader part epilog
.entry:
%a = fmul half %u, %v
@@ -129,16 +126,14 @@ define amdgpu_vs float @test_f16_f32_add_ext_fma_mul_rhs(float %x, half %y, half
; GFX10: ; %bb.0: ; %.entry
; GFX10-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX10-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX10-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX10-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX10-NEXT: ; return to shader part epilog
;
; GFX10-CONTRACT-LABEL: test_f16_f32_add_ext_fma_mul_rhs:
; GFX10-CONTRACT: ; %bb.0: ; %.entry
; GFX10-CONTRACT-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-CONTRACT-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX10-CONTRACT-NEXT: ; return to shader part epilog
;
; GFX10-DENORM-LABEL: test_f16_f32_add_ext_fma_mul_rhs:
@@ -146,8 +141,7 @@ define amdgpu_vs float @test_f16_f32_add_ext_fma_mul_rhs(float %x, half %y, half
; GFX10-DENORM-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX10-DENORM-NEXT: v_mul_f16_e32 v1, v1, v2
; GFX10-DENORM-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GFX10-DENORM-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v0, v1, 1, v0 op_sel_hi:[1,0,0]
; GFX10-DENORM-NEXT: ; return to shader part epilog
.entry:
%a = fmul half %u, %v
@@ -230,48 +224,36 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul(<4 x half> %x, <4
; GFX9-DENORM-NEXT: v_pk_mul_f16 v9, v9, v11
; GFX9-DENORM-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX9-DENORM-NEXT: v_pk_mul_f16 v1, v1, v3
-; GFX9-DENORM-NEXT: v_pk_add_f16 v0, v0, v8
-; GFX9-DENORM-NEXT: v_pk_add_f16 v1, v1, v9
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_e32 v8, v1
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_sdwa v9, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-DENORM-NEXT: v_add_f32_e32 v0, v2, v4
-; GFX9-DENORM-NEXT: v_add_f32_e32 v1, v3, v5
-; GFX9-DENORM-NEXT: v_add_f32_e32 v2, v8, v6
-; GFX9-DENORM-NEXT: v_add_f32_e32 v3, v9, v7
+; GFX9-DENORM-NEXT: v_pk_add_f16 v2, v0, v8
+; GFX9-DENORM-NEXT: v_pk_add_f16 v3, v1, v9
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v0, v2, 1, v4 op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v1, v2, 1, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v2, v3, 1, v6 op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v3, v3, 1, v7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX9-DENORM-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: test_v4f16_v4f32_add_ext_fma_mul:
; GFX10: ; %bb.0: ; %.entry
; GFX10-NEXT: v_pk_mul_f16 v8, v8, v10
; GFX10-NEXT: v_pk_mul_f16 v9, v9, v11
-; GFX10-NEXT: v_pk_fma_f16 v0, v0, v2, v8
-; GFX10-NEXT: v_pk_fma_f16 v1, v1, v3, v9
-; GFX10-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-NEXT: v_cvt_f32_f16_e32 v8, v1
-; GFX10-NEXT: v_cvt_f32_f16_sdwa v9, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-NEXT: v_add_f32_e32 v0, v2, v4
-; GFX10-NEXT: v_add_f32_e32 v1, v3, v5
-; GFX10-NEXT: v_add_f32_e32 v2, v8, v6
-; GFX10-NEXT: v_add_f32_e32 v3, v9, v7
+; GFX10-NEXT: v_pk_fma_f16 v2, v0, v2, v8
+; GFX10-NEXT: v_pk_fma_f16 v3, v1, v3, v9
+; GFX10-NEXT: v_fma_mix_f32 v0, v2, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v1, v2, 1, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v2, v3, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v3, v3, 1, v7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-NEXT: ; return to shader part epilog
;
; GFX10-CONTRACT-LABEL: test_v4f16_v4f32_add_ext_fma_mul:
; GFX10-CONTRACT: ; %bb.0: ; %.entry
; GFX10-CONTRACT-NEXT: v_pk_mul_f16 v8, v8, v10
; GFX10-CONTRACT-NEXT: v_pk_mul_f16 v9, v9, v11
-; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v0, v0, v2, v8
-; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v1, v1, v3, v9
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v8, v1
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_sdwa v9, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v0, v2, v4
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v1, v3, v5
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v2, v8, v6
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v3, v9, v7
+; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v2, v0, v2, v8
+; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v3, v1, v3, v9
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v0, v2, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v1, v2, 1, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v2, v3, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v3, v3, 1, v7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-CONTRACT-NEXT: ; return to shader part epilog
;
; GFX10-DENORM-LABEL: test_v4f16_v4f32_add_ext_fma_mul:
@@ -280,16 +262,12 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul(<4 x half> %x, <4
; GFX10-DENORM-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX10-DENORM-NEXT: v_pk_mul_f16 v2, v9, v11
; GFX10-DENORM-NEXT: v_pk_mul_f16 v1, v1, v3
-; GFX10-DENORM-NEXT: v_pk_add_f16 v0, v0, v8
-; GFX10-DENORM-NEXT: v_pk_add_f16 v1, v1, v2
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v8, v1
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_sdwa v9, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-DENORM-NEXT: v_add_f32_e32 v0, v2, v4
-; GFX10-DENORM-NEXT: v_add_f32_e32 v1, v3, v5
-; GFX10-DENORM-NEXT: v_add_f32_e32 v2, v8, v6
-; GFX10-DENORM-NEXT: v_add_f32_e32 v3, v9, v7
+; GFX10-DENORM-NEXT: v_pk_add_f16 v3, v0, v8
+; GFX10-DENORM-NEXT: v_pk_add_f16 v8, v1, v2
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v0, v3, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v1, v3, 1, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v2, v8, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v3, v8, 1, v7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-DENORM-NEXT: ; return to shader part epilog
.entry:
%a = fmul <4 x half> %u, %v
@@ -374,14 +352,10 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul_rhs(<4 x float> %
; GFX9-DENORM-NEXT: v_pk_mul_f16 v5, v5, v7
; GFX9-DENORM-NEXT: v_pk_add_f16 v4, v4, v8
; GFX9-DENORM-NEXT: v_pk_add_f16 v5, v5, v9
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_e32 v6, v4
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_sdwa v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_e32 v7, v5
-; GFX9-DENORM-NEXT: v_cvt_f32_f16_sdwa v5, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-DENORM-NEXT: v_add_f32_e32 v0, v0, v6
-; GFX9-DENORM-NEXT: v_add_f32_e32 v1, v1, v4
-; GFX9-DENORM-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX9-DENORM-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v0, v4, 1, v0 op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v1, v4, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v2, v5, 1, v2 op_sel_hi:[1,0,0]
+; GFX9-DENORM-NEXT: v_mad_mix_f32 v3, v5, 1, v3 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX9-DENORM-NEXT: ; return to shader part epilog
;
; GFX10-LABEL: test_v4f16_v4f32_add_ext_fma_mul_rhs:
@@ -390,14 +364,10 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul_rhs(<4 x float> %
; GFX10-NEXT: v_pk_mul_f16 v9, v9, v11
; GFX10-NEXT: v_pk_fma_f16 v4, v4, v6, v8
; GFX10-NEXT: v_pk_fma_f16 v5, v5, v7, v9
-; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v4
-; GFX10-NEXT: v_cvt_f32_f16_sdwa v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-NEXT: v_cvt_f32_f16_e32 v7, v5
-; GFX10-NEXT: v_cvt_f32_f16_sdwa v5, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-NEXT: v_add_f32_e32 v0, v0, v6
-; GFX10-NEXT: v_add_f32_e32 v1, v1, v4
-; GFX10-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-NEXT: v_fma_mix_f32 v0, v4, 1, v0 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v1, v4, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v2, v5, 1, v2 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_fma_mix_f32 v3, v5, 1, v3 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-NEXT: ; return to shader part epilog
;
; GFX10-CONTRACT-LABEL: test_v4f16_v4f32_add_ext_fma_mul_rhs:
@@ -406,14 +376,10 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul_rhs(<4 x float> %
; GFX10-CONTRACT-NEXT: v_pk_mul_f16 v9, v9, v11
; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v4, v4, v6, v8
; GFX10-CONTRACT-NEXT: v_pk_fma_f16 v5, v5, v7, v9
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v6, v4
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_sdwa v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_e32 v7, v5
-; GFX10-CONTRACT-NEXT: v_cvt_f32_f16_sdwa v5, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v0, v0, v6
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v1, v1, v4
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-CONTRACT-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v0, v4, 1, v0 op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v1, v4, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v2, v5, 1, v2 op_sel_hi:[1,0,0]
+; GFX10-CONTRACT-NEXT: v_fma_mix_f32 v3, v5, 1, v3 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-CONTRACT-NEXT: ; return to shader part epilog
;
; GFX10-DENORM-LABEL: test_v4f16_v4f32_add_ext_fma_mul_rhs:
@@ -424,14 +390,10 @@ define amdgpu_vs <4 x float> @test_v4f16_v4f32_add_ext_fma_mul_rhs(<4 x float> %
; GFX10-DENORM-NEXT: v_pk_mul_f16 v5, v5, v7
; GFX10-DENORM-NEXT: v_pk_add_f16 v4, v4, v8
; GFX10-DENORM-NEXT: v_pk_add_f16 v5, v5, v6
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v6, v4
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_sdwa v4, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_e32 v7, v5
-; GFX10-DENORM-NEXT: v_cvt_f32_f16_sdwa v5, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-DENORM-NEXT: v_add_f32_e32 v0, v0, v6
-; GFX10-DENORM-NEXT: v_add_f32_e32 v1, v1, v4
-; GFX10-DENORM-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-DENORM-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v0, v4, 1, v0 op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v1, v4, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v2, v5, 1, v2 op_sel_hi:[1,0,0]
+; GFX10-DENORM-NEXT: v_fma_mix_f32 v3, v5, 1, v3 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX10-DENORM-NEXT: ; return to shader part epilog
.entry:
%a = fmul <4 x half> %u, %v
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-mul.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-mul.ll
index 4d603f7487754..26f8e41c9351a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-mul.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/combine-fma-add-ext-mul.ll
@@ -49,21 +49,16 @@ define amdgpu_vs <5 x float> @test_5xf16_5xf32_add_ext_mul(<5 x half> inreg %x,
; GFX9-FAST-DENORM-LABEL: test_5xf16_5xf32_add_ext_mul:
; GFX9-FAST-DENORM: ; %bb.0: ; %.entry
; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s3
-; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v1, s4
-; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v2, s5
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v0, s0, v0
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v1, s1, v1
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v2, s2, v2
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v3, v0
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_sdwa v4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v5, v1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_sdwa v6, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v7, v2
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v0, s6, v3
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v1, s7, v4
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v2, s8, v5
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v3, s9, v6
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v4, s10, v7
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v1, s0, v0
+; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v3, s1, v0
+; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s5
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v4, s2, v0
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v0, v1, 1, s6 op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v1, v1, 1, s7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v2, v3, 1, s8 op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v3, v3, 1, s9 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v4, v4, 1, s10 op_sel_hi:[1,0,0]
; GFX9-FAST-DENORM-NEXT: ; return to shader part epilog
;
; GFX10-FAST-DENORM-LABEL: test_5xf16_5xf32_add_ext_mul:
@@ -94,23 +89,17 @@ define amdgpu_vs <6 x float> @test_6xf16_6xf32_add_ext_mul_rhs(<6 x half> inreg
; GFX9-FAST-DENORM-LABEL: test_6xf16_6xf32_add_ext_mul_rhs:
; GFX9-FAST-DENORM: ; %bb.0: ; %.entry
; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s3
-; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v1, s4
-; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v2, s5
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v0, s0, v0
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v1, s1, v1
-; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v2, s2, v2
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v3, v0
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_sdwa v4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v5, v1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_sdwa v6, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_e32 v7, v2
-; GFX9-FAST-DENORM-NEXT: v_cvt_f32_f16_sdwa v8, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v0, s6, v3
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v1, s7, v4
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v2, s8, v5
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v3, s9, v6
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v4, s10, v7
-; GFX9-FAST-DENORM-NEXT: v_add_f32_e32 v5, s11, v8
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v1, s0, v0
+; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v3, s1, v0
+; GFX9-FAST-DENORM-NEXT: v_mov_b32_e32 v0, s5
+; GFX9-FAST-DENORM-NEXT: v_pk_mul_f16 v5, s2, v0
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v0, v1, 1, s6 op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v1, v1, 1, s7 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v2, v3, 1, s8 op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v3, v3, 1, s9 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v4, v5, 1, s10 op_sel_hi:[1,0,0]
+; GFX9-FAST-DENORM-NEXT: v_mad_mix_f32 v5, v5, 1, s11 op_sel:[1,0,0] op_sel_hi:[1,0,0]
; GFX9-FAST-DENORM-NEXT: ; return to shader part epilog
;
; GFX10-FAST-DENORM-LABEL: test_6xf16_6xf32_add_ext_mul_rhs:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll
index 1aee6ab24eea0..ed80469052d17 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll
@@ -118,9 +118,8 @@ define half @v_fdiv_f16(half %a, half %b) {
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mac_f32_e32 v3, v4, v2
; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
@@ -135,18 +134,17 @@ define half @v_fdiv_f16(half %a, half %b) {
; GFX10-IEEE: ; %bb.0:
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v0
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, v0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v2
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, v0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v4, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v3
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v1, v0
; GFX10-IEEE-NEXT: s_setpc_b64 s[30:31]
@@ -157,7 +155,7 @@ define half @v_fdiv_f16(half %a, half %b) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v3, v2
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v5, v4
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v3
; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v5, v4
@@ -172,10 +170,9 @@ define half @v_fdiv_f16(half %a, half %b) {
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v3, v3, v2
+; GFX11-NEXT: v_fma_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_fmac_f32_e32 v3, v4, v2
; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
@@ -328,9 +325,8 @@ define half @v_fdiv_f16_ulp25(half %a, half %b) {
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mac_f32_e32 v3, v4, v2
; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
@@ -345,18 +341,17 @@ define half @v_fdiv_f16_ulp25(half %a, half %b) {
; GFX10-IEEE: ; %bb.0:
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v0
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, v0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v2
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, v0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v4, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v3
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v1, v0
; GFX10-IEEE-NEXT: s_setpc_b64 s[30:31]
@@ -367,7 +362,7 @@ define half @v_fdiv_f16_ulp25(half %a, half %b) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v3, v2
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v5, v4
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v3
; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v5, v4
@@ -382,10 +377,9 @@ define half @v_fdiv_f16_ulp25(half %a, half %b) {
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v3, v3, v2
+; GFX11-NEXT: v_fma_mix_f32 v3, v0, v2, 0 op_sel_hi:[1,0,0]
; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_fmac_f32_e32 v3, v4, v2
; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v3, v0 op_sel_hi:[1,0,1]
@@ -942,31 +936,29 @@ define <2 x half> @v_fdiv_v2f16(<2 x half> %a, <2 x half> %b) {
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v0
-; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v7, v5
+; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, v3
-; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v7, v7
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v4, v0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v8, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v4, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, v0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v4
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v6, v7
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v7
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v6, v7
-; GFX9-FLUSH-NEXT: v_and_b32_e32 v6, 0xff800000, v6
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v4, v6, v4
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v7, v6, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX9-FLUSH-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v4, v4, v7
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v4, v4
+; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v1, v0
-; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v5, v3
+; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v3, v5
; GFX9-FLUSH-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-FLUSH-NEXT: s_setpc_b64 s[30:31]
;
@@ -975,32 +967,30 @@ define <2 x half> @v_fdiv_v2f16(<2 x half> %a, <2 x half> %b) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v5, 16, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v8, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v6, v3
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v9, v5
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v7, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v8, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v11, v9, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v12, -v3, v10
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v13, -v4, v11
-; GFX10-IEEE-NEXT: v_add_f32_e32 v12, v12, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v13, v13, v9
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v12, v12, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v13, v13, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v12, v10
-; GFX10-IEEE-NEXT: v_add_f32_e32 v11, v13, v11
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v10
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v4, -v4, v11
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v9
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v7
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v4
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -v1, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, v0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, v0, 1, v8 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v8, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v8, v6
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -v1, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, v0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, v0, 1, v8 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v8, v4
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v10
-; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v11
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v5, 16, v0
+; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v6
