[llvm] 33d401f - [AMDGPU][True16][CodeGen] true16 codegen for icmp and is_fpclass (#124757)

Thu Jan 30 09:18:04 PST 2025

Author: Brox Chen
Date: 2025-01-30T12:18:00-05:00
New Revision: 33d401fb15148f212990b94775c5f59e0b72bb92

URL: https://github.com/llvm/llvm-project/commit/33d401fb15148f212990b94775c5f59e0b72bb92
DIFF: https://github.com/llvm/llvm-project/commit/33d401fb15148f212990b94775c5f59e0b72bb92.diff

LOG: [AMDGPU][True16][CodeGen] true16 codegen for icmp and is_fpclass (#124757)

True16 codegen pattern for icmp patterns and is_fpclass

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
    llvm/lib/Target/AMDGPU/VOPCInstructions.td
    llvm/lib/Target/AMDGPU/VOPInstructions.td
    llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w32.mir
    llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w64.mir
    llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
    llvm/test/CodeGen/AMDGPU/fp-classify.ll
    llvm/test/CodeGen/AMDGPU/icmp.i16.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 40eaba2c09209d..3bbbbcf71d8aec 100644

--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -1207,9 +1207,8 @@ static int getV_CMPOpcode(CmpInst::Predicate P, unsigned Size,
                           unsigned FakeS16Opc, unsigned S32Opc,
                           unsigned S64Opc) {
     if (Size == 16)
-      // FIXME-TRUE16 use TrueS16Opc when realtrue16 is supported for CMP code
       return ST.hasTrue16BitInsts()
-                 ? ST.useRealTrue16Insts() ? FakeS16Opc : FakeS16Opc
+                 ? ST.useRealTrue16Insts() ? TrueS16Opc : FakeS16Opc
                  : S16Opc;
     if (Size == 32)
       return S32Opc;

diff  --git a/llvm/lib/Target/AMDGPU/VOPCInstructions.td b/llvm/lib/Target/AMDGPU/VOPCInstructions.td
index e16ac4423265ec..00a3381b3fd49e 100644
--- a/llvm/lib/Target/AMDGPU/VOPCInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPCInstructions.td
@@ -1035,6 +1035,20 @@ multiclass VOPCClassPat64<string inst_name> {
   >;
 }
 
+multiclass VOPCClassPat64_t16<string inst_name> {
+  defvar inst = !cast<VOP_Pseudo>(inst_name#"_t16_e64");
+  defvar P = inst.Pfl;
+  def : GCNPat <
+    (i1:$sdst
+      (AMDGPUfp_class
+        (P.Src0VT (VOP3ModsNonCanonicalizing P.Src0VT:$src0, i32:$src0_modifiers)),
+        i32:$src1)),
+    (inst i32:$src0_modifiers, VSrcT_f16:$src0,
+          0 /* src1_modifiers */, (f16 (EXTRACT_SUBREG VGPR_32:$src1, lo16)),
+          0) /* op_sel */
+  >;
+}
+
 multiclass VOPCClassPat64_fake16<string inst_name> {
   defvar inst = !cast<VOP_Pseudo>(inst_name#"_fake16_e64");
   defvar P = inst.Pfl;
@@ -1158,6 +1172,7 @@ multiclass VOPC_CLASS_F16 <string opName> {
   }
   let True16Predicate = UseRealTrue16Insts in {
     defm _t16 : VOPC_Class_Pseudos <opName#"_t16", VOPC_I1_F16_I16_t16, 0>;
+    defm : VOPCClassPat64_t16<NAME>;
   }
   let True16Predicate = UseFakeTrue16Insts in {
     defm _fake16 : VOPC_Class_Pseudos <opName#"_fake16", VOPC_I1_F16_I16_fake16, 0>;
@@ -1207,27 +1222,30 @@ defm V_CMPX_CLASS_F16 : VOPCX_CLASS_F16 <"v_cmpx_class_f16">;
 
