From e71da01f0f908417723a54cf8829a734a37fa173 Mon Sep 17 00:00:00 2001
From: Guo Chen <guochen2@amd.com>
Date: Tue, 24 Mar 2026 15:03:00 -0400
Subject: [PATCH] [AMDGPU][True16] add true16 pattern for cvt_pk_fp32_f8
 (#180096)

---
 llvm/lib/Target/AMDGPU/VOP1Instructions.td    |  21 +-
 .../CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.ll     | 414 ++++++++++++------
 2 files changed, 299 insertions(+), 136 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/VOP1Instructions.td b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
index 1c1282c0791f..d40378a5ac4b 100644
--- a/llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -821,12 +821,25 @@ class Cvt_PK_F32_F8_Pat_OpSel<SDPatternOperator node, int index,
          (inst_e32 $src))
 >;
 
+class Cvt_PK_F32_F8_Pat_t16<SDPatternOperator node, int index,
+    VOP3_Pseudo inst_e64> : GCNPat<
+    (v2f32 (node i32:$src, index)),
+    (inst_e64 0, (EXTRACT_SUBREG (i32 (COPY_TO_REGCLASS $src, VGPR_32)),
+      !if(index, hi16, lo16)), 0)
+>;
+
 let SubtargetPredicate = isGFX11Plus, OtherPredicates = [HasFP8ConversionInsts] in {
   foreach Index = [0, -1] in {
-    def : Cvt_PK_F32_F8_Pat_OpSel<int_amdgcn_cvt_pk_f32_fp8, Index,
-                                  V_CVT_PK_F32_FP8_fake16_e32, V_CVT_PK_F32_FP8_fake16_e64>;
-    def : Cvt_PK_F32_F8_Pat_OpSel<int_amdgcn_cvt_pk_f32_bf8, Index,
-                                  V_CVT_PK_F32_BF8_fake16_e32, V_CVT_PK_F32_BF8_fake16_e64>;
+    let True16Predicate = UseFakeTrue16Insts in {
+      def : Cvt_PK_F32_F8_Pat_OpSel<int_amdgcn_cvt_pk_f32_fp8, Index,
+                                    V_CVT_PK_F32_FP8_fake16_e32, V_CVT_PK_F32_FP8_fake16_e64>;
+      def : Cvt_PK_F32_F8_Pat_OpSel<int_amdgcn_cvt_pk_f32_bf8, Index,
+                                    V_CVT_PK_F32_BF8_fake16_e32, V_CVT_PK_F32_BF8_fake16_e64>;
+    }
+    let True16Predicate = UseRealTrue16Insts in {
+      def : Cvt_PK_F32_F8_Pat_t16<int_amdgcn_cvt_pk_f32_fp8, Index, V_CVT_PK_F32_FP8_t16_e64>;
+      def : Cvt_PK_F32_F8_Pat_t16<int_amdgcn_cvt_pk_f32_bf8, Index, V_CVT_PK_F32_BF8_t16_e64>;
+    }
   }
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.ll
index d847e8afcb0e..a267d2a515ae 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.cvt.fp8.ll
@@ -305,28 +305,51 @@ define <2 x float> @test_cvt_pk_f32_bf8_word0(i32 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_cvt_pk_f32_bf8_word0:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.l
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_cvt_pk_f32_bf8_word0:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_cvt_pk_f32_bf8_word0:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.l
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.l
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_cvt_pk_f32_bf8_word0:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 false)
   ret <2 x float> %ret
 }
@@ -338,28 +361,51 @@ define <2 x float> @test_cvt_pk_f32_bf8_word1(i32 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_cvt_pk_f32_bf8_word1:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_cvt_pk_f32_bf8_word1:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_cvt_pk_f32_bf8_word1:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_cvt_pk_f32_bf8_word1:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a, i1 true)
   ret <2 x float> %ret
 }
@@ -371,28 +417,51 @@ define <2 x float> @test_cvt_pk_f32_fp8_word0(i32 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_cvt_pk_f32_fp8_word0:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_cvt_pk_f32_fp8_word0:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_cvt_pk_f32_fp8_word0:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_cvt_pk_f32_fp8_word0:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 false)
   ret <2 x float> %ret
 }
@@ -404,28 +473,51 @@ define <2 x float> @test_cvt_pk_f32_fp8_word1(i32 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_fp8_sdwa v[0:1], v0 src0_sel:WORD_1
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_cvt_pk_f32_fp8_word1:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.h
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_cvt_pk_f32_fp8_word1:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_cvt_pk_f32_fp8_word1:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.h
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.h
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_cvt_pk_f32_fp8_word1:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a, i1 true)
   ret <2 x float> %ret
 }
@@ -1124,34 +1216,63 @@ define <2 x float> @test_sext_cvt_pk_f32_bf8_word1(i16 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_bf8_sdwa v[0:1], v0 src0_sel:WORD_1
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_sext_cvt_pk_f32_bf8_word1:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX1170-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1170-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1170-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_sext_cvt_pk_f32_bf8_word1:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1170-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_sext_cvt_pk_f32_bf8_word1:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX12-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1250-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_bf8_e32 v[0:1], v0.h
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_sext_cvt_pk_f32_bf8_word1:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_bf8_e64 v[0:1], v0 op_sel:[1,0]
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %a.sext = sext i16 %a to i32
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.bf8(i32 %a.sext, i1 true)
   ret <2 x float> %ret
@@ -1165,34 +1286,63 @@ define <2 x float> @test_sext_cvt_pk_f32_fp8_word0(i16 %a) {
 ; GFX9X-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
 ; GFX9X-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1170-LABEL: test_sext_cvt_pk_f32_fp8_word0:
-; GFX1170:       ; %bb.0:
-; GFX1170-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1170-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX1170-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1170-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX1170-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-TRUE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX1170-TRUE16:       ; %bb.0:
+; GFX1170-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1170-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1170-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX1170-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX12-LABEL: test_sext_cvt_pk_f32_fp8_word0:
-; GFX12:       ; %bb.0:
-; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT:    s_wait_expcnt 0x0
-; GFX12-NEXT:    s_wait_samplecnt 0x0
-; GFX12-NEXT:    s_wait_bvhcnt 0x0
-; GFX12-NEXT:    s_wait_kmcnt 0x0
-; GFX12-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX12-NEXT:    s_setpc_b64 s[30:31]
+; GFX1170-FAKE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX1170-FAKE16:       ; %bb.0:
+; GFX1170-FAKE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1170-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1170-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1170-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX1170-FAKE16-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: test_sext_cvt_pk_f32_fp8_word0:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_bfe_i32 v0, v0, 0, 16
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1250-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX12-TRUE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX12-TRUE16:       ; %bb.0:
+; GFX12-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX12-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX12-TRUE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-FAKE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX12-FAKE16:       ; %bb.0:
+; GFX12-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_expcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_samplecnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX12-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX12-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX12-FAKE16-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX1250-TRUE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX1250-TRUE16:       ; %bb.0:
+; GFX1250-TRUE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-TRUE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-TRUE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1250-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-TRUE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0.l
+; GFX1250-TRUE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-FAKE16-LABEL: test_sext_cvt_pk_f32_fp8_word0:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_bfe_i32 v0, v0, 0, 16
+; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_f32_fp8_e32 v[0:1], v0
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
   %a.sext = sext i16 %a to i32
   %ret = tail call <2 x float> @llvm.amdgcn.cvt.pk.f32.fp8(i32 %a.sext, i1 false)
   ret <2 x float> %ret