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v3, v1, v0
@@ -1013,30 +1003,30 @@ define <2 x half> @v_fdiv_v2f16(<2 x half> %a, <2 x half> %b) {
; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v0
+; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v7, 16, v0
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v8, v0
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v6, v3
-; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v9, v5
-; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v7, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v10, v8, v6
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v11, v9, v7
-; GFX10-FLUSH-NEXT: v_mad_f32 v12, -v3, v10, v8
-; GFX10-FLUSH-NEXT: v_mad_f32 v13, -v4, v11, v9
-; GFX10-FLUSH-NEXT: v_mac_f32_e32 v10, v12, v6
-; GFX10-FLUSH-NEXT: v_mac_f32_e32 v11, v13, v7
-; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v10, v8
-; GFX10-FLUSH-NEXT: v_mad_f32 v4, -v4, v11, v9
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v7
+; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
+; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v10, v7
+; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v6, v4
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v9, v0, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v11, v0, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_mad_f32 v12, -v3, v9, v8
+; GFX10-FLUSH-NEXT: v_mad_f32 v13, -v4, v11, v10
+; GFX10-FLUSH-NEXT: v_mac_f32_e32 v9, v12, v5
+; GFX10-FLUSH-NEXT: v_mac_f32_e32 v11, v13, v6
+; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v8
+; GFX10-FLUSH-NEXT: v_mad_f32 v4, -v4, v11, v10
+; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v6
; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX10-FLUSH-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v10
+; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9
; GFX10-FLUSH-NEXT: v_add_f32_e32 v4, v4, v11
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v3, v1, v0
-; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v2, v5
+; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v2, v7
; GFX10-FLUSH-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-FLUSH-NEXT: s_setpc_b64 s[30:31]
;
@@ -1045,25 +1035,24 @@ define <2 x half> @v_fdiv_v2f16(<2 x half> %a, <2 x half> %b) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX11-NEXT: v_lshrrev_b32_e32 v5, 16, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v6, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
-; GFX11-NEXT: v_cvt_f32_f16_e32 v7, v5
; GFX11-NEXT: v_rcp_f32_e32 v4, v4
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_dual_mul_f32 v6, v6, v3 :: v_dual_mul_f32 v7, v7, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v9, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_fmac_f32 v6, v8, v3 :: v_dual_fmac_f32 v7, v9, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v9, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v3, v8, v3 :: v_dual_mul_f32 v4, v9, v4
-; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v6 :: v_dual_and_b32 v4, 0xff800000, v4
-; GFX11-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
+; GFX11-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v5, v7, v3 :: v_dual_fmac_f32 v6, v8, v4
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v4, v8, v4
+; GFX11-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-NEXT: v_dual_add_f32 v4, v4, v6 :: v_dual_and_b32 v3, 0xff800000, v3
+; GFX11-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX11-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX11-NEXT: v_cvt_f16_f32_e32 v4, v4
+; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v0, v3, v1, v0
; GFX11-NEXT: v_div_fixup_f16 v1, v4, v2, v5
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
@@ -1320,31 +1309,29 @@ define <2 x half> @v_fdiv_v2f16_ulp25(<2 x half> %a, <2 x half> %b) {
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v0
-; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v7, v5
+; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1
+; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v0
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, v3
-; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v7, v7
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v4, v0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v8, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v4, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, v0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v4
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v6, v7
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v7
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v6, v7
-; GFX9-FLUSH-NEXT: v_and_b32_e32 v6, 0xff800000, v6
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v4, v6, v4
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v7, v6, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX9-FLUSH-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v4, v4, v7
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v4, v4
+; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v1, v0
-; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v5, v3
+; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v3, v5
; GFX9-FLUSH-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-FLUSH-NEXT: s_setpc_b64 s[30:31]
;
@@ -1353,32 +1340,30 @@ define <2 x half> @v_fdiv_v2f16_ulp25(<2 x half> %a, <2 x half> %b) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v5, 16, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v8, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v6, v3
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v9, v5
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v7, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v8, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v11, v9, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v12, -v3, v10
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v13, -v4, v11
-; GFX10-IEEE-NEXT: v_add_f32_e32 v12, v12, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v13, v13, v9
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v12, v12, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v13, v13, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v12, v10
-; GFX10-IEEE-NEXT: v_add_f32_e32 v11, v13, v11
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v10
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v4, -v4, v11
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v9
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v7
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v4
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -v1, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, v0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, v0, 1, v8 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v8, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v8, v6
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -v1, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, v0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, v0, 1, v8 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v8, v4
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v10
-; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v11
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v5, 16, v0
+; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v6
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v3, v1, v0
@@ -1391,30 +1376,30 @@ define <2 x half> @v_fdiv_v2f16_ulp25(<2 x half> %a, <2 x half> %b) {
; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v5, 16, v0
+; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v7, 16, v0
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v8, v0
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v6, v3
-; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v9, v5
-; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v7, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v10, v8, v6
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v11, v9, v7
-; GFX10-FLUSH-NEXT: v_mad_f32 v12, -v3, v10, v8
-; GFX10-FLUSH-NEXT: v_mad_f32 v13, -v4, v11, v9
-; GFX10-FLUSH-NEXT: v_mac_f32_e32 v10, v12, v6
-; GFX10-FLUSH-NEXT: v_mac_f32_e32 v11, v13, v7
-; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v10, v8
-; GFX10-FLUSH-NEXT: v_mad_f32 v4, -v4, v11, v9
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v7
+; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
+; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v10, v7
+; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v6, v4
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v9, v0, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v11, v0, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_mad_f32 v12, -v3, v9, v8
+; GFX10-FLUSH-NEXT: v_mad_f32 v13, -v4, v11, v10
+; GFX10-FLUSH-NEXT: v_mac_f32_e32 v9, v12, v5
+; GFX10-FLUSH-NEXT: v_mac_f32_e32 v11, v13, v6
+; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v8
+; GFX10-FLUSH-NEXT: v_mad_f32 v4, -v4, v11, v10
+; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v6
; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX10-FLUSH-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v10
+; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9
; GFX10-FLUSH-NEXT: v_add_f32_e32 v4, v4, v11
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v4, v4
; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v3, v1, v0
-; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v2, v5
+; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v4, v2, v7
; GFX10-FLUSH-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX10-FLUSH-NEXT: s_setpc_b64 s[30:31]
;
@@ -1423,25 +1408,24 @@ define <2 x half> @v_fdiv_v2f16_ulp25(<2 x half> %a, <2 x half> %b) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX11-NEXT: v_lshrrev_b32_e32 v5, 16, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v6, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
-; GFX11-NEXT: v_cvt_f32_f16_e32 v7, v5
; GFX11-NEXT: v_rcp_f32_e32 v4, v4
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_dual_mul_f32 v6, v6, v3 :: v_dual_mul_f32 v7, v7, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v9, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_fmac_f32 v6, v8, v3 :: v_dual_fmac_f32 v7, v9, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v9, -v1, v7, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v3, v8, v3 :: v_dual_mul_f32 v4, v9, v4
-; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v6 :: v_dual_and_b32 v4, 0xff800000, v4
-; GFX11-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
+; GFX11-NEXT: v_fma_mix_f32 v5, v0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, v0, v4, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v5, v7, v3 :: v_dual_fmac_f32 v6, v8, v4
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, v0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v0 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v4, v8, v4
+; GFX11-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-NEXT: v_dual_add_f32 v4, v4, v6 :: v_dual_and_b32 v3, 0xff800000, v3
+; GFX11-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX11-NEXT: v_lshrrev_b32_e32 v5, 16, v0
; GFX11-NEXT: v_cvt_f16_f32_e32 v4, v4
+; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v0, v3, v1, v0
; GFX11-NEXT: v_div_fixup_f16 v1, v4, v2, v5
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
@@ -1631,23 +1615,22 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) {
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0
; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, 1.0, v1, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v1
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v5, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0
@@ -1660,30 +1643,29 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, 1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v6, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, 1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
@@ -1700,8 +1682,8 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, 1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -1726,26 +1708,24 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -1934,23 +1914,22 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) {
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0
; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -1.0, v1, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v1
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v6, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v5, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0
@@ -1963,30 +1942,29 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, -1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v6, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
@@ -2003,8 +1981,8 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, -1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -2029,26 +2007,24 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -2251,23 +2227,22 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) {
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, 1.0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v5, v4
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v7, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v4
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -|v0|, v7, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v7, v6, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v7, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v5
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v7
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0
@@ -2278,38 +2253,37 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) {
; GFX10-IEEE-LABEL: v_rcp_v2f16_fabs:
; GFX10-IEEE: ; %bb.0:
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0
-; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GFX10-IEEE-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0
+; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5
-; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
+; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v4
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -|v0|, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, 1.0, 1, v8 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v8, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v8, v6
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v0, -|v0|, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v0, 1.0, 1, v0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
-; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0xff800000, v0
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v6
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
+; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
-; GFX10-IEEE-NEXT: v_pack_b32_f16 v0, v0, v1
+; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v0, v2, 1.0
+; GFX10-IEEE-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX10-IEEE-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-FLUSH-LABEL: v_rcp_v2f16_fabs:
@@ -2322,8 +2296,8 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, 1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -2346,30 +2320,31 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) {
; GFX11-LABEL: v_rcp_v2f16_fabs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_cvt_f32_f16_e32 v5, 1.0
; GFX11-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX11-NEXT: v_rcp_f32_e32 v3, v3
-; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v6, v5, v3
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v3
; GFX11-NEXT: v_rcp_f32_e32 v4, v4
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1]
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v5, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v8, v4
-; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v0, v0, v4
+; GFX11-NEXT: v_fma_mix_f32 v6, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fmac_f32_e32 v6, v8, v4
+; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX11-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX11-NEXT: v_dual_add_f32 v0, v0, v5 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_add_f32_e32 v3, v3, v6
+; GFX11-NEXT: v_add_f32_e32 v0, v0, v6
+; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
-; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
+; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v0, v0, v2, 1.0
+; GFX11-NEXT: s_waitcnt_depctr 0xfff
+; GFX11-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -2573,23 +2548,22 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) {
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, -1.0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v5, v4
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v7, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v4
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -|v0|, v7, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v7, v6, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v7, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v5
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v7
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0
@@ -2600,38 +2574,37 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) {
; GFX10-IEEE-LABEL: v_neg_rcp_v2f16_fabs:
; GFX10-IEEE: ; %bb.0:
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, -1.0
-; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GFX10-IEEE-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0
+; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5
-; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
+; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v4
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -|v0|, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v8, -1.0, 1, v8 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v8, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v8, v6
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v0, -|v0|, v6, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v0, -1.0, 1, v0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
-; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0xff800000, v0
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v6
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
+; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
-; GFX10-IEEE-NEXT: v_pack_b32_f16 v0, v0, v1
+; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v0, v2, -1.0
+; GFX10-IEEE-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX10-IEEE-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-FLUSH-LABEL: v_neg_rcp_v2f16_fabs:
@@ -2644,8 +2617,8 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, -1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -2668,30 +2641,31 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) {
; GFX11-LABEL: v_neg_rcp_v2f16_fabs:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_cvt_f32_f16_e32 v5, -1.0
; GFX11-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX11-NEXT: v_rcp_f32_e32 v3, v3
-; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v6, v5, v3
; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v3
; GFX11-NEXT: v_rcp_f32_e32 v4, v4
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1]
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v5, v4
-; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v8, v4
-; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v0, v0, v4
+; GFX11-NEXT: v_fma_mix_f32 v6, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v6, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fmac_f32_e32 v6, v8, v4
+; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v6, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_mul_f32_e32 v0, v0, v4
; GFX11-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX11-NEXT: v_dual_add_f32 v0, v0, v5 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_add_f32_e32 v3, v3, v6
+; GFX11-NEXT: v_add_f32_e32 v0, v0, v6
+; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0
-; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
+; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v0, v0, v2, -1.0
+; GFX11-NEXT: s_waitcnt_depctr 0xfff
+; GFX11-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_add_f32_e32 v3, v3, v5
+; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v1, v0
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -3041,23 +3015,22 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) {
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0
; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, 1.0, v1, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v1
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v5, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0