 // We need to use COPY_TO_REGCLASS to w/a the problem when ReplaceAllUsesWith()
 // complaints it cannot replace i1 <-> i64/i32 if node was not morphed in place.
-multiclass ICMP_Pattern <PatFrags cond, Instruction inst, ValueType vt> {
+multiclass ICMP_Pattern <PatFrags cond, Instruction inst, ValueType vt, dag dstInst = (inst $src0, $src1)> {
   let WaveSizePredicate = isWave64 in
   def : GCNPat <
     (i64 (AMDGPUsetcc vt:$src0, vt:$src1, cond)),
-    (i64 (COPY_TO_REGCLASS (inst $src0, $src1), SReg_64))
+    (i64 (COPY_TO_REGCLASS dstInst, SReg_64))
   >;
 
   let WaveSizePredicate = isWave32 in {
     def : GCNPat <
       (i32 (AMDGPUsetcc vt:$src0, vt:$src1, cond)),
-      (i32 (COPY_TO_REGCLASS (inst $src0, $src1), SReg_32))
+      (i32 (COPY_TO_REGCLASS dstInst, SReg_32))
     >;
 
     // Support codegen of i64 setcc in wave32 mode.
     def : GCNPat <
       (i64 (AMDGPUsetcc vt:$src0, vt:$src1, cond)),
-      (i64 (REG_SEQUENCE SReg_64, (inst $src0, $src1), sub0, (S_MOV_B32 (i32 0)), sub1))
+      (i64 (REG_SEQUENCE SReg_64, dstInst, sub0, (S_MOV_B32 (i32 0)), sub1))
     >;
   }
 }
 
+multiclass ICMP_Pattern_t16<PatFrags cond, Instruction inst, ValueType vt>
+    : ICMP_Pattern<cond, inst, vt, (inst 0, $src0, 0, $src1)>;
+
 defm : ICMP_Pattern <COND_EQ, V_CMP_EQ_U32_e64, i32>;
 defm : ICMP_Pattern <COND_NE, V_CMP_NE_U32_e64, i32>;
 defm : ICMP_Pattern <COND_UGT, V_CMP_GT_U32_e64, i32>;
@@ -1250,6 +1268,19 @@ defm : ICMP_Pattern <COND_SGE, V_CMP_GE_I64_e64, i64>;
 defm : ICMP_Pattern <COND_SLT, V_CMP_LT_I64_e64, i64>;
 defm : ICMP_Pattern <COND_SLE, V_CMP_LE_I64_e64, i64>;
 
+let True16Predicate = UseRealTrue16Insts in {
+defm : ICMP_Pattern_t16 <COND_EQ, V_CMP_EQ_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_NE, V_CMP_NE_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_UGT, V_CMP_GT_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_UGE, V_CMP_GE_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_ULT, V_CMP_LT_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_ULE, V_CMP_LE_U16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_SGT, V_CMP_GT_I16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_SGE, V_CMP_GE_I16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_SLT, V_CMP_LT_I16_t16_e64, i16>;
+defm : ICMP_Pattern_t16 <COND_SLE, V_CMP_LE_I16_t16_e64, i16>;
+} // End True16Predicate = UseRealTrue16Insts
+
 let True16Predicate = UseFakeTrue16Insts in {
 defm : ICMP_Pattern <COND_EQ, V_CMP_EQ_U16_fake16_e64, i16>;
 defm : ICMP_Pattern <COND_NE, V_CMP_NE_U16_fake16_e64, i16>;
@@ -1335,6 +1366,24 @@ defm : FCMP_Pattern <COND_UGE, V_CMP_NLT_F64_e64, f64>;
 defm : FCMP_Pattern <COND_ULT, V_CMP_NGE_F64_e64, f64>;
 defm : FCMP_Pattern <COND_ULE, V_CMP_NGT_F64_e64, f64>;
 