@@ -3070,30 +3043,29 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, 1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v3
+; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v6, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v5, v7, v5
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 0 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, 1.0, 1, v6 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, 1.0, 1, v7 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
@@ -3110,8 +3082,8 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, 1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -3136,26 +3108,24 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -3510,36 +3480,34 @@ define amdgpu_ps i16 @s_fdiv_f16(i16 inreg %a.arg, i16 inreg %b.arg) {
; GFX9-FLUSH-LABEL: s_fdiv_f16:
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s0
-; GFX9-FLUSH-NEXT: v_mov_b32_e32 v2, s1
+; GFX9-FLUSH-NEXT: v_mov_b32_e32 v1, s1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v0, v0
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v0
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, -v2, v1, s0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v1, v3, v0
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, -v2, v1, s0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v2, s0, v0, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, -v1, v2, s0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v2, v3, v0
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, -v1, v2, s0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v3, v0
; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0
-; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v0, v2, s0
+; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v0, v1, s0
; GFX9-FLUSH-NEXT: v_readfirstlane_b32 s0, v0
; GFX9-FLUSH-NEXT: ; return to shader part epilog
;
; GFX10-IEEE-LABEL: s_fdiv_f16:
; GFX10-IEEE: ; %bb.0:
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, s0
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v1, v0
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v2, v1
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v4, -v0, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v4, v4, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v1
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v4, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v0, -v0, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v0, v0
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v1, s0, v0, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v2, -s1, v1, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v2, s0, 1, v2 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v0
+; GFX10-IEEE-NEXT: v_add_f32_e32 v1, v2, v1
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v2, -s1, v1, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v2, s0, 1, v2 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v2, v0
; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v3
+; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v0, s1, s0
; GFX10-IEEE-NEXT: v_readfirstlane_b32 s0, v0
@@ -3550,7 +3518,7 @@ define amdgpu_ps i16 @s_fdiv_f16(i16 inreg %a.arg, i16 inreg %b.arg) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s1
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, s0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v1, v0
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v2, v1
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v3, s0, v1, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v4, -v0, v3, v2
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v3, v4, v1
; GFX10-FLUSH-NEXT: v_mad_f32 v0, -v0, v3, v2
@@ -3565,10 +3533,9 @@ define amdgpu_ps i16 @s_fdiv_f16(i16 inreg %a.arg, i16 inreg %b.arg) {
; GFX11-LABEL: s_fdiv_f16:
; GFX11: ; %bb.0:
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, s1
-; GFX11-NEXT: v_cvt_f32_f16_e32 v1, s0
; GFX11-NEXT: v_rcp_f32_e32 v0, v0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v1, v1, v0
+; GFX11-NEXT: v_fma_mix_f32 v1, s0, v0, 0 op_sel_hi:[1,0,0]
; GFX11-NEXT: v_fma_mix_f32 v2, -s1, v1, s0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_fmac_f32_e32 v1, v2, v0
; GFX11-NEXT: v_fma_mix_f32 v2, -s1, v1, s0 op_sel_hi:[1,0,1]
@@ -3894,33 +3861,31 @@ define amdgpu_ps i32 @s_fdiv_v2f16(i32 inreg %a.arg, i32 inreg %b.arg) {
; GFX9-FLUSH-LABEL: s_fdiv_v2f16:
; GFX9-FLUSH: ; %bb.0:
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s0
-; GFX9-FLUSH-NEXT: s_lshr_b32 s3, s1, 16
+; GFX9-FLUSH-NEXT: s_lshr_b32 s2, s1, 16
+; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s2
; GFX9-FLUSH-NEXT: v_mov_b32_e32 v2, s1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v0, v0
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, s3
-; GFX9-FLUSH-NEXT: s_lshr_b32 s2, s0, 16
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v0
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v2, v1, s0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v1, v4, v0
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v2, v1, s0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: s_lshr_b32 s3, s0, 16
+; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v3, s0, v0, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v2, v3, s0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v3, v4, v0
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v2, v3, s0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v4, v0
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, s2
-; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX9-FLUSH-NEXT: v_mov_b32_e32 v1, s3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v4, s2 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v4, s2 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3
-; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v3
+; GFX9-FLUSH-NEXT: v_mov_b32_e32 v3, s2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, s3, v1, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v3, v4, s3 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v1
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v3, v4, s3 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v5, v1
+; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v4
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0
-; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
+; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v0, v2, s0
-; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v3, v1, s2
+; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v1, v3, s3
; GFX9-FLUSH-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX9-FLUSH-NEXT: v_readfirstlane_b32 s0, v0
; GFX9-FLUSH-NEXT: ; return to shader part epilog
@@ -3931,30 +3896,28 @@ define amdgpu_ps i32 @s_fdiv_v2f16(i32 inreg %a.arg, i32 inreg %b.arg) {
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v1, s2
; GFX10-IEEE-NEXT: s_lshr_b32 s3, s0, 16
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, s0
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v5, s3
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v0
-; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v1
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v4, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v5, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v8, -v0, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v1, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v8, v4
-; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v8, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v8, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v0, -v0, v6
-; GFX10-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7
-; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v4
-; GFX10-IEEE-NEXT: v_add_f32_e32 v1, v1, v5
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v0, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v1, v1, v3
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v0, v0
+; GFX10-IEEE-NEXT: v_rcp_f32_e32 v1, v1
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v2, s0, v0, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v3, s3, v1, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -s1, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -s2, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, s0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, s3, 1, v5 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v0
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v5, v1
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v4, v2
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v5, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -s1, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -s2, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, s0, 1, v4 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, s3, 1, v5 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v0, v4, v0
+; GFX10-IEEE-NEXT: v_mul_f32_e32 v1, v5, v1
; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0xff800000, v0
; GFX10-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1
-; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v6
-; GFX10-IEEE-NEXT: v_add_f32_e32 v1, v1, v7
+; GFX10-IEEE-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-IEEE-NEXT: v_add_f32_e32 v1, v1, v3
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v0, s1, s0
@@ -3970,22 +3933,22 @@ define amdgpu_ps i32 @s_fdiv_v2f16(i32 inreg %a.arg, i32 inreg %b.arg) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s2
; GFX10-FLUSH-NEXT: s_lshr_b32 s3, s0, 16
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, s0
-; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, s3
+; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, s3
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v2, v0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v3, v1
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v6, v4, v2
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v5, v3
-; GFX10-FLUSH-NEXT: v_mad_f32 v8, -v0, v6, v4
-; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v1, v7, v5
-; GFX10-FLUSH-NEXT: v_mac_f32_e32 v6, v8, v2
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v5, s0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, s3, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_mad_f32 v8, -v0, v5, v4
+; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v1, v7, v6
+; GFX10-FLUSH-NEXT: v_mac_f32_e32 v5, v8, v2
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v3
-; GFX10-FLUSH-NEXT: v_mad_f32 v0, -v0, v6, v4
-; GFX10-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v5
+; GFX10-FLUSH-NEXT: v_mad_f32 v0, -v0, v5, v4
+; GFX10-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v6
; GFX10-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v2
; GFX10-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v3
; GFX10-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0
; GFX10-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1
-; GFX10-FLUSH-NEXT: v_add_f32_e32 v0, v0, v6
+; GFX10-FLUSH-NEXT: v_add_f32_e32 v0, v0, v5
; GFX10-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0
; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1
@@ -4001,12 +3964,11 @@ define amdgpu_ps i32 @s_fdiv_v2f16(i32 inreg %a.arg, i32 inreg %b.arg) {
; GFX11-NEXT: v_cvt_f32_f16_e32 v0, s1
; GFX11-NEXT: v_cvt_f32_f16_e32 v1, s2
; GFX11-NEXT: s_lshr_b32 s3, s0, 16
-; GFX11-NEXT: v_cvt_f32_f16_e32 v2, s0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v3, s3
; GFX11-NEXT: v_rcp_f32_e32 v0, v0
; GFX11-NEXT: v_rcp_f32_e32 v1, v1
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_dual_mul_f32 v2, v2, v0 :: v_dual_mul_f32 v3, v3, v1
+; GFX11-NEXT: v_fma_mix_f32 v2, s0, v0, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v3, s3, v1, 0 op_sel_hi:[1,0,0]
; GFX11-NEXT: v_fma_mix_f32 v4, -s1, v2, s0 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_fma_mix_f32 v5, -s2, v3, s3 op_sel_hi:[1,0,1]
; GFX11-NEXT: v_dual_fmac_f32 v2, v4, v0 :: v_dual_fmac_f32 v3, v5, v1
@@ -4434,25 +4396,24 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) {
; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v0, s0
; GFX9-FLUSH-NEXT: s_lshr_b32 s0, s0, 16
; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, s0
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v5, v2
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v6, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v5, v3
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v6, -1.0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v7, v2
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v6, v3
+; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
@@ -4466,25 +4427,24 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) {
; GFX10-IEEE-NEXT: s_lshr_b32 s1, s0, 16
; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v0, s0
; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, s1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v6, v2
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
@@ -4503,8 +4463,8 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, -1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -4530,28 +4490,26 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) {
; GFX11-NEXT: s_lshr_b32 s1, s0, 16
; GFX11-NEXT: v_sqrt_f16_e32 v0, s0
; GFX11-NEXT: v_sqrt_f16_e32 v1, s1
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: v_readfirstlane_b32 s0, v0
@@ -5549,25 +5507,24 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) {
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0
; GFX9-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v7, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v2
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v3
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v6, v2
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v5
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0
@@ -5580,25 +5537,24 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, v0
; GFX10-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v6, v2
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1]
; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0
@@ -5616,8 +5572,8 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, 1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, 1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -5642,7 +5598,6 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_sqrt_f16_e32 v0, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0
; GFX11-NEXT: v_sqrt_f16_e32 v1, v1
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
@@ -5650,21 +5605,20 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) {
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, 1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, 1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, 1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
@@ -5863,25 +5817,24 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) {
; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0
; GFX9-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2
-; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v7, v3
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v2
+; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v3
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v6, v2
; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v5
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0
@@ -5894,25 +5847,24 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) {
; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, v0
; GFX10-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1
; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v6, v2
+; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1]
; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2
; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3
; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2
; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v4
+; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2
; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0
@@ -5930,8 +5882,8 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) {
; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2
; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4
-; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v7, -1.0, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-FLUSH-NEXT: v_fma_mix_f32 v8, -1.0, v5, 0 op_sel_hi:[1,0,0]
; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6
; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6
; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4
@@ -5956,7 +5908,6 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) {
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX11-NEXT: v_sqrt_f16_e32 v0, v0
-; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0
; GFX11-NEXT: v_sqrt_f16_e32 v1, v1
; GFX11-NEXT: s_waitcnt_depctr 0xfff
; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0
@@ -5964,21 +5915,20 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) {
; GFX11-NEXT: v_rcp_f32_e32 v2, v2
; GFX11-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2
-; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2
-; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3
-; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1]
-; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3
-; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_fma_mix_f32 v4, -1.0, v2, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v5, -1.0, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_fmac_f32 v4, v6, v2 :: v_dual_fmac_f32 v5, v7, v3
+; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v4, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v5, -1.0 op_sel_hi:[1,0,1]
+; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_mul_f32 v3, v7, v3
+; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX11-NEXT: v_dual_add_f32 v3, v3, v5 :: v_dual_and_b32 v2, 0xff800000, v2
+; GFX11-NEXT: v_add_f32_e32 v2, v2, v4
; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3
-; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2
+; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0
; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0
; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1
; GFX11-NEXT: s_setpc_b64 s[30:31]
diff --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
index 9e240238c1066..f8bfe2bf1e2f1 100644
--- a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
+++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
@@ -625,26 +625,22 @@ define amdgpu_kernel void @introduced_copy_to_sgpr(i64 %arg, i32 %arg1, i32 %arg
; GFX908-NEXT: s_cbranch_vccnz .LBB3_7
; GFX908-NEXT: ; %bb.6: ; %bb51
; GFX908-NEXT: ; in Loop: Header=BB3_5 Depth=2
-; GFX908-NEXT: v_cvt_f32_f16_sdwa v22, v21 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT: v_cvt_f32_f16_e32 v21, v21
-; GFX908-NEXT: v_cvt_f32_f16_sdwa v23, v20 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT: v_cvt_f32_f16_e32 v20, v20
-; GFX908-NEXT: v_add_f32_e32 v24, v18, v12
-; GFX908-NEXT: v_add_f32_e32 v25, v19, v13
-; GFX908-NEXT: v_add_f32_e32 v26, 0, v12
-; GFX908-NEXT: v_add_f32_e32 v27, 0, v13
-; GFX908-NEXT: v_add_f32_e32 v15, v22, v15
-; GFX908-NEXT: v_add_f32_e32 v14, v21, v14
-; GFX908-NEXT: v_add_f32_e32 v13, v23, v13
-; GFX908-NEXT: v_add_f32_e32 v12, v20, v12
-; GFX908-NEXT: v_add_f32_e32 v5, v5, v25
-; GFX908-NEXT: v_add_f32_e32 v4, v4, v24
-; GFX908-NEXT: v_add_f32_e32 v7, v7, v27
-; GFX908-NEXT: v_add_f32_e32 v6, v6, v26
-; GFX908-NEXT: v_add_f32_e32 v8, v8, v14
+; GFX908-NEXT: v_add_f32_e32 v22, v18, v12
+; GFX908-NEXT: v_add_f32_e32 v23, v19, v13
+; GFX908-NEXT: v_add_f32_e32 v24, 0, v12
+; GFX908-NEXT: v_add_f32_e32 v25, 0, v13
+; GFX908-NEXT: v_fma_mix_f32 v14, v21, 1, v14 op_sel_hi:[1,0,0]
+; GFX908-NEXT: v_fma_mix_f32 v15, v21, 1, v15 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX908-NEXT: v_fma_mix_f32 v12, v20, 1, v12 op_sel_hi:[1,0,0]
+; GFX908-NEXT: v_fma_mix_f32 v13, v20, 1, v13 op_sel:[1,0,0] op_sel_hi:[1,0,0]
+; GFX908-NEXT: v_add_f32_e32 v5, v5, v23
+; GFX908-NEXT: v_add_f32_e32 v4, v4, v22
+; GFX908-NEXT: v_add_f32_e32 v7, v7, v25
+; GFX908-NEXT: v_add_f32_e32 v6, v6, v24
; GFX908-NEXT: v_add_f32_e32 v9, v9, v15
-; GFX908-NEXT: v_add_f32_e32 v10, v10, v12
+; GFX908-NEXT: v_add_f32_e32 v8, v8, v14
; GFX908-NEXT: v_add_f32_e32 v11, v11, v13
+; GFX908-NEXT: v_add_f32_e32 v10, v10, v12
; GFX908-NEXT: s_branch .LBB3_4
; GFX908-NEXT: .LBB3_7: ; in Loop: Header=BB3_5 Depth=2
; GFX908-NEXT: s_mov_b64 s[22:23], s[18:19]
diff --git a/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll b/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll
index 210e09fd9169a..fe6a75d4896ca 100644