+let True16Predicate = UseRealTrue16Insts in {
+defm : FCMP_Pattern <COND_O, V_CMP_O_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_UO, V_CMP_U_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_OEQ, V_CMP_EQ_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_ONE, V_CMP_NEQ_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_OGT, V_CMP_GT_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_OGE, V_CMP_GE_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_OLT, V_CMP_LT_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_OLE, V_CMP_LE_F16_t16_e64, f16>;
+
+defm : FCMP_Pattern <COND_UEQ, V_CMP_NLG_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_UNE, V_CMP_NEQ_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_UGT, V_CMP_NLE_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_UGE, V_CMP_NLT_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_ULT, V_CMP_NGE_F16_t16_e64, f16>;
+defm : FCMP_Pattern <COND_ULE, V_CMP_NGT_F16_t16_e64, f16>;
+} // End True16Predicate = UseRealTrue16Insts
+
 let True16Predicate = UseFakeTrue16Insts in {
 defm : FCMP_Pattern <COND_O, V_CMP_O_F16_fake16_e64, f16>;
 defm : FCMP_Pattern <COND_UO, V_CMP_U_F16_fake16_e64, f16>;

diff  --git a/llvm/lib/Target/AMDGPU/VOPInstructions.td b/llvm/lib/Target/AMDGPU/VOPInstructions.td
index eb18cabe368ce0..d6ad01c8f9b358 100644
--- a/llvm/lib/Target/AMDGPU/VOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPInstructions.td
@@ -2019,6 +2019,10 @@ def : ClassPat<V_CMP_CLASS_F16_e64, f16> {
   let True16Predicate = NotHasTrue16BitInsts;
 }
 
+def : ClassPat_t16<V_CMP_CLASS_F16_t16_e64, f16> {
+  let True16Predicate = UseRealTrue16Insts;
+}
+
 def : ClassPat_t16<V_CMP_CLASS_F16_fake16_e64, f16> {
   let True16Predicate = UseFakeTrue16Insts;
 }

diff  --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w32.mir
index cdb67caea12cf0..49383135ab0c58 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w32.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w32.mir
@@ -17,11 +17,9 @@ body: |
     ; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
     ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_]]
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_1:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_1]]
-    ; GFX11-TRUE16-NEXT: [[V_CMP_F_F16_fake16_e64_:%[0-9]+]]:sreg_32 = V_CMP_F_F16_fake16_e64 0, [[COPY2]], 0, [[COPY3]], 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_F_F16_fake16_e64_]]
+    ; GFX11-TRUE16-NEXT: [[V_CMP_F_F16_t16_e64_:%[0-9]+]]:sreg_32 = V_CMP_F_F16_t16_e64 0, [[V_CVT_F16_F32_t16_e64_]], 0, [[V_CVT_F16_F32_t16_e64_1]], 0, 0, implicit $mode, implicit $exec
+    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_F_F16_t16_e64_]]
     ;
     ; GFX11-FAKE16-LABEL: name: fcmp_false_f16
     ; GFX11-FAKE16: liveins: $vgpr0, $vgpr1
@@ -55,11 +53,9 @@ body: |
     ; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
     ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_]]
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_1:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_1]]
-    ; GFX11-TRUE16-NEXT: [[V_CMP_TRU_F16_fake16_e64_:%[0-9]+]]:sreg_32 = V_CMP_TRU_F16_fake16_e64 0, [[COPY2]], 0, [[COPY3]], 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_TRU_F16_fake16_e64_]]
+    ; GFX11-TRUE16-NEXT: [[V_CMP_TRU_F16_t16_e64_:%[0-9]+]]:sreg_32 = V_CMP_TRU_F16_t16_e64 0, [[V_CVT_F16_F32_t16_e64_]], 0, [[V_CVT_F16_F32_t16_e64_1]], 0, 0, implicit $mode, implicit $exec
+    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_TRU_F16_t16_e64_]]
     ;
     ; GFX11-FAKE16-LABEL: name: fcmp_true_f16
     ; GFX11-FAKE16: liveins: $vgpr0, $vgpr1