--- a/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fdiv.f16.ll
@@ -89,10 +89,9 @@ define amdgpu_kernel void @v_fdiv_f16(
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] glc
; GFX9-NEXT: s_waitcnt vmcnt(0)
-; GFX9-NEXT: v_cvt_f32_f16_e32 v4, v1
; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX9-NEXT: v_rcp_f32_e32 v3, v3
-; GFX9-NEXT: v_mul_f32_e32 v4, v4, v3
+; GFX9-NEXT: v_mad_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX9-NEXT: v_mac_f32_e32 v4, v5, v3
; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
@@ -115,16 +114,16 @@ define amdgpu_kernel void @v_fdiv_f16(
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] glc dlc
; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: v_cvt_f32_f16_e32 v5, v1
+; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v1
; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX10-NEXT: v_rcp_f32_e32 v4, v3
-; GFX10-NEXT: v_mul_f32_e32 v6, v5, v4
-; GFX10-NEXT: v_mad_f32 v7, -v3, v6, v5
-; GFX10-NEXT: v_mac_f32_e32 v6, v7, v4
-; GFX10-NEXT: v_mad_f32 v3, -v3, v6, v5
+; GFX10-NEXT: v_fma_mix_f32 v5, v1, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_mad_f32 v7, -v3, v5, v6
+; GFX10-NEXT: v_mac_f32_e32 v5, v7, v4
+; GFX10-NEXT: v_mad_f32 v3, -v3, v5, v6
; GFX10-NEXT: v_mul_f32_e32 v3, v3, v4
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-NEXT: v_add_f32_e32 v3, v3, v6
+; GFX10-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-NEXT: v_div_fixup_f16 v1, v3, v2, v1
; GFX10-NEXT: global_store_short v0, v1, s[0:1]
@@ -143,23 +142,22 @@ define amdgpu_kernel void @v_fdiv_f16(
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] glc dlc
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v0.l
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v4, v3
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v4, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v4, v7, v3
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v4, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX11-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v4
+; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.l, v0.h, v1.l, v0.l
@@ -179,12 +177,11 @@ define amdgpu_kernel void @v_fdiv_f16(
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] glc dlc
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v1
; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-FAKE16-NEXT: v_mul_f32_e32 v4, v4, v3
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
diff --git a/llvm/test/CodeGen/AMDGPU/fpext-free.ll b/llvm/test/CodeGen/AMDGPU/fpext-free.ll
index b88cb210c91e8..428b80ebf396e 100644
--- a/llvm/test/CodeGen/AMDGPU/fpext-free.ll
+++ b/llvm/test/CodeGen/AMDGPU/fpext-free.ll
@@ -14,18 +14,16 @@ define float @fadd_fpext_fmul_f16_to_f32(half %x, half %y, float %z) #0 {
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v0, 1, v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fadd_fpext_fmul_f16_to_f32:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v0, 1, v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fadd_fpext_fmul_f16_to_f32:
@@ -118,18 +116,16 @@ define float @fadd_fpext_fmul_f16_to_f32_commute(half %x, half %y, float %z) #0
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v2, v0
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v0, 1, v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fadd_fpext_fmul_f16_to_f32_commute:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v2, v0
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v0, 1, v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fadd_fpext_fmul_f16_to_f32_commute:
@@ -380,12 +376,10 @@ define float @fadd_fpext_fmuladd_f16_to_f32(float %x, half %y, half %z, half %u,
; GFX11-TRUE16-LABEL: fadd_fpext_fmuladd_f16_to_f32:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.h, v3.l, v4.l
+; GFX11-TRUE16-NEXT: v_mul_f16_e32 v3.l, v3.l, v4.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v1.h, v1.l, v2.l
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v3.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fadd_fpext_fmuladd_f16_to_f32:
@@ -394,9 +388,7 @@ define float @fadd_fpext_fmuladd_f16_to_f32(float %x, half %y, half %z, half %u,
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fadd_fpext_fmuladd_f16_to_f32:
@@ -426,12 +418,10 @@ define float @fadd_fpext_fma_f16_to_f32(float %x, half %y, half %z, half %u, hal
; GFX11-TRUE16-LABEL: fadd_fpext_fma_f16_to_f32:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.h, v3.l, v4.l
+; GFX11-TRUE16-NEXT: v_mul_f16_e32 v3.l, v3.l, v4.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v1.h, v1.l, v2.l
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v3.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fadd_fpext_fma_f16_to_f32:
@@ -440,9 +430,7 @@ define float @fadd_fpext_fma_f16_to_f32(float %x, half %y, half %z, half %u, hal
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fadd_fpext_fma_f16_to_f32:
@@ -472,12 +460,10 @@ define float @fadd_fpext_fma_f16_to_f32_commute(float %x, half %y, half %z, half
; GFX11-TRUE16-LABEL: fadd_fpext_fma_f16_to_f32_commute:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.h, v3.l, v4.l
+; GFX11-TRUE16-NEXT: v_mul_f16_e32 v3.l, v3.l, v4.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v1.h, v1.l, v2.l
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v0, v1, v0
+; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v3.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fadd_fpext_fma_f16_to_f32_commute:
@@ -486,9 +472,7 @@ define float @fadd_fpext_fma_f16_to_f32_commute(float %x, half %y, half %z, half
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v0, v1, v0
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v3, 1, v0 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fadd_fpext_fma_f16_to_f32_commute:
@@ -521,18 +505,16 @@ define float @fsub_fpext_fmul_f16_to_f32(half %x, half %y, float %z) #0 {
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; GFX11-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fsub_fpext_fmul_f16_to_f32:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX11-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fsub_fpext_fmul_f16_to_f32:
@@ -568,18 +550,16 @@ define float @fsub_fpext_fmul_f16_to_f32_commute(float %x, half %y, half %z) #0
; GFX11-F32DENORM-TRUE16: ; %bb.0: ; %entry
; GFX11-F32DENORM-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-F32DENORM-TRUE16-NEXT: v_mul_f16_e32 v1.l, v1.l, v2.l
-; GFX11-F32DENORM-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-F32DENORM-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
-; GFX11-F32DENORM-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX11-F32DENORM-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-F32DENORM-TRUE16-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[0,0,1]
; GFX11-F32DENORM-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-F32DENORM-FAKE16-LABEL: fsub_fpext_fmul_f16_to_f32_commute:
; GFX11-F32DENORM-FAKE16: ; %bb.0: ; %entry
; GFX11-F32DENORM-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-F32DENORM-FAKE16-NEXT: v_mul_f16_e32 v1, v1, v2
-; GFX11-F32DENORM-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-F32DENORM-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GFX11-F32DENORM-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX11-F32DENORM-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-F32DENORM-FAKE16-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[0,0,1]
; GFX11-F32DENORM-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fsub_fpext_fmul_f16_to_f32_commute:
@@ -609,18 +589,16 @@ define float @fsub_fpext_fneg_fmul_f16_to_f32(half %x, half %y, float %z) #0 {
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_mul_f16_e64 v0.l, v0.l, -v1.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; GFX11-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fsub_fpext_fneg_fmul_f16_to_f32:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_mul_f16_e64 v0, v0, -v1
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX11-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fsub_fpext_fneg_fmul_f16_to_f32:
@@ -651,18 +629,16 @@ define float @fsub_fneg_fpext_fmul_f16_to_f32(half %x, half %y, float %z) #0 {
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_mul_f16_e64 v0.l, v0.l, -v1.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; GFX11-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fsub_fneg_fpext_fmul_f16_to_f32:
; GFX11-FAKE16: ; %bb.0: ; %entry
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-FAKE16-NEXT: v_mul_f16_e64 v0, v0, -v1
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX11-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-F32FLUSH-LABEL: fsub_fneg_fpext_fmul_f16_to_f32:
@@ -738,12 +714,10 @@ define float @fsub_fpext_muladd_mul_f16_to_f32(half %x, half %y, float %z, half
; GFX11-TRUE16-LABEL: fsub_fpext_muladd_mul_f16_to_f32:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_mul_f16_e32 v0.h, v3.l, v4.l
+; GFX11-TRUE16-NEXT: v_mul_f16_e32 v3.l, v3.l, v4.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v0.h, v0.l, v1.l
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v3.l, v0.l, v1.l
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, v3, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fsub_fpext_muladd_mul_f16_to_f32:
@@ -752,19 +726,25 @@ define float @fsub_fpext_muladd_mul_f16_to_f32(half %x, half %y, float %z, half
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f16_e32 v3, v0, v1
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v3
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, v3, 1, -v2 op_sel_hi:[1,0,0]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX89-LABEL: fsub_fpext_muladd_mul_f16_to_f32:
-; GFX89: ; %bb.0: ; %entry
-; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX89-NEXT: v_mul_f16_e32 v3, v3, v4
-; GFX89-NEXT: v_fma_f16 v0, v0, v1, v3
-; GFX89-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX89-NEXT: v_sub_f32_e32 v0, v0, v2
-; GFX89-NEXT: s_setpc_b64 s[30:31]
+; GFX9-F32FLUSH-LABEL: fsub_fpext_muladd_mul_f16_to_f32:
+; GFX9-F32FLUSH: ; %bb.0: ; %entry
+; GFX9-F32FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-F32FLUSH-NEXT: v_mul_f16_e32 v3, v3, v4
+; GFX9-F32FLUSH-NEXT: v_fma_f16 v0, v0, v1, v3
+; GFX9-F32FLUSH-NEXT: v_mad_mix_f32 v0, v0, 1, -v2 op_sel_hi:[1,0,0]
+; GFX9-F32FLUSH-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-F32DENORM-LABEL: fsub_fpext_muladd_mul_f16_to_f32:
+; GFX9-F32DENORM: ; %bb.0: ; %entry
+; GFX9-F32DENORM-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-F32DENORM-NEXT: v_mul_f16_e32 v3, v3, v4
+; GFX9-F32DENORM-NEXT: v_fma_f16 v0, v0, v1, v3
+; GFX9-F32DENORM-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9-F32DENORM-NEXT: v_sub_f32_e32 v0, v0, v2
+; GFX9-F32DENORM-NEXT: s_setpc_b64 s[30:31]
entry:
%mul = fmul half %u, %v
%fma = call half @llvm.fmuladd.f16(half %x, half %y, half %mul)
@@ -824,12 +804,10 @@ define float @fsub_fpext_muladd_mul_f16_to_f32_commute(float %x, half %y, half %
; GFX11-TRUE16-LABEL: fsub_fpext_muladd_mul_f16_to_f32_commute:
; GFX11-TRUE16: ; %bb.0: ; %entry
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_mul_f16_e32 v1.h, v3.l, v4.l
+; GFX11-TRUE16-NEXT: v_mul_f16_e32 v3.l, v3.l, v4.l
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v1.h, v1.l, v2.l
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.h
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX11-TRUE16-NEXT: v_fmac_f16_e32 v3.l, v1.l, v2.l
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v0, -v0, 1, v3 op_sel_hi:[0,0,1]
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-FAKE16-LABEL: fsub_fpext_muladd_mul_f16_to_f32_commute:
@@ -838,19 +816,25 @@ define float @fsub_fpext_muladd_mul_f16_to_f32_commute(float %x, half %y, half %
; GFX11-FAKE16-NEXT: v_mul_f16_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_fmac_f16_e32 v3, v1, v2
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v3
-; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v0, -v0, 1, v3 op_sel_hi:[0,0,1]
; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX89-LABEL: fsub_fpext_muladd_mul_f16_to_f32_commute:
-; GFX89: ; %bb.0: ; %entry
-; GFX89-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX89-NEXT: v_mul_f16_e32 v3, v3, v4
-; GFX89-NEXT: v_fma_f16 v1, v1, v2, v3
-; GFX89-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GFX89-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX89-NEXT: s_setpc_b64 s[30:31]
+; GFX9-F32FLUSH-LABEL: fsub_fpext_muladd_mul_f16_to_f32_commute:
+; GFX9-F32FLUSH: ; %bb.0: ; %entry
+; GFX9-F32FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-F32FLUSH-NEXT: v_mul_f16_e32 v3, v3, v4
+; GFX9-F32FLUSH-NEXT: v_fma_f16 v1, v1, v2, v3
+; GFX9-F32FLUSH-NEXT: v_mad_mix_f32 v0, -v0, 1, v1 op_sel_hi:[0,0,1]
+; GFX9-F32FLUSH-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-F32DENORM-LABEL: fsub_fpext_muladd_mul_f16_to_f32_commute:
+; GFX9-F32DENORM: ; %bb.0: ; %entry
+; GFX9-F32DENORM-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-F32DENORM-NEXT: v_mul_f16_e32 v3, v3, v4
+; GFX9-F32DENORM-NEXT: v_fma_f16 v1, v1, v2, v3
+; GFX9-F32DENORM-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9-F32DENORM-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX9-F32DENORM-NEXT: s_setpc_b64 s[30:31]
entry:
%mul = fmul half %u, %v
%fma = call half @llvm.fmuladd.f16(half %y, half %z, half %mul)
diff --git a/llvm/test/CodeGen/AMDGPU/frem.ll b/llvm/test/CodeGen/AMDGPU/frem.ll
index c4a38dcd7b5f3..1032656301220 100644
--- a/llvm/test/CodeGen/AMDGPU/frem.ll
+++ b/llvm/test/CodeGen/AMDGPU/frem.ll
@@ -1380,18 +1380,16 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[2:3]
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
-; GFX9-NEXT: s_waitcnt vmcnt(1)
-; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX9-NEXT: s_waitcnt vmcnt(0)
-; GFX9-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX9-NEXT: v_rcp_f32_e32 v4, v4
-; GFX9-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX9-NEXT: v_mac_f32_e32 v3, v5, v4
-; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX9-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX9-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX9-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v2
+; GFX9-NEXT: v_rcp_f32_e32 v3, v3
+; GFX9-NEXT: v_mad_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX9-NEXT: v_mac_f32_e32 v4, v5, v3
+; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX9-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX9-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX9-NEXT: v_add_f32_e32 v3, v3, v4
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-NEXT: v_div_fixup_f16 v3, v3, v2, v1
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
@@ -1410,17 +1408,17 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX10-NEXT: global_load_ushort v1, v0, s[2:3]
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
; GFX10-NEXT: s_waitcnt vmcnt(1)
-; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v1
+; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v1
; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-NEXT: v_rcp_f32_e32 v5, v4
-; GFX10-NEXT: v_mul_f32_e32 v6, v3, v5
-; GFX10-NEXT: v_mad_f32 v7, -v4, v6, v3
-; GFX10-NEXT: v_mac_f32_e32 v6, v7, v5
-; GFX10-NEXT: v_mad_f32 v3, -v4, v6, v3
-; GFX10-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v2
+; GFX10-NEXT: v_rcp_f32_e32 v4, v3
+; GFX10-NEXT: v_fma_mix_f32 v5, v1, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_mad_f32 v7, -v3, v5, v6
+; GFX10-NEXT: v_mac_f32_e32 v5, v7, v4
+; GFX10-NEXT: v_mad_f32 v3, -v3, v5, v6
+; GFX10-NEXT: v_mul_f32_e32 v3, v3, v4
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-NEXT: v_add_f32_e32 v3, v3, v6
+; GFX10-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-NEXT: v_div_fixup_f16 v3, v3, v2, v1
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
@@ -1439,24 +1437,23 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX11-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
+; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
+; GFX11-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -1476,23 +1473,21 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX11-FAKE16-NEXT: s_clause 0x1
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
+; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
-; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
+; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
@@ -1513,23 +1508,22 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX1150-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX1150-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX1150-TRUE16-NEXT: s_waitcnt vmcnt(1)
-; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX1150-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1150-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
+; GFX1150-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1150-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1150-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1150-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1150-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1150-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1150-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX1150-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -1551,22 +1545,20 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX1150-FAKE16-NEXT: s_clause 0x1
; GFX1150-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX1150-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX1150-FAKE16-NEXT: s_waitcnt vmcnt(1)
-; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX1150-FAKE16-NEXT: s_waitcnt vmcnt(0)
-; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1150-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1150-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1150-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1150-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1150-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1150-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1150-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1150-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX1150-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
@@ -1589,23 +1581,22 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX1200-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX1200-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX1200-TRUE16-NEXT: s_wait_loadcnt 0x1
-; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX1200-TRUE16-NEXT: s_wait_loadcnt 0x0
-; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1200-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
+; GFX1200-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1200-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1200-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1200-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1200-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1200-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1200-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX1200-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -1627,22 +1618,20 @@ define amdgpu_kernel void @fast_frem_f16(ptr addrspace(1) %out, ptr addrspace(1)
; GFX1200-FAKE16-NEXT: s_clause 0x1
; GFX1200-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX1200-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX1200-FAKE16-NEXT: s_wait_loadcnt 0x1
-; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX1200-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1200-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1200-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1200-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1200-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1200-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1200-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1200-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1200-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX1200-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
@@ -1787,18 +1776,16 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[2:3]
; GFX9-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
-; GFX9-NEXT: s_waitcnt vmcnt(1)
-; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX9-NEXT: s_waitcnt vmcnt(0)
-; GFX9-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX9-NEXT: v_rcp_f32_e32 v4, v4
-; GFX9-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX9-NEXT: v_mac_f32_e32 v3, v5, v4
-; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX9-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX9-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX9-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX9-NEXT: v_cvt_f32_f16_e32 v3, v2