diff  --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w64.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w64.mir
index ed811d37c3d0fc..828eb5d3fb40ac 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w64.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.fcmp.constants.w64.mir
@@ -17,11 +17,9 @@ body: |
     ; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
     ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_]]
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_1:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_1]]
-    ; GFX11-TRUE16-NEXT: [[V_CMP_F_F16_fake16_e64_:%[0-9]+]]:sreg_64 = V_CMP_F_F16_fake16_e64 0, [[COPY2]], 0, [[COPY3]], 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_F_F16_fake16_e64_]]
+    ; GFX11-TRUE16-NEXT: [[V_CMP_F_F16_t16_e64_:%[0-9]+]]:sreg_64 = V_CMP_F_F16_t16_e64 0, [[V_CVT_F16_F32_t16_e64_]], 0, [[V_CVT_F16_F32_t16_e64_1]], 0, 0, implicit $mode, implicit $exec
+    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_F_F16_t16_e64_]]
     ;
     ; GFX11-FAKE16-LABEL: name: fcmp_false_f16
     ; GFX11-FAKE16: liveins: $vgpr0, $vgpr1
@@ -55,11 +53,9 @@ body: |
     ; GFX11-TRUE16-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
     ; GFX11-TRUE16-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_]]
     ; GFX11-TRUE16-NEXT: [[V_CVT_F16_F32_t16_e64_1:%[0-9]+]]:vgpr_16 = nofpexcept V_CVT_F16_F32_t16_e64 0, [[COPY1]], 0, 0, 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[V_CVT_F16_F32_t16_e64_1]]
-    ; GFX11-TRUE16-NEXT: [[V_CMP_TRU_F16_fake16_e64_:%[0-9]+]]:sreg_64 = V_CMP_TRU_F16_fake16_e64 0, [[COPY2]], 0, [[COPY3]], 0, implicit $mode, implicit $exec
-    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_TRU_F16_fake16_e64_]]
+    ; GFX11-TRUE16-NEXT: [[V_CMP_TRU_F16_t16_e64_:%[0-9]+]]:sreg_64 = V_CMP_TRU_F16_t16_e64 0, [[V_CVT_F16_F32_t16_e64_]], 0, [[V_CVT_F16_F32_t16_e64_1]], 0, 0, implicit $mode, implicit $exec
+    ; GFX11-TRUE16-NEXT: S_ENDPGM 0, implicit [[V_CMP_TRU_F16_t16_e64_]]
     ;
     ; GFX11-FAKE16-LABEL: name: fcmp_true_f16
     ; GFX11-FAKE16: liveins: $vgpr0, $vgpr1

diff  --git a/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll b/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
index a00abecabdffb9..9b98bd6c9ab03d 100644
--- a/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcmp.f16.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
 ; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=SI %s
 ; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=fiji -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=VI %s
-; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX11 %s
+; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,+real-true16 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX11-TRUE16 %s
+; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1100 -mattr=-flat-for-global,-real-true16 -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX11-FAKE16 %s
 ; RUN:  llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx1200 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -enable-var-scope -check-prefixes=GFX12 %s
 