+; GFX9-NEXT: v_rcp_f32_e32 v3, v3
+; GFX9-NEXT: v_mad_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX9-NEXT: v_mac_f32_e32 v4, v5, v3
+; GFX9-NEXT: v_mad_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX9-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX9-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX9-NEXT: v_add_f32_e32 v3, v3, v4
; GFX9-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX9-NEXT: v_div_fixup_f16 v3, v3, v2, v1
; GFX9-NEXT: v_trunc_f16_e32 v3, v3
@@ -1817,17 +1804,17 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX10-NEXT: global_load_ushort v1, v0, s[2:3]
; GFX10-NEXT: global_load_ushort v2, v0, s[6:7] offset:8
; GFX10-NEXT: s_waitcnt vmcnt(1)
-; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v1
+; GFX10-NEXT: v_cvt_f32_f16_e32 v6, v1
; GFX10-NEXT: s_waitcnt vmcnt(0)
-; GFX10-NEXT: v_cvt_f32_f16_e32 v4, v2
-; GFX10-NEXT: v_rcp_f32_e32 v5, v4
-; GFX10-NEXT: v_mul_f32_e32 v6, v3, v5
-; GFX10-NEXT: v_mad_f32 v7, -v4, v6, v3
-; GFX10-NEXT: v_mac_f32_e32 v6, v7, v5
-; GFX10-NEXT: v_mad_f32 v3, -v4, v6, v3
-; GFX10-NEXT: v_mul_f32_e32 v3, v3, v5
+; GFX10-NEXT: v_cvt_f32_f16_e32 v3, v2
+; GFX10-NEXT: v_rcp_f32_e32 v4, v3
+; GFX10-NEXT: v_fma_mix_f32 v5, v1, v4, 0 op_sel_hi:[1,0,0]
+; GFX10-NEXT: v_mad_f32 v7, -v3, v5, v6
+; GFX10-NEXT: v_mac_f32_e32 v5, v7, v4
+; GFX10-NEXT: v_mad_f32 v3, -v3, v5, v6
+; GFX10-NEXT: v_mul_f32_e32 v3, v3, v4
; GFX10-NEXT: v_and_b32_e32 v3, 0xff800000, v3
-; GFX10-NEXT: v_add_f32_e32 v3, v3, v6
+; GFX10-NEXT: v_add_f32_e32 v3, v3, v5
; GFX10-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX10-NEXT: v_div_fixup_f16 v3, v3, v2, v1
; GFX10-NEXT: v_trunc_f16_e32 v3, v3
@@ -1846,24 +1833,23 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX11-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX11-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(1)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
+; GFX11-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX11-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-TRUE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
-; GFX11-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
+; GFX11-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
+; GFX11-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX11-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
+; GFX11-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX11-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX11-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -1883,23 +1869,21 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX11-FAKE16-NEXT: s_clause 0x1
; GFX11-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX11-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(1)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0)
-; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX11-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
+; GFX11-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
; GFX11-FAKE16-NEXT: s_waitcnt_depctr 0xfff
-; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
-; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
-; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
+; GFX11-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
+; GFX11-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
-; GFX11-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
+; GFX11-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
+; GFX11-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX11-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX11-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
@@ -1920,23 +1904,22 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX1150-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX1150-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX1150-TRUE16-NEXT: s_waitcnt vmcnt(1)
-; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX1150-TRUE16-NEXT: s_waitcnt vmcnt(0)
-; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1150-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1150-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX1150-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
+; GFX1150-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1150-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1150-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1150-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1150-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1150-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1150-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1150-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX1150-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1150-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX1150-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -1958,22 +1941,20 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX1150-FAKE16-NEXT: s_clause 0x1
; GFX1150-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX1150-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX1150-FAKE16-NEXT: s_waitcnt vmcnt(1)
-; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX1150-FAKE16-NEXT: s_waitcnt vmcnt(0)
-; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX1150-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1150-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1150-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1150-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1150-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1150-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1150-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1150-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1150-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1150-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1150-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX1150-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1150-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX1150-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
@@ -1996,23 +1977,22 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX1200-TRUE16-NEXT: global_load_d16_b16 v0, v2, s[2:3]
; GFX1200-TRUE16-NEXT: global_load_d16_b16 v1, v2, s[4:5] offset:8
; GFX1200-TRUE16-NEXT: s_wait_loadcnt 0x1
-; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v0.l
+; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v4.l, v0.l
; GFX1200-TRUE16-NEXT: s_wait_loadcnt 0x0
-; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v4, v1.l
-; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v5.l, v1.l
-; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v6.l, v0.l
-; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1200-TRUE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1200-TRUE16-NEXT: v_cvt_f32_f16_e32 v3, v1.l
+; GFX1200-TRUE16-NEXT: v_mov_b16_e32 v6.l, v1.l
+; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(TRANS32_DEP_1)
+; GFX1200-TRUE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v5, v4, v3, 0 op_sel_hi:[1,0,0]
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v7, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1200-TRUE16-NEXT: v_fmac_f32_e32 v3, v7, v4
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v7, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1200-TRUE16-NEXT: v_fmac_f32_e32 v5, v7, v3
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v5, -v5, v3, v6 op_sel_hi:[1,0,1]
-; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1200-TRUE16-NEXT: v_fma_mix_f32 v4, -v6, v5, v4 op_sel_hi:[1,0,1]
+; GFX1200-TRUE16-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-TRUE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1200-TRUE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1200-TRUE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1200-TRUE16-NEXT: v_add_f32_e32 v3, v3, v5
; GFX1200-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1200-TRUE16-NEXT: v_cvt_f16_f32_e32 v0.h, v3
; GFX1200-TRUE16-NEXT: v_div_fixup_f16 v0.h, v0.h, v1.l, v0.l
@@ -2034,22 +2014,20 @@ define amdgpu_kernel void @unsafe_frem_f16(ptr addrspace(1) %out, ptr addrspace(
; GFX1200-FAKE16-NEXT: s_clause 0x1
; GFX1200-FAKE16-NEXT: global_load_u16 v1, v0, s[2:3]
; GFX1200-FAKE16-NEXT: global_load_u16 v2, v0, s[4:5] offset:8
-; GFX1200-FAKE16-NEXT: s_wait_loadcnt 0x1
-; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v1
; GFX1200-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v4, v2
+; GFX1200-FAKE16-NEXT: v_cvt_f32_f16_e32 v3, v2
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(TRANS32_DEP_1)
-; GFX1200-FAKE16-NEXT: v_rcp_f32_e32 v4, v4
-; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v3, v3, v4
+; GFX1200-FAKE16-NEXT: v_rcp_f32_e32 v3, v3
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v4, v1, v3, 0 op_sel_hi:[1,0,0]
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1200-FAKE16-NEXT: v_fmac_f32_e32 v3, v5, v4
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1200-FAKE16-NEXT: v_fmac_f32_e32 v4, v5, v3
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v3, v1 op_sel_hi:[1,0,1]
-; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX1200-FAKE16-NEXT: v_fma_mix_f32 v5, -v2, v4, v1 op_sel_hi:[1,0,1]
+; GFX1200-FAKE16-NEXT: v_mul_f32_e32 v3, v5, v3
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1200-FAKE16-NEXT: v_and_b32_e32 v4, 0xff800000, v4
-; GFX1200-FAKE16-NEXT: v_add_f32_e32 v3, v4, v3
+; GFX1200-FAKE16-NEXT: v_and_b32_e32 v3, 0xff800000, v3
+; GFX1200-FAKE16-NEXT: v_add_f32_e32 v3, v3, v4
; GFX1200-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1200-FAKE16-NEXT: v_cvt_f16_f32_e32 v3, v3
; GFX1200-FAKE16-NEXT: v_div_fixup_f16 v3, v3, v2, v1
diff --git a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
index b67a1c513c49f..31f1c502f3b61 100644
--- a/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
@@ -13656,21 +13656,17 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-TRUE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-TRUE16-NEXT: .LBB54_1: ; %atomicrmw.start
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, 0
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v7, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v5, v5, v6
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v6.l
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.l
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, v3, v6
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -13693,7 +13689,7 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v3, v0 :: v_dual_lshlrev_b32 v2, 16, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v0
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, -4, v3
@@ -13709,14 +13705,12 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v7, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v5, v5, v2
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, v3, v5
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -14253,21 +14247,17 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine_
; GFX1250-TRUE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-TRUE16-NEXT: .LBB55_1: ; %atomicrmw.start
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, 0
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v7, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v5, v5, v6
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v6.l
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.l
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, v3, v6
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -14292,13 +14282,13 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_bitop2_b32 v3, 3, v4 bitop3:0x40
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-FAKE16-NEXT: .LBB55_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
@@ -14306,14 +14296,12 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v7, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v5, v5, v2
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, v3, v5
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -14868,21 +14856,17 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_neg__amdgpu_no_fine_
; GFX1250-TRUE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-TRUE16-NEXT: .LBB56_1: ; %atomicrmw.start
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, 0
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v7, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v5, v5, v6
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v6.l
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.l
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, v3, v6
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -14906,16 +14890,15 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_neg__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[0:1], lit64(0xfffffffffffff800)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], s[0:1], v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-FAKE16-NEXT: .LBB56_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
@@ -14923,14 +14906,12 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_neg__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v7, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v5, v5, v2
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, v3, v5
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -15487,17 +15468,14 @@ define void @global_agent_atomic_fadd_noret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v7.l
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, 0
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v3, v7
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -15521,7 +15499,7 @@ define void @global_agent_atomic_fadd_noret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v3, v0 :: v_dual_lshlrev_b32 v2, 16, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v0
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v0, -4, v3
@@ -15536,13 +15514,12 @@ define void @global_agent_atomic_fadd_noret_bf16__amdgpu_no_fine_grained_memory(
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v4, v4, v2
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v3, v4
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -16066,17 +16043,14 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine_
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v7.l
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, 0
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v3, v7
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -16102,36 +16076,35 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-FAKE16-NEXT: v_dual_lshlrev_b32 v6, 16, v2 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v4, 3, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: global_load_b32 v3, v[0:1], off
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 3, v4
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v5, v4, 0xffff
-; GFX1250-FAKE16-NEXT: v_not_b32_e32 v5, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
+; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_not_b32_e32 v6, v4
; GFX1250-FAKE16-NEXT: .LBB58_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v4, v3
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v2, v2, v6
+; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
+; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, v4, v2
-; GFX1250-FAKE16-NEXT: v_and_or_b32 v2, v3, v5, v2
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v3, v4
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v2, v[0:1], v[2:3], off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v4, v[0:1], v[4:5], off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: global_inv scope:SCOPE_DEV
-; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3
+; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v5
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v5, v4
; GFX1250-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
@@ -16663,17 +16636,14 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b_neg__amdgpu_no_fine_
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v7.l
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, 0
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v3, v7
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -16698,39 +16668,37 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b_neg__amdgpu_no_fine_
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: s_mov_b64 s[0:1], lit64(0xfffffffffffff800)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v6, 16, v2
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], s[0:1], v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 3, v4
-; GFX1250-FAKE16-NEXT: global_load_b32 v3, v[0:1], off
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 3, v4
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v5, v4, 0xffff
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_not_b32_e32 v5, v5
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
+; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
+; GFX1250-FAKE16-NEXT: v_not_b32_e32 v6, v4
; GFX1250-FAKE16-NEXT: .LBB59_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v4, v3
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v2, v2, v6
+; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
+; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, v4, v2
-; GFX1250-FAKE16-NEXT: v_and_or_b32 v2, v3, v5, v2
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v3, v4
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v2, v[0:1], v[2:3], off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v4, v[0:1], v[4:5], off th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: global_inv scope:SCOPE_DEV
-; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3
+; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v5
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v5, v4
; GFX1250-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
@@ -17254,17 +17222,13 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_pos__align4__amdgpu_
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v5, v3
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v5
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v3
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.h, v3.l
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.l, v4.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v3
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v5, 1, v2 op_sel_hi:[1,0,1]
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v3, v3, s0
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v4.l, v3.l
+; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_xcnt 0x0
@@ -17287,19 +17251,16 @@ define bfloat @global_agent_atomic_fadd_ret_bf16__offset12b_pos__align4__amdgpu_
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: global_load_b32 v3, v[0:1], off offset:2046
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
; GFX1250-FAKE16-NEXT: .LBB60_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v5, v3
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 16, v5
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v3, v3, v2
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v3, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v3, v3, s0
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v3
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -17739,17 +17700,14 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b__align4_pos__amdgpu_
; GFX1250-TRUE16-NEXT: s_mov_b32 s0, 0
; GFX1250-TRUE16-NEXT: .LBB61_1: ; %atomicrmw.start
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.h, v2.l
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v5
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v3
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.h, v3.l
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v3.l, v4.l
-; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v3
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v3, v5, 1, v2 op_sel_hi:[1,0,1]
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v4.h, 0
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v3, v3, s0
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v4.l, v3.l
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
; GFX1250-TRUE16-NEXT: s_wait_xcnt 0x0
@@ -17772,28 +17730,26 @@ define void @global_agent_atomic_fadd_noret_bf16__offset12b__align4_pos__amdgpu_
; GFX1250-FAKE16: ; %bb.0:
; GFX1250-FAKE16-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
-; GFX1250-FAKE16-NEXT: global_load_b32 v3, v[0:1], off offset:2046
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 16, v2
+; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off offset:2046
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
; GFX1250-FAKE16-NEXT: .LBB61_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v3
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v2, v2, v4
-; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v3, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2
-; GFX1250-FAKE16-NEXT: v_and_or_b32 v2, 0xffff0000, v3, v2
+; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v3, v3, s0
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 0xffff, v3
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, 0xffff0000, v5, v3
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v2, v[0:1], v[2:3], off offset:2046 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
+; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v3, v[0:1], v[4:5], off offset:2046 th:TH_ATOMIC_RETURN scope:SCOPE_DEV
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: global_inv scope:SCOPE_DEV
-; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3
+; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v3, v5
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v5, v3
; GFX1250-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
@@ -18218,21 +18174,17 @@ define bfloat @global_system_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine
; GFX1250-TRUE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-TRUE16-NEXT: .LBB62_1: ; %atomicrmw.start
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, 0
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_mov_b32_e32 v7, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v5, v5, v6
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.h, v6.l
-; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v6.l, v5.l
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v5, v3, v6
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -18257,13 +18209,13 @@ define bfloat @global_system_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-FAKE16-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_bitop2_b32 v3, 3, v4 bitop3:0x40
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_not_b32_e32 v4, v4
; GFX1250-FAKE16-NEXT: .LBB62_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
@@ -18271,14 +18223,12 @@ define bfloat @global_system_atomic_fadd_ret_bf16__offset12b_pos__amdgpu_no_fine
; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v7, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v5, v3, v7
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, 16, v5
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v5, v5, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v5, v5, v2
; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v5, v5, s0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_b32_e32 v5, 0xffff, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v5, v3, v5
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-FAKE16-NEXT: v_and_or_b32 v6, v7, v4, v5
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
@@ -18839,17 +18789,14 @@ define void @global_system_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine
; GFX1250-TRUE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-TRUE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-TRUE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, 0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v2.l
-; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, 16, v4
-; GFX1250-TRUE16-NEXT: v_add_f32_e32 v4, v4, v7
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, v7.l
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.h, 0
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
; GFX1250-TRUE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
-; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT: v_mov_b16_e32 v7.l, v4.l
; GFX1250-TRUE16-NEXT: v_lshlrev_b32_e32 v4, v3, v7
+; GFX1250-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-TRUE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-TRUE16-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-TRUE16-NEXT: s_wait_storecnt 0x0
@@ -18875,36 +18822,35 @@ define void @global_system_atomic_fadd_noret_bf16__offset12b_pos__amdgpu_no_fine
; GFX1250-FAKE16-NEXT: s_wait_kmcnt 0x0
; GFX1250-FAKE16-NEXT: v_add_nc_u64_e32 v[4:5], 0x7fe, v[0:1]
; GFX1250-FAKE16-NEXT: s_mov_b32 s0, 0
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX1250-FAKE16-NEXT: v_dual_lshlrev_b32 v6, 16, v2 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v4, 3, v4 bitop3:0x40
-; GFX1250-FAKE16-NEXT: global_load_b32 v3, v[0:1], off
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, 3, v4
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v5, v4, 0xffff
-; GFX1250-FAKE16-NEXT: v_not_b32_e32 v5, v5
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_dual_mov_b32 v1, v5 :: v_dual_bitop2_b32 v0, -4, v4 bitop3:0x40
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v3, 3, v4
+; GFX1250-FAKE16-NEXT: global_load_b32 v5, v[0:1], off
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v3, 3, v3
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e64 v4, v3, 0xffff
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_not_b32_e32 v6, v4
; GFX1250-FAKE16-NEXT: .LBB63_1: ; %atomicrmw.start
; GFX1250-FAKE16-NEXT: ; =>This Inner Loop Header: Depth=1
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
-; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v2, v4, v3
-; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, 16, v2
-; GFX1250-FAKE16-NEXT: v_add_f32_e32 v2, v2, v6
+; GFX1250-FAKE16-NEXT: v_lshrrev_b32_e32 v4, v3, v5
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v2, v2, s0
-; GFX1250-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2
+; GFX1250-FAKE16-NEXT: v_fma_mix_f32_bf16 v4, v4, 1, v2 op_sel_hi:[1,0,1]
+; GFX1250-FAKE16-NEXT: v_cvt_pk_bf16_f32 v4, v4, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v2, v4, v2
-; GFX1250-FAKE16-NEXT: v_and_or_b32 v2, v3, v5, v2
+; GFX1250-FAKE16-NEXT: v_and_b32_e32 v4, 0xffff, v4
+; GFX1250-FAKE16-NEXT: v_lshlrev_b32_e32 v4, v3, v4
+; GFX1250-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT: v_and_or_b32 v4, v5, v6, v4
; GFX1250-FAKE16-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-FAKE16-NEXT: s_wait_storecnt 0x0
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v2, v[0:1], v[2:3], off th:TH_ATOMIC_RETURN scope:SCOPE_SYS
+; GFX1250-FAKE16-NEXT: global_atomic_cmpswap_b32 v4, v[0:1], v[4:5], off th:TH_ATOMIC_RETURN scope:SCOPE_SYS
; GFX1250-FAKE16-NEXT: s_wait_loadcnt 0x0
; GFX1250-FAKE16-NEXT: global_inv scope:SCOPE_SYS
-; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v2, v3
+; GFX1250-FAKE16-NEXT: v_cmp_eq_u32_e32 vcc_lo, v4, v5
; GFX1250-FAKE16-NEXT: s_wait_xcnt 0x0
-; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v3, v2
+; GFX1250-FAKE16-NEXT: v_mov_b32_e32 v5, v4
; GFX1250-FAKE16-NEXT: s_or_b32 s0, vcc_lo, s0
; GFX1250-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX1250-FAKE16-NEXT: s_and_not1_b32 exec_lo, exec_lo, s0
diff --git a/llvm/test/CodeGen/AMDGPU/mad-mix-bf16.ll b/llvm/test/CodeGen/AMDGPU/mad-mix-bf16.ll
index c96ba754c0811..b01cef5231481 100644
--- a/llvm/test/CodeGen/AMDGPU/mad-mix-bf16.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad-mix-bf16.ll
@@ -389,10 +389,9 @@ define float @v_mad_mix_f32_bf16lo_bf16lo_bf16lo_f32_denormals_fmulfadd(bfloat %
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
-; GFX1250-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1
-; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_mul_f32 v0, v0, v1
-; GFX1250-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v2, 1, v0 op_sel_hi:[1,0,0]
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%src0.ext = fpext bfloat %src0 to float
%src1.ext = fpext bfloat %src1 to float
@@ -407,9 +406,8 @@ define float @v_mad_mix_f32_bf16lo_bf16lo_f32_denormals_fmulfadd(bfloat %src0, b
; GFX1250: ; %bb.0:
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_wait_kmcnt 0x0
-; GFX1250-NEXT: v_dual_lshlrev_b32 v0, 16, v0 :: v_dual_lshlrev_b32 v1, 16, v1
-; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX1250-NEXT: v_add_f32_e32 v0, v0, v2
; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%src0.ext = fpext bfloat %src0 to float
@@ -616,6 +614,644 @@ entry:
ret void
}
+define float @v_mad_mix_f32_bf16lo_add_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, v1 op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16hi_add_bf16hi_int(i32 %src0, i32 %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16hi_add_bf16hi_int:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = lshr i32 %src0, 16
+ %src1.hi = lshr i32 %src1, 16
+ %src0.i16 = trunc i32 %src0.hi to i16
+ %src1.i16 = trunc i32 %src1.hi to i16
+ %src0.fp16 = bitcast i16 %src0.i16 to bfloat
+ %src1.fp16 = bitcast i16 %src1.i16 to bfloat
+ %src0.ext = fpext bfloat %src0.fp16 to float
+ %src1.ext = fpext bfloat %src1.fp16 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16hi_and_bf16hi_elt(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16hi_and_bf16hi_elt:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = extractelement <2 x bfloat> %src0, i32 1
+ %src1.hi = extractelement <2 x bfloat> %src1, i32 1
+ %src0.ext = fpext bfloat %src0.hi to float
+ %src1.ext = fpext bfloat %src1.hi to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_cvt_add(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_v2f32_cvt_add:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v3, 0xffff0000, v0
+; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v4, 16, v1
+; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_add_f32 v[0:1], v[2:3], v[4:5]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext <2 x bfloat> %src0 to <2 x float>
+ %src1.ext = fpext <2 x bfloat> %src1 to <2 x float>
+ %result = fadd <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_shuffle_cvt_add(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_lshlrev_b32 v3, 16, v0 :: v_dual_lshlrev_b32 v4, 16, v1
+; GFX1250-NEXT: v_and_b32_e32 v2, 0xffff0000, v0
+; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_add_f32 v[0:1], v[2:3], v[4:5]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.shuf = shufflevector <2 x bfloat> %src0, <2 x bfloat> poison, <2 x i32> <i32 1, i32 0>
+ %src1.shuf = shufflevector <2 x bfloat> %src1, <2 x bfloat> poison, <2 x i32> <i32 0, i32 1>
+ %src0.ext = fpext <2 x bfloat> %src0.shuf to <2 x float>
+ %src1.ext = fpext <2 x bfloat> %src1.shuf to <2 x float>
+ %result = fadd <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define float @v_mad_mix_f32_negbf16lo_add_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negbf16lo_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.neg = fneg float %src0.ext
+ %result = fadd float %src0.ext.neg, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_absbf16lo_add_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_absbf16lo_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %result = fadd float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_negabsbf16lo_add_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negabsbf16lo_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v1, 1, -|v0| op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src0.ext.neg.abs = fneg float %src0.ext.abs
+ %result = fadd float %src0.ext.neg.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_add_f32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_add_f32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, v1 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %result = fadd float %src0.ext, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_add_negf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_add_negf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, -v1 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.neg = fneg float %src1
+ %result = fadd float %src0.ext, %src1.neg
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_add_absf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_add_absf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, |v1| op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %result = fadd float %src0.ext, %src1.abs
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_add_negabsf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_add_negabsf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, -|v1| op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %src1.neg.abs = fneg float %src1.abs
+ %result = fadd float %src0.ext, %src1.neg.abs
+ ret float %result
+}
+
+define float @no_mix_simple_cvt_add(float %src0, float %src1) {
+; GFX1250-LABEL: no_mix_simple_cvt_add:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %result = fadd float %src0, %src1
+ ret float %result
+}
+
+define float @no_mix_simple_fabs_cvt_add(float %src0, float %src1) {
+; GFX1250-LABEL: no_mix_simple_fabs_cvt_add:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.fabs = call float @llvm.fabs.f32(float %src0)
+ %result = fadd float %src0.fabs, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_clamp_f32_bf16hi_add_bf16hi_elt(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_clamp_f32_bf16hi_add_bf16hi_elt:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1] clamp
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = extractelement <2 x bfloat> %src0, i32 1
+ %src1.hi = extractelement <2 x bfloat> %src1, i32 1
+ %src0.ext = fpext bfloat %src0.hi to float
+ %src1.ext = fpext bfloat %src1.hi to float
+ %result = fadd float %src0.ext, %src1.ext
+ %max = call float @llvm.maxnum.f32(float %result, float 0.0)
+ %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
+ ret float %clamp
+}
+
+define float @v_mad_mix_f32_negprecvtbf16lo_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negprecvtbf16lo_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 0
+ %src0.neg = fneg bfloat %src0
+ %src0.ext = fpext bfloat %src0.neg to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtnegbf16hi_abs_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_precvtnegbf16hi_abs_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 1
+ %src0.neg = fneg bfloat %src0
+ %src0.ext = fpext bfloat %src0.neg to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtabsbf16hi_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_precvtabsbf16hi_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 1
+ %src0.abs = call bfloat @llvm.fabs.bf16(bfloat %src0)
+ %src0.ext = fpext bfloat %src0.abs to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfneg_bf16hi_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfneg_bf16hi_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -v0, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fneg = fneg <2 x bfloat> %src0.arg.bc
+ %src0 = extractelement <2 x bfloat> %fneg, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabs_bf16hi_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfabs_bf16hi_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fabs = call <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat> %src0.arg.bc)
+ %src0 = extractelement <2 x bfloat> %fabs, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabsfneg_bf16hi_add_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfabsfneg_bf16hi_add_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -|v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fabs = call <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat> %src0.arg.bc)
+ %fneg.fabs = fneg <2 x bfloat> %fabs
+ %src0 = extractelement <2 x bfloat> %fneg.fabs, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_mul_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16hi_mul_bf16hi_int(i32 %src0, i32 %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16hi_mul_bf16hi_int:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = lshr i32 %src0, 16
+ %src1.hi = lshr i32 %src1, 16
+ %src0.i16 = trunc i32 %src0.hi to i16
+ %src1.i16 = trunc i32 %src1.hi to i16
+ %src0.fp16 = bitcast i16 %src0.i16 to bfloat
+ %src1.fp16 = bitcast i16 %src1.i16 to bfloat
+ %src0.ext = fpext bfloat %src0.fp16 to float
+ %src1.ext = fpext bfloat %src1.fp16 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16hi_mul_bf16hi_elt(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16hi_mul_bf16hi_elt:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = extractelement <2 x bfloat> %src0, i32 1
+ %src1.hi = extractelement <2 x bfloat> %src1, i32 1
+ %src0.ext = fpext bfloat %src0.hi to float
+ %src1.ext = fpext bfloat %src1.hi to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_cvt_mul(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v3, 0xffff0000, v0
+; GFX1250-NEXT: v_dual_lshlrev_b32 v2, 16, v0 :: v_dual_lshlrev_b32 v4, 16, v1
+; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_f32 v[0:1], v[2:3], v[4:5]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext <2 x bfloat> %src0 to <2 x float>
+ %src1.ext = fpext <2 x bfloat> %src1 to <2 x float>
+ %result = fmul <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_shuffle_cvt_mul(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_lshlrev_b32 v3, 16, v0 :: v_dual_lshlrev_b32 v4, 16, v1
+; GFX1250-NEXT: v_and_b32_e32 v2, 0xffff0000, v0
+; GFX1250-NEXT: v_and_b32_e32 v5, 0xffff0000, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_f32 v[0:1], v[2:3], v[4:5]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.shuf = shufflevector <2 x bfloat> %src0, <2 x bfloat> poison, <2 x i32> <i32 1, i32 0>
+ %src1.shuf = shufflevector <2 x bfloat> %src1, <2 x bfloat> poison, <2 x i32> <i32 0, i32 1>
+ %src0.ext = fpext <2 x bfloat> %src0.shuf to <2 x float>
+ %src1.ext = fpext <2 x bfloat> %src1.shuf to <2 x float>
+ %result = fmul <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define float @v_mad_mix_f32_negbf16lo_mul_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negbf16lo_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.neg = fneg float %src0.ext
+ %result = fmul float %src0.ext.neg, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_absbf16lo_mul_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_absbf16lo_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %result = fmul float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_negabsbf16lo_mul_bf16lo(bfloat %src0, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negabsbf16lo_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -|v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src0.ext.neg.abs = fneg float %src0.ext.abs
+ %result = fmul float %src0.ext.neg.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_mul_f32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_mul_f32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %result = fmul float %src0.ext, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_mul_negf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_mul_negf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, -v1, 0 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.neg = fneg float %src1
+ %result = fmul float %src0.ext, %src1.neg
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_mul_absf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_mul_absf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, |v1|, 0 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %result = fmul float %src0.ext, %src1.abs
+ ret float %result
+}
+
+define float @v_mad_mix_f32_bf16lo_mul_negabsf32(bfloat %src0, float %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_bf16lo_mul_negabsf32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, -|v1|, 0 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %src1.neg.abs = fneg float %src1.abs
+ %result = fmul float %src0.ext, %src1.neg.abs
+ ret float %result
+}
+
+define float @no_mix_simple_cvt_mul(float %src0, float %src1) {
+; GFX1250-LABEL: no_mix_simple_cvt_mul:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %result = fmul float %src0, %src1
+ ret float %result
+}
+
+define float @no_mix_simple_fabs_cvt_mul(float %src0, float %src1) {
+; GFX1250-LABEL: no_mix_simple_fabs_cvt_mul:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.fabs = call float @llvm.fabs.f32(float %src0)
+ %result = fmul float %src0.fabs, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_clamp_f32_bf16hi_mul_bf16hi_elt(<2 x bfloat> %src0, <2 x bfloat> %src1) {
+; GFX1250-LABEL: v_mad_mix_clamp_f32_bf16hi_mul_bf16hi_elt:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0] clamp
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.hi = extractelement <2 x bfloat> %src0, i32 1
+ %src1.hi = extractelement <2 x bfloat> %src1, i32 1
+ %src0.ext = fpext bfloat %src0.hi to float
+ %src1.ext = fpext bfloat %src1.hi to float
+ %result = fmul float %src0.ext, %src1.ext
+ %max = call float @llvm.maxnum.f32(float %result, float 0.0)
+ %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
+ ret float %clamp
+}
+
+define float @v_mad_mix_f32_negprecvtbf16lo_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_negprecvtbf16lo_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 0