 define amdgpu_kernel void @fcmp_f16_lt(
@@ -55,30 +56,57 @@ define amdgpu_kernel void @fcmp_f16_lt(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_lt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_lt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_lt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_lt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -167,31 +195,58 @@ define amdgpu_kernel void @fcmp_f16_lt_abs(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_lt_abs:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_lt_f16_e64 s2, |v0|, |v1|
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s2
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_lt_abs:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e64 s2, |v0.l|, |v0.h|
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s2
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_lt_abs:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_lt_f16_e64 s2, |v0|, |v1|
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s2
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_lt_abs:
 ; GFX12:       ; %bb.0: ; %entry
@@ -285,30 +340,57 @@ define amdgpu_kernel void @fcmp_f16_eq(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_eq:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_eq:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_eq:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_eq:
 ; GFX12:       ; %bb.0: ; %entry
@@ -397,30 +479,57 @@ define amdgpu_kernel void @fcmp_f16_le(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_le:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_le:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_le:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_le:
 ; GFX12:       ; %bb.0: ; %entry
@@ -509,30 +618,57 @@ define amdgpu_kernel void @fcmp_f16_gt(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_gt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_gt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_gt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_gt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -621,30 +757,57 @@ define amdgpu_kernel void @fcmp_f16_lg(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_lg:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_lg:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_lg:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_lg:
 ; GFX12:       ; %bb.0: ; %entry
@@ -733,30 +896,57 @@ define amdgpu_kernel void @fcmp_f16_ge(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_ge:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_ge:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_ge:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_ge:
 ; GFX12:       ; %bb.0: ; %entry
@@ -845,30 +1035,57 @@ define amdgpu_kernel void @fcmp_f16_o(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_o:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_o:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_o:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_o:
 ; GFX12:       ; %bb.0: ; %entry
@@ -957,30 +1174,57 @@ define amdgpu_kernel void @fcmp_f16_u(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_u:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_u:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_u:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_u:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1069,30 +1313,57 @@ define amdgpu_kernel void @fcmp_f16_nge(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_nge:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_nge:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_nge:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_nge:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1181,30 +1452,57 @@ define amdgpu_kernel void @fcmp_f16_nlg(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_nlg:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_nlg:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_nlg:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_nlg:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1293,30 +1591,57 @@ define amdgpu_kernel void @fcmp_f16_ngt(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_ngt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_ngt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_ngt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_ngt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1405,30 +1730,57 @@ define amdgpu_kernel void @fcmp_f16_nle(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_nle:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_nle:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_nle:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_nle:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1517,30 +1869,57 @@ define amdgpu_kernel void @fcmp_f16_neq(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_neq:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_neq:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_neq:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_neq:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1629,30 +2008,57 @@ define amdgpu_kernel void @fcmp_f16_nlt(
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_f16_nlt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0, v1
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_f16_nlt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.h, v1.l
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0.l, v0.h
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_f16_nlt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v0, off, s[12:15], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    buffer_load_u16 v1, off, s[4:7], 0 glc dlc
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v0, v1
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b32 v0, off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_f16_nlt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1751,35 +2157,65 @@ define amdgpu_kernel void @fcmp_v2f16_lt(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_lt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_lt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_lt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_lt_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_lt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -1885,35 +2321,65 @@ define amdgpu_kernel void @fcmp_v2f16_eq(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_eq:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_eq:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_eq:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_eq_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_eq:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2018,35 +2484,65 @@ define amdgpu_kernel void @fcmp_v2f16_le(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_le:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_le_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_le:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_le:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_le_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_le:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2151,35 +2647,65 @@ define amdgpu_kernel void @fcmp_v2f16_gt(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_gt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_gt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_gt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_gt_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_gt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2285,35 +2811,65 @@ define amdgpu_kernel void @fcmp_v2f16_lg(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_lg:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_lg:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_lg:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_lg_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_lg:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2419,35 +2975,65 @@ define amdgpu_kernel void @fcmp_v2f16_ge(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_ge:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_ge:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_ge:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_ge_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_ge:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2553,35 +3139,65 @@ define amdgpu_kernel void @fcmp_v2f16_o(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_o:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_o_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_o:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_o:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_o_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_o:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2687,35 +3303,65 @@ define amdgpu_kernel void @fcmp_v2f16_u(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_u:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_u_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_u:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_u:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_u_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_u:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2820,35 +3466,65 @@ define amdgpu_kernel void @fcmp_v2f16_nge(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_nge:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_nge:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_nge:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_nge_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_nge:
 ; GFX12:       ; %bb.0: ; %entry
@@ -2953,35 +3629,65 @@ define amdgpu_kernel void @fcmp_v2f16_nlg(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_nlg:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_nlg:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_nlg:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_nlg_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_nlg:
 ; GFX12:       ; %bb.0: ; %entry
@@ -3087,35 +3793,65 @@ define amdgpu_kernel void @fcmp_v2f16_ngt(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_ngt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_ngt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_ngt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_ngt_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_ngt:
 ; GFX12:       ; %bb.0: ; %entry
@@ -3220,35 +3956,65 @@ define amdgpu_kernel void @fcmp_v2f16_nle(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_nle:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_nle:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_nle:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_nle_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_nle:
 ; GFX12:       ; %bb.0: ; %entry
@@ -3353,35 +4119,65 @@ define amdgpu_kernel void @fcmp_v2f16_neq(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_neq:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_neq:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_neq:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_neq_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_neq:
 ; GFX12:       ; %bb.0: ; %entry
@@ -3486,35 +4282,65 @@ define amdgpu_kernel void @fcmp_v2f16_nlt(
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: fcmp_v2f16_nlt:
-; GFX11:       ; %bb.0: ; %entry
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
-; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
-; GFX11-NEXT:    s_mov_b32 s10, -1
-; GFX11-NEXT:    s_mov_b32 s11, 0x31016000
-; GFX11-NEXT:    s_mov_b32 s6, s10
-; GFX11-NEXT:    s_mov_b32 s7, s11
-; GFX11-NEXT:    s_mov_b32 s14, s10
-; GFX11-NEXT:    s_mov_b32 s15, s11
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    s_mov_b32 s12, s2
-; GFX11-NEXT:    s_mov_b32 s13, s3
-; GFX11-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
-; GFX11-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
-; GFX11-NEXT:    s_mov_b32 s8, s0
-; GFX11-NEXT:    s_mov_b32 s9, s1
-; GFX11-NEXT:    s_waitcnt vmcnt(1)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
-; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v1, v0
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_3)
-; GFX11-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v3, v2
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
-; GFX11-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: fcmp_v2f16_nlt:
+; GFX11-TRUE16:       ; %bb.0: ; %entry
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-TRUE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-TRUE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-TRUE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-TRUE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-TRUE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-TRUE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-TRUE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-TRUE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-TRUE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v1.l, v0.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-TRUE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v3.l, v2.l
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-TRUE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: fcmp_v2f16_nlt:
+; GFX11-FAKE16:       ; %bb.0: ; %entry
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    s_load_b64 s[4:5], s[4:5], 0x34
+; GFX11-FAKE16-NEXT:    s_mov_b32 s10, -1
+; GFX11-FAKE16-NEXT:    s_mov_b32 s11, 0x31016000
+; GFX11-FAKE16-NEXT:    s_mov_b32 s6, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s7, s11
+; GFX11-FAKE16-NEXT:    s_mov_b32 s14, s10
+; GFX11-FAKE16-NEXT:    s_mov_b32 s15, s11
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    s_mov_b32 s12, s2
+; GFX11-FAKE16-NEXT:    s_mov_b32 s13, s3
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v0, off, s[4:7], 0
+; GFX11-FAKE16-NEXT:    buffer_load_b32 v1, off, s[12:15], 0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s8, s0
+; GFX11-FAKE16-NEXT:    s_mov_b32 s9, s1
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(1)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
+; GFX11-FAKE16-NEXT:    s_waitcnt vmcnt(0)
+; GFX11-FAKE16-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
+; GFX11-FAKE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v1, v0
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_3)
+; GFX11-FAKE16-NEXT:    v_cmp_nlt_f16_e32 vcc_lo, v3, v2
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc_lo
+; GFX11-FAKE16-NEXT:    buffer_store_b64 v[0:1], off, s[8:11], 0
+; GFX11-FAKE16-NEXT:    s_endpgm
 ;
 ; GFX12-LABEL: fcmp_v2f16_nlt:
 ; GFX12:       ; %bb.0: ; %entry