+ %src0.neg = fneg bfloat %src0
+ %src0.ext = fpext bfloat %src0.neg to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtnegbf16hi_abs_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_precvtnegbf16hi_abs_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 1
+ %src0.neg = fneg bfloat %src0
+ %src0.ext = fpext bfloat %src0.neg to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtabsbf16hi_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_precvtabsbf16hi_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %src0 = extractelement <2 x bfloat> %src0.arg.bc, i32 1
+ %src0.abs = call bfloat @llvm.fabs.bf16(bfloat %src0)
+ %src0.ext = fpext bfloat %src0.abs to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfneg_bf16hi_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfneg_bf16hi_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -v0, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fneg = fneg <2 x bfloat> %src0.arg.bc
+ %src0 = extractelement <2 x bfloat> %fneg, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabs_bf16hi_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfabs_bf16hi_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fabs = call <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat> %src0.arg.bc)
+ %src0 = extractelement <2 x bfloat> %fabs, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabsfneg_bf16hi_mul_bf16lo(i32 %src0.arg, bfloat %src1) {
+; GFX1250-LABEL: v_mad_mix_f32_preextractfabsfneg_bf16hi_mul_bf16lo:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_fma_mix_f32_bf16 v0, -|v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x bfloat>
+ %fabs = call <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat> %src0.arg.bc)
+ %fneg.fabs = fneg <2 x bfloat> %fabs
+ %src0 = extractelement <2 x bfloat> %fneg.fabs, i32 1
+ %src0.ext = fpext bfloat %src0 to float
+ %src1.ext = fpext bfloat %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
declare bfloat @llvm.fabs.bf16(bfloat) #2
declare <2 x bfloat> @llvm.fabs.v2bf16(<2 x bfloat>) #2
declare float @llvm.fabs.f32(float) #2
diff --git a/llvm/test/CodeGen/AMDGPU/mad-mix.ll b/llvm/test/CodeGen/AMDGPU/mad-mix.ll
index a4878539b1c74..5eb1a4649ff98 100644
--- a/llvm/test/CodeGen/AMDGPU/mad-mix.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad-mix.ll
@@ -1866,27 +1866,13 @@ define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals(half %src0, half %src1, fl
}
define float @v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd(half %src0, half %src1, half %src2) #1 {
-; SDAG-GFX1100-TRUE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
-; SDAG-GFX1100-TRUE16: ; %bb.0:
-; SDAG-GFX1100-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; SDAG-GFX1100-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
-; SDAG-GFX1100-TRUE16-NEXT: v_cvt_f32_f16_e32 v2, v2.l
-; SDAG-GFX1100-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-TRUE16-NEXT: v_mul_f32_e32 v0, v0, v1
-; SDAG-GFX1100-TRUE16-NEXT: v_add_f32_e32 v0, v0, v2
-; SDAG-GFX1100-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; SDAG-GFX1100-FAKE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
-; SDAG-GFX1100-FAKE16: ; %bb.0:
-; SDAG-GFX1100-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; SDAG-GFX1100-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
-; SDAG-GFX1100-FAKE16-NEXT: v_cvt_f32_f16_e32 v2, v2
-; SDAG-GFX1100-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-FAKE16-NEXT: v_mul_f32_e32 v0, v0, v1
-; SDAG-GFX1100-FAKE16-NEXT: v_add_f32_e32 v0, v0, v2
-; SDAG-GFX1100-FAKE16-NEXT: s_setpc_b64 s[30:31]
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v2, 1, v0 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
;
; GFX900-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
; GFX900: ; %bb.0:
@@ -1901,11 +1887,8 @@ define float @v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd(half %src0,
; GFX906-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
; GFX906: ; %bb.0:
; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX906-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GFX906-NEXT: v_cvt_f32_f16_e32 v2, v2
-; GFX906-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX906-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: v_fma_mix_f32 v0, v2, 1, v0 op_sel_hi:[1,0,0]
; GFX906-NEXT: s_setpc_b64 s[30:31]
;
; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
@@ -1935,17 +1918,6 @@ define float @v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd(half %src0,
; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v2
; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
;
-; GISEL-GFX1100-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
-; GISEL-GFX1100: ; %bb.0:
-; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v2, v2
-; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX1100-NEXT: v_mul_f32_e32 v0, v0, v1
-; GISEL-GFX1100-NEXT: v_add_f32_e32 v0, v0, v2
-; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
-;
; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd:
; GISEL-CI: ; %bb.0:
; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -1964,25 +1936,13 @@ define float @v_mad_mix_f32_f16lo_f16lo_f16lo_f32_denormals_fmulfadd(half %src0,
}
define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd(half %src0, half %src1, float %src2) #1 {
-; SDAG-GFX1100-TRUE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
-; SDAG-GFX1100-TRUE16: ; %bb.0:
-; SDAG-GFX1100-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-TRUE16-NEXT: v_cvt_f32_f16_e32 v0, v0.l
-; SDAG-GFX1100-TRUE16-NEXT: v_cvt_f32_f16_e32 v1, v1.l
-; SDAG-GFX1100-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-TRUE16-NEXT: v_mul_f32_e32 v0, v0, v1
-; SDAG-GFX1100-TRUE16-NEXT: v_add_f32_e32 v0, v0, v2
-; SDAG-GFX1100-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; SDAG-GFX1100-FAKE16-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
-; SDAG-GFX1100-FAKE16: ; %bb.0:
-; SDAG-GFX1100-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SDAG-GFX1100-FAKE16-NEXT: v_cvt_f32_f16_e32 v0, v0
-; SDAG-GFX1100-FAKE16-NEXT: v_cvt_f32_f16_e32 v1, v1
-; SDAG-GFX1100-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; SDAG-GFX1100-FAKE16-NEXT: v_mul_f32_e32 v0, v0, v1
-; SDAG-GFX1100-FAKE16-NEXT: v_add_f32_e32 v0, v0, v2
-; SDAG-GFX1100-FAKE16-NEXT: s_setpc_b64 s[30:31]
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1100-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
;
; GFX900-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
; GFX900: ; %bb.0:
@@ -1996,9 +1956,7 @@ define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd(half %src0, half
; GFX906-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
; GFX906: ; %bb.0:
; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX906-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GFX906-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GFX906-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
; GFX906-NEXT: v_add_f32_e32 v0, v0, v2
; GFX906-NEXT: s_setpc_b64 s[30:31]
;
@@ -2027,16 +1985,6 @@ define float @v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd(half %src0, half
; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v2
; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
;
-; GISEL-GFX1100-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
-; GISEL-GFX1100: ; %bb.0:
-; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v0, v0
-; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v1, v1
-; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GISEL-GFX1100-NEXT: v_mul_f32_e32 v0, v0, v1
-; GISEL-GFX1100-NEXT: v_add_f32_e32 v0, v0, v2
-; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
-;
; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_f16lo_f32_denormals_fmulfadd:
; GISEL-CI: ; %bb.0:
; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -2658,6 +2606,2915 @@ define float @v_mad_mix_f32_preextractfabsfneg_f16hi_f16lo_f16lo(i32 %src0.arg,
ret float %result
}
+define float @v_mad_mix_f32_f16lo_add_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16hi_add_f16hi(i32 %src0, i32 %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_mad_mix_f32_f16hi_add_f16hi:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_add_f32_e32 v0, v0, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = lshr i32 %src0, 16
+ %src1.hi = lshr i32 %src1, 16
+ %src0.i16 = trunc i32 %src0.hi to i16
+ %src1.i16 = trunc i32 %src1.hi to i16
+ %src0.fp16 = bitcast i16 %src0.i16 to half
+ %src1.fp16 = bitcast i16 %src1.i16 to half
+ %src0.ext = fpext half %src0.fp16 to float
+ %src1.ext = fpext half %src1.fp16 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16hi_add_f16hi_elt(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v1, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16hi_add_f16hi_elt:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v3
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = extractelement <2 x half> %src0, i32 1
+ %src1.hi = extractelement <2 x half> %src1, i32 1
+ %src0.ext = fpext half %src0.hi to float
+ %src1.ext = fpext half %src1.hi to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_cvt_add(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_v2f32_cvt_add:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v2, v0, 1, v1 op_sel_hi:[1,0,1]
+; GFX1100-NEXT: v_fma_mix_f32 v1, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1100-NEXT: v_mov_b32_e32 v0, v2
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_v2f32_cvt_add:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-GFX900-NEXT: v_add_f32_e32 v1, v2, v3
+; SDAG-GFX900-NEXT: v_add_f32_e32 v0, v0, v4
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_v2f32_cvt_add:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v2, v0, 1, v1 op_sel_hi:[1,0,1]
+; GFX906-NEXT: v_fma_mix_f32 v1, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1]
+; GFX906-NEXT: v_mov_b32_e32 v0, v2
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_v2f32_cvt_add:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-GFX9GEN-NEXT: v_add_f32_e32 v1, v2, v3
+; SDAG-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v4
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_v2f32_cvt_add:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-VI-NEXT: v_add_f32_e32 v1, v2, v3
+; SDAG-VI-NEXT: v_add_f32_e32 v0, v0, v4
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_v2f32_cvt_add:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v2
+; SDAG-CI-NEXT: v_add_f32_e32 v1, v1, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_v2f32_cvt_add:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v2, v0
+; GISEL-GFX900-NEXT: v_add_f32_e32 v1, v3, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_v2f32_cvt_add:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v2, v0
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v1, v3, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_v2f32_cvt_add:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v2, v0
+; GISEL-VI-NEXT: v_add_f32_e32 v1, v3, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_v2f32_cvt_add:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v2
+; GISEL-CI-NEXT: v_add_f32_e32 v1, v1, v3
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext <2 x half> %src0 to <2 x float>
+ %src1.ext = fpext <2 x half> %src1 to <2 x float>
+ %result = fadd <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_shuffle_cvt_add(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v2, v0, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: v_fma_mix_f32 v1, v0, 1, v1 op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1100-NEXT: v_mov_b32_e32 v0, v2
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v3, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_add_f32_e32 v0, v2, v0
+; GFX900-NEXT: v_add_f32_e32 v1, v3, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v2, v0, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX906-NEXT: v_fma_mix_f32 v1, v0, 1, v1 op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; GFX906-NEXT: v_mov_b32_e32 v0, v2
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v3, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v2, v0
+; GFX9GEN-NEXT: v_add_f32_e32 v1, v3, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v3, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_add_f32_e32 v0, v2, v0
+; VI-NEXT: v_add_f32_e32 v1, v3, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v4, v0
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v1, v2
+; SDAG-CI-NEXT: v_add_f32_e32 v1, v4, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_v2f32_shuffle_cvt_add:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v4, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v2
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v3
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v1, v0
+; GISEL-CI-NEXT: v_add_f32_e32 v1, v4, v2
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.shuf = shufflevector <2 x half> %src0, <2 x half> poison, <2 x i32> <i32 1, i32 0>
+ %src1.shuf = shufflevector <2 x half> %src1, <2 x half> poison, <2 x i32> <i32 0, i32 1>
+ %src0.ext = fpext <2 x half> %src0.shuf to <2 x float>
+ %src1.ext = fpext <2 x half> %src1.shuf to <2 x float>
+ %result = fadd <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define float @v_mad_mix_f32_negf16lo_add_f16lo(half %src0, half %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[1,0,1]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[1,0,1]
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negf16lo_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.neg = fneg float %src0.ext
+ %result = fadd float %src0.ext.neg, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_absf16lo_add_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_absf16lo_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %result = fadd float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_negabsf16lo_add_f16lo(half %src0, half %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v1, 1, -|v0| op_sel_hi:[1,0,1]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v1, 1, -|v0| op_sel_hi:[1,0,1]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GISEL-GFX1100-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX906-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX906-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negabsf16lo_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src0.ext.neg.abs = fneg float %src0.ext.abs
+ %result = fadd float %src0.ext.neg.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_add_f32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_add_f32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %result = fadd float %src0.ext, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_add_negf32(half %src0, float %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, -v1 op_sel_hi:[1,0,0]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, -v1 op_sel_hi:[1,0,0]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_sub_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_sub_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GISEL-GFX1100-NEXT: v_sub_f32_e32 v0, v0, v1
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX906-NEXT: v_sub_f32_e32 v0, v0, v1
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_add_negf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_sub_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.neg = fneg float %src1
+ %result = fadd float %src0.ext, %src1.neg
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_add_absf32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, |v1| op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, |v1| op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_add_f32_e64 v0, v0, |v1|
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e64 v0, v0, |v1|
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_add_absf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %result = fadd float %src0.ext, %src1.abs
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_add_negabsf32(half %src0, float %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, -|v1| op_sel_hi:[1,0,0]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, -|v1| op_sel_hi:[1,0,0]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GISEL-GFX1100-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-GFX906-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_add_negabsf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %src1.neg.abs = fneg float %src1.abs
+ %result = fadd float %src0.ext, %src1.neg.abs
+ ret float %result
+}
+
+define float @no_mix_simple_cvt_add(float %src0, float %src1) {
+; GFX1100-LABEL: no_mix_simple_cvt_add:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: no_mix_simple_cvt_add:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: no_mix_simple_cvt_add:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: no_mix_simple_cvt_add:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: no_mix_simple_cvt_add:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: no_mix_simple_cvt_add:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_add_f32_e32 v0, v0, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %result = fadd float %src0, %src1
+ ret float %result
+}
+
+define float @no_mix_simple_fabs_cvt_add(float %src0, float %src1) {
+; GFX1100-LABEL: no_mix_simple_fabs_cvt_add:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: no_mix_simple_fabs_cvt_add:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: no_mix_simple_fabs_cvt_add:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: no_mix_simple_fabs_cvt_add:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: no_mix_simple_fabs_cvt_add:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: no_mix_simple_fabs_cvt_add:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.fabs = call float @llvm.fabs.f32(float %src0)
+ %result = fadd float %src0.fabs, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_clamp_f32_f16hi_add_f16hi(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1] clamp
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_add_f32_e64 v0, v0, v1 clamp
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, 1, v1 op_sel:[1,0,1] op_sel_hi:[1,0,1] clamp
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_add_f32_e64 v0, v0, v1 clamp
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_add_f32_e64 v0, v0, v1 clamp
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_add_f32_e64 v0, v1, v3 clamp
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_clamp_f32_f16hi_add_f16hi:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v3
+; GISEL-CI-NEXT: v_add_f32_e64 v0, v0, v1 clamp
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = extractelement <2 x half> %src0, i32 1
+ %src1.hi = extractelement <2 x half> %src1, i32 1
+ %src0.ext = fpext half %src0.hi to float
+ %src1.ext = fpext half %src1.hi to float
+ %result = fadd float %src0.ext, %src1.ext
+ %max = call float @llvm.maxnum.f32(float %result, float 0.0)
+ %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
+ ret float %clamp
+}
+
+define float @v_mad_mix_f32_negprecvtf16lo_add_f16lo(i32 %src0.arg, half %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel_hi:[1,0,1]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-CI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[1,0,1]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel_hi:[1,0,1]
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negprecvtf16lo_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 0
+ %src0.neg = fneg half %src0
+ %src0.ext = fpext half %src0.neg to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo(i32 %src0.arg, half %src1) {
+; SDAG-GFX1100-TRUE16-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; SDAG-GFX1100-TRUE16: ; %bb.0:
+; SDAG-GFX1100-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v0.h
+; SDAG-GFX1100-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; SDAG-GFX1100-TRUE16-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; SDAG-GFX1100-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX1100-FAKE16-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; SDAG-GFX1100-FAKE16: ; %bb.0:
+; SDAG-GFX1100-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-GFX1100-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; SDAG-GFX1100-FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; SDAG-GFX1100-FAKE16-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; SDAG-GFX1100-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: s_mov_b32 s4, 0x8000