diff  --git a/llvm/test/CodeGen/AMDGPU/fp-classify.ll b/llvm/test/CodeGen/AMDGPU/fp-classify.ll
index e7c425a2d2752d..cc11e256d5544e 100644
--- a/llvm/test/CodeGen/AMDGPU/fp-classify.ll
+++ b/llvm/test/CodeGen/AMDGPU/fp-classify.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
 ; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=SI %s
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI %s
-; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-TRUE16 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-FAKE16 %s
 
 declare float @llvm.fabs.f32(float) #1
 declare double @llvm.fabs.f64(double) #1
@@ -619,18 +620,32 @@ define amdgpu_kernel void @test_isinf_pattern_f16(ptr addrspace(1) nocapture %ou
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: test_isinf_pattern_f16:
-; GFX11:       ; %bb.0:
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
-; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GFX11-NEXT:    v_mov_b32_e32 v0, 0
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x204
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
-; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: test_isinf_pattern_f16:
+; GFX11-TRUE16:       ; %bb.0:
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, s2
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_class_f16_e64 s2, v0.l, 0x204
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s2
+; GFX11-TRUE16-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: test_isinf_pattern_f16:
+; GFX11-FAKE16:       ; %bb.0:
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x204
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
+; GFX11-FAKE16-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-FAKE16-NEXT:    s_endpgm
   %fabs = tail call half @llvm.fabs.f16(half %x) #1
   %cmp = fcmp oeq half %fabs, 0xH7C00
   %ext = zext i1 %cmp to i32
@@ -669,18 +684,32 @@ define amdgpu_kernel void @test_isfinite_pattern_0_f16(ptr addrspace(1) nocaptur
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: test_isfinite_pattern_0_f16:
-; GFX11:       ; %bb.0:
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
-; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GFX11-NEXT:    v_mov_b32_e32 v0, 0
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x1f8
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
-; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: test_isfinite_pattern_0_f16:
+; GFX11-TRUE16:       ; %bb.0:
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, s2
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_class_f16_e64 s2, v0.l, 0x1f8
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s2
+; GFX11-TRUE16-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: test_isfinite_pattern_0_f16:
+; GFX11-FAKE16:       ; %bb.0:
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x1f8
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
+; GFX11-FAKE16-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-FAKE16-NEXT:    s_endpgm
   %ord = fcmp ord half %x, 0.0
   %x.fabs = tail call half @llvm.fabs.f16(half %x) #1
   %ninf = fcmp une half %x.fabs, 0xH7C00
@@ -718,18 +747,32 @@ define amdgpu_kernel void @test_isfinite_pattern_4_f16(ptr addrspace(1) nocaptur
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
-; GFX11-LABEL: test_isfinite_pattern_4_f16:
-; GFX11:       ; %bb.0:
-; GFX11-NEXT:    s_clause 0x1
-; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x2c
-; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GFX11-NEXT:    v_mov_b32_e32 v0, 0
-; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX11-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x1f8
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
-; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
-; GFX11-NEXT:    s_endpgm
+; GFX11-TRUE16-LABEL: test_isfinite_pattern_4_f16:
+; GFX11-TRUE16:       ; %bb.0:
+; GFX11-TRUE16-NEXT:    s_clause 0x1
+; GFX11-TRUE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-TRUE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-TRUE16-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-TRUE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, s2
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_cmp_class_f16_e64 s2, v0.l, 0x1f8
+; GFX11-TRUE16-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s2
+; GFX11-TRUE16-NEXT:    global_store_b32 v1, v0, s[0:1]
+; GFX11-TRUE16-NEXT:    s_endpgm
+;
+; GFX11-FAKE16-LABEL: test_isfinite_pattern_4_f16:
+; GFX11-FAKE16:       ; %bb.0:
+; GFX11-FAKE16-NEXT:    s_clause 0x1
+; GFX11-FAKE16-NEXT:    s_load_b32 s2, s[4:5], 0x2c
+; GFX11-FAKE16-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
+; GFX11-FAKE16-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-FAKE16-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-FAKE16-NEXT:    v_cmp_class_f16_e64 s2, s2, 0x1f8
+; GFX11-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT:    v_cndmask_b32_e64 v1, 0, 1, s2
+; GFX11-FAKE16-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-FAKE16-NEXT:    s_endpgm
   %ord = fcmp ord half %x, 0.0
   %x.fabs = tail call half @llvm.fabs.f16(half %x) #1
   %ninf = fcmp one half %x.fabs, 0xH7C00

diff  --git a/llvm/test/CodeGen/AMDGPU/icmp.i16.ll b/llvm/test/CodeGen/AMDGPU/icmp.i16.ll
index c1a074a81b2aa2..790a457c2b3371 100644
--- a/llvm/test/CodeGen/AMDGPU/icmp.i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/icmp.i16.ll
@@ -1,5 +1,9 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn -verify-machineinstrs < %s| FileCheck -check-prefix=GCN -check-prefix=SI %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 -verify-machineinstrs < %s| FileCheck -check-prefixes=GCN,GFX11-FAKE16 %s
+; FIXME-TRUE16. In true16 flow, the codegen introduces addtional s2v copy and mov, and revert the operand order thus picking 
diff erent cmp instructions
+; This should be corrected after addtional mov/copy is removed
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 -verify-machineinstrs < %s| FileCheck -check-prefixes=GCN,GFX11-TRUE16 %s
 