+; GFX906-NEXT: v_xor_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GISEL-GFX1100-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel_hi:[1,0,1]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_precvtnegf16hi_abs_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 1
+ %src0.neg = fneg half %src0
+ %src0.ext = fpext half %src0.neg to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtabsf16hi_add_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_add_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_precvtabsf16hi_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 1
+ %src0.abs = call half @llvm.fabs.f16(half %src0)
+ %src0.ext = fpext half %src0.abs to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfneg_f16hi_add_f16lo(i32 %src0.arg, half %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v1, 1, -v0 op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_fma_mix_f32 v0, -v0, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fneg = fneg <2 x half> %src0.arg.bc
+ %src0 = extractelement <2 x half> %fneg, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabs_f16hi_add_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_preextractfabs_f16hi_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %src0.arg.bc)
+ %src0 = extractelement <2 x half> %fabs, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo(i32 %src0.arg, half %src1) {
+; SDAG-GFX1100-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-GFX1100: ; %bb.0:
+; SDAG-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-NEXT: v_fma_mix_f32 v0, v1, 1, -|v0| op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; SDAG-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX906-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-GFX906: ; %bb.0:
+; SDAG-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX906-NEXT: v_fma_mix_f32 v0, v1, 1, -|v0| op_sel:[0,0,1] op_sel_hi:[1,0,1]
+; SDAG-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_sub_f32_e32 v0, v1, v0
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
+; SDAG-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, -|v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX906-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-GFX906: ; %bb.0:
+; GISEL-GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX906-NEXT: v_fma_mix_f32 v0, -|v0|, 1, v1 op_sel:[1,0,0] op_sel_hi:[1,0,1]
+; GISEL-GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_preextractfabsfneg_f16hi_add_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_add_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %src0.arg.bc)
+ %fneg.fabs = fneg <2 x half> %fabs
+ %src0 = extractelement <2 x half> %fneg.fabs, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fadd float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_mul_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16hi_mul_f16hi_int(i32 %src0, i32 %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_int:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = lshr i32 %src0, 16
+ %src1.hi = lshr i32 %src1, 16
+ %src0.i16 = trunc i32 %src0.hi to i16
+ %src1.i16 = trunc i32 %src1.hi to i16
+ %src0.fp16 = bitcast i16 %src0.i16 to half
+ %src1.fp16 = bitcast i16 %src1.i16 to half
+ %src0.ext = fpext half %src0.fp16 to float
+ %src1.ext = fpext half %src1.fp16 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16hi_mul_f16hi_elt(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v1, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16hi_mul_f16hi_elt:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v3
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = extractelement <2 x half> %src0, i32 1
+ %src1.hi = extractelement <2 x half> %src1, i32 1
+ %src0.ext = fpext half %src0.hi to float
+ %src1.ext = fpext half %src1.hi to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_cvt_mul(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v2, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: v_fma_mix_f32 v1, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1100-NEXT: v_mov_b32_e32 v0, v2
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_v2f32_cvt_mul:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v1, v2, v3
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v0, v0, v4
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v2, v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: v_fma_mix_f32 v1, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: v_mov_b32_e32 v0, v2
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_v2f32_cvt_mul:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v1, v2, v3
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v4
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_v2f32_cvt_mul:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v3, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v4, v1
+; SDAG-VI-NEXT: v_mul_f32_e32 v1, v2, v3
+; SDAG-VI-NEXT: v_mul_f32_e32 v0, v0, v4
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_v2f32_cvt_mul:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v2
+; SDAG-CI-NEXT: v_mul_f32_e32 v1, v1, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v0, v2, v0
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v1, v3, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v0, v2, v0
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v1, v3, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v2, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_mul_f32_e32 v0, v2, v0
+; GISEL-VI-NEXT: v_mul_f32_e32 v1, v3, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_v2f32_cvt_mul:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v2
+; GISEL-CI-NEXT: v_mul_f32_e32 v1, v1, v3
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext <2 x half> %src0 to <2 x float>
+ %src1.ext = fpext <2 x half> %src1 to <2 x float>
+ %result = fmul <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define <2 x float> @v_mad_mix_v2f32_shuffle_cvt_mul(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v2, v0, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: v_fma_mix_f32 v1, v0, v1, 0 op_sel:[0,1,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX1100-NEXT: v_mov_b32_e32 v0, v2
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v3, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_mul_f32_e32 v0, v2, v0
+; GFX900-NEXT: v_mul_f32_e32 v1, v3, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v2, v0, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: v_fma_mix_f32 v1, v0, v1, 0 op_sel:[0,1,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: v_mov_b32_e32 v0, v2
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v3, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v2, v0
+; GFX9GEN-NEXT: v_mul_f32_e32 v1, v3, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v3, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_mul_f32_e32 v0, v2, v0
+; VI-NEXT: v_mul_f32_e32 v1, v3, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-CI-NEXT: v_cvt_f32_f16_e32 v4, v0
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v1, v2
+; SDAG-CI-NEXT: v_mul_f32_e32 v1, v4, v3
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_v2f32_shuffle_cvt_mul:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v4, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v2
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v3
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v1, v0
+; GISEL-CI-NEXT: v_mul_f32_e32 v1, v4, v2
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.shuf = shufflevector <2 x half> %src0, <2 x half> poison, <2 x i32> <i32 1, i32 0>
+ %src1.shuf = shufflevector <2 x half> %src1, <2 x half> poison, <2 x i32> <i32 0, i32 1>
+ %src0.ext = fpext <2 x half> %src0.shuf to <2 x float>
+ %src1.ext = fpext <2 x half> %src1.shuf to <2 x float>
+ %result = fmul <2 x float> %src0.ext, %src1.ext
+ ret <2 x float> %result
+}
+
+define float @v_mad_mix_f32_negf16lo_mul_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, -v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negf16lo_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.neg = fneg float %src0.ext
+ %result = fmul float %src0.ext.neg, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_absf16lo_mul_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_absf16lo_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %result = fmul float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_negabsf16lo_mul_f16lo(half %src0, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, -|v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, -|v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negabsf16lo_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src0.ext.neg.abs = fneg float %src0.ext.abs
+ %result = fmul float %src0.ext.neg.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_mul_f32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_mul_f32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %result = fmul float %src0.ext, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_mul_negf32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, -v1, 0 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_mul_f32_e64 v0, v0, -v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, -v1, 0 op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, v0, -v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_mul_f32_e64 v0, v0, -v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, v0, -v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_mul_negf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, v0, -v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.neg = fneg float %src1
+ %result = fmul float %src0.ext, %src1.neg
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_mul_absf32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, |v1|, 0 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, |v1|, 0 op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_mul_absf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %result = fmul float %src0.ext, %src1.abs
+ ret float %result
+}
+
+define float @v_mad_mix_f32_f16lo_mul_negabsf32(half %src0, float %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, -|v1|, 0 op_sel_hi:[1,0,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX900-NEXT: v_mul_f32_e64 v0, v0, -|v1|
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, -|v1|, 0 op_sel_hi:[1,0,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, v0, -|v1|
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; VI-NEXT: v_mul_f32_e64 v0, v0, -|v1|
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, v0, -|v1|
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_f16lo_mul_negabsf32:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, v0, -|v1|
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.ext = fpext half %src0 to float
+ %src1.abs = call float @llvm.fabs.f32(float %src1)
+ %src1.neg.abs = fneg float %src1.abs
+ %result = fmul float %src0.ext, %src1.neg.abs
+ ret float %result
+}
+
+define float @no_mix_simple_cvt_mul(float %src0, float %src1) {
+; GFX1100-LABEL: no_mix_simple_cvt_mul:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: no_mix_simple_cvt_mul:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: no_mix_simple_cvt_mul:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: no_mix_simple_cvt_mul:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: no_mix_simple_cvt_mul:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: no_mix_simple_cvt_mul:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %result = fmul float %src0, %src1
+ ret float %result
+}
+
+define float @no_mix_simple_fabs_cvt_mul(float %src0, float %src1) {
+; GFX1100-LABEL: no_mix_simple_fabs_cvt_mul:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: no_mix_simple_fabs_cvt_mul:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: no_mix_simple_fabs_cvt_mul:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: no_mix_simple_fabs_cvt_mul:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: no_mix_simple_fabs_cvt_mul:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: no_mix_simple_fabs_cvt_mul:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.fabs = call float @llvm.fabs.f32(float %src0)
+ %result = fmul float %src0.fabs, %src1
+ ret float %result
+}
+
+define float @v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt(<2 x half> %src0, <2 x half> %src1) {
+; GFX1100-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0] clamp
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_mul_f32_e64 v0, v0, v1 clamp
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, v0, v1, 0 op_sel:[1,1,0] op_sel_hi:[1,1,0] clamp
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, v0, v1 clamp
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_sdwa v1, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_mul_f32_e64 v0, v0, v1 clamp
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_mul_f32_e64 v0, v1, v3 clamp
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_clamp_f32_f16hi_mul_f16hi_elt:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v1
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v3
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, v0, v1 clamp
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.hi = extractelement <2 x half> %src0, i32 1
+ %src1.hi = extractelement <2 x half> %src1, i32 1
+ %src0.ext = fpext half %src0.hi to float
+ %src1.ext = fpext half %src1.hi to float
+ %result = fmul float %src0.ext, %src1.ext
+ %max = call float @llvm.maxnum.f32(float %result, float 0.0)
+ %clamp = call float @llvm.minnum.f32(float %max, float 1.0)
+ ret float %clamp
+}
+
+define float @v_mad_mix_f32_negprecvtf16lo_mul_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_negprecvtf16lo_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 0
+ %src0.neg = fneg half %src0
+ %src0.ext = fpext half %src0.neg to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo(i32 %src0.arg, half %src1) {
+; SDAG-GFX1100-TRUE16-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; SDAG-GFX1100-TRUE16: ; %bb.0:
+; SDAG-GFX1100-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-TRUE16-NEXT: v_xor_b16 v0.l, 0x8000, v0.h
+; SDAG-GFX1100-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; SDAG-GFX1100-TRUE16-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; SDAG-GFX1100-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX1100-FAKE16-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; SDAG-GFX1100-FAKE16: ; %bb.0:
+; SDAG-GFX1100-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX1100-FAKE16-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-GFX1100-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; SDAG-GFX1100-FAKE16-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; SDAG-GFX1100-FAKE16-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; SDAG-GFX1100-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: s_mov_b32 s4, 0x8000
+; GFX906-NEXT: v_xor_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX1100-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; GISEL-GFX1100: ; %bb.0:
+; GISEL-GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX1100-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-GFX1100-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GISEL-GFX1100-NEXT: v_xor_b32_e32 v0, 0x8000, v0
+; GISEL-GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel_hi:[1,1,0]
+; GISEL-GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_fma_mix_f32_precvtnegf16hi_abs_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 1
+ %src0.neg = fneg half %src0
+ %src0.ext = fpext half %src0.neg to float
+ %src0.ext.abs = call float @llvm.fabs.f32(float %src0.ext)
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext.abs, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_precvtabsf16hi_mul_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX900-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; GFX900: ; %bb.0:
+; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX900-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9GEN-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; GFX9GEN: ; %bb.0:
+; GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; VI-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_precvtabsf16hi_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %src0 = extractelement <2 x half> %src0.arg.bc, i32 1
+ %src0.abs = call half @llvm.fabs.f16(half %src0)
+ %src0.ext = fpext half %src0.abs to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, -v0, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, -v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_mad_mix_f32_preextractfneg_f16hi_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fneg = fneg <2 x half> %src0.arg.bc
+ %src0 = extractelement <2 x half> %fneg, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, |v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, |v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_fma_mix_f32_preextractfabs_f16hi_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %src0.arg.bc)
+ %src0 = extractelement <2 x half> %fabs, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+define float @v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo(i32 %src0.arg, half %src1) {
+; GFX1100-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GFX1100: ; %bb.0:
+; GFX1100-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1100-NEXT: v_fma_mix_f32 v0, -|v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX1100-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX900-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; SDAG-GFX900: ; %bb.0:
+; SDAG-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX906-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GFX906: ; %bb.0:
+; GFX906-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX906-NEXT: v_fma_mix_f32 v0, -|v0|, v1, 0 op_sel:[1,0,0] op_sel_hi:[1,1,0]
+; GFX906-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-GFX9GEN-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; SDAG-GFX9GEN: ; %bb.0:
+; SDAG-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-VI-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; SDAG-VI: ; %bb.0:
+; SDAG-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-VI-NEXT: v_cvt_f32_f16_sdwa v0, -|v0| dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; SDAG-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; SDAG-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; SDAG-CI-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; SDAG-CI: ; %bb.0:
+; SDAG-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; SDAG-CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
+; SDAG-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; SDAG-CI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX900-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GISEL-GFX900: ; %bb.0:
+; GISEL-GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX900-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX900-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX900-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX900-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-GFX9GEN-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GISEL-GFX9GEN: ; %bb.0:
+; GISEL-GFX9GEN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-GFX9GEN-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-GFX9GEN-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-GFX9GEN-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-GFX9GEN-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-VI-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GISEL-VI: ; %bb.0:
+; GISEL-VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-VI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-VI-NEXT: v_cvt_f32_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GISEL-VI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-VI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-CI-LABEL: v_fma_mix_f32_preextractfabsfneg_f16hi_mul_f16lo:
+; GISEL-CI: ; %bb.0:
+; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-CI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
+; GISEL-CI-NEXT: v_lshrrev_b32_e32 v0, 16, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; GISEL-CI-NEXT: v_mul_f32_e32 v0, v0, v1
+; GISEL-CI-NEXT: s_setpc_b64 s[30:31]
+ %src0.arg.bc = bitcast i32 %src0.arg to <2 x half>
+ %fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %src0.arg.bc)
+ %fneg.fabs = fneg <2 x half> %fabs
+ %src0 = extractelement <2 x half> %fneg.fabs, i32 1
+ %src0.ext = fpext half %src0 to float
+ %src1.ext = fpext half %src1 to float
+ %result = fmul float %src0.ext, %src1.ext
+ ret float %result
+}
+
+
declare half @llvm.fabs.f16(half) #2
declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #2
declare float @llvm.fabs.f32(float) #2
More information about the llvm-commits
mailing list