 ;;;==========================================================================;;;
 ;; 16-bit integer comparisons
@@ -8,6 +12,8 @@
 ; GCN-LABEL: {{^}}i16_eq:
 ; VI: v_cmp_eq_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_eq_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_eq_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_eq_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_eq(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -26,6 +32,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ne:
 ; VI: v_cmp_ne_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ne_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ne_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ne_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_ne(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -44,6 +52,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ugt:
 ; VI: v_cmp_gt_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_gt_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_gt_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_gt_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_ugt(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -62,6 +72,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_uge:
 ; VI: v_cmp_ge_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ge_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ge_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ge_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_uge(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -80,6 +92,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ult:
 ; VI: v_cmp_lt_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_lt_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_lt_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_lt_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_ult(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -98,6 +112,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ule:
 ; VI: v_cmp_le_u16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_le_u32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_le_u16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_le_u16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_ule(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -117,6 +133,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sgt:
 ; VI: v_cmp_gt_i16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_gt_i32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_gt_i16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_gt_i16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_sgt(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -135,6 +153,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sge:
 ; VI: v_cmp_ge_i16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ge_i32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ge_i16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ge_i16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_sge(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -153,6 +173,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_slt:
 ; VI: v_cmp_lt_i16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_lt_i32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_lt_i16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_lt_i16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_slt(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -171,6 +193,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sle:
 ; VI: v_cmp_le_i16_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_le_i32_e32 vcc, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_le_i16_e32 vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_le_i16_e32 vcc_lo, v{{[0-9]+}}.l, v{{[0-9]+}}.h
 define amdgpu_kernel void @i16_sle(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, ptr addrspace(1) %b.ptr) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -190,6 +214,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_eq_v_s:
 ; VI: v_cmp_eq_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_eq_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_eq_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_eq_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_eq_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -206,6 +232,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ne_v_s:
 ; VI: v_cmp_ne_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ne_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ne_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ne_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_ne_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -222,6 +250,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ugt_v_s:
 ; VI: v_cmp_lt_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_lt_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_lt_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_gt_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_ugt_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -238,6 +268,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_uge_v_s:
 ; VI: v_cmp_le_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_le_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_le_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ge_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_uge_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -254,6 +286,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ult_v_s:
 ; VI: v_cmp_gt_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_gt_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_gt_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_lt_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_ult_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -270,6 +304,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_ule_v_s:
 ; VI: v_cmp_ge_u16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ge_u32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ge_u16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_le_u16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_ule_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -286,6 +322,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sgt_v_s:
 ; VI: v_cmp_lt_i16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_lt_i32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_lt_i16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_gt_i16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_sgt_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -302,6 +340,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sge_v_s:
 ; VI: v_cmp_le_i16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_le_i32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_le_i16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_ge_i16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_sge_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -318,6 +358,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_slt_v_s:
 ; VI: v_cmp_gt_i16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_gt_i32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_gt_i16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_lt_i16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_slt_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -334,6 +376,8 @@ entry:
 ; GCN-LABEL: {{^}}i16_sle_v_s:
 ; VI: v_cmp_ge_i16_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
 ; SI: v_cmp_ge_i32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-FAKE16: v_cmp_ge_i16_e32 vcc_lo, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX11-TRUE16: v_cmp_le_i16_e32 vcc_lo, v{{[0-9]+}}.h, v{{[0-9]+}}.l
 define amdgpu_kernel void @i16_sle_v_s(ptr addrspace(1) %out, ptr addrspace(1) %a.ptr, i16 %b) #0 {
 entry:
   %tid = call i32 @llvm.amdgcn.workitem.id.x()


        
