[AMDGPU][True16][CodeGen] true16 codegen pattern for v_pack_b32_f16 (llvm#121988)

broxigarchen · web-flow · commit 26e13091ea5a · 2025-01-13T12:26:36.000-05:00
true16 codegen pattern for v_pack_b32_f16
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -3390,6 +3390,9 @@ let SubtargetPredicate = isGFX9Plus in {
 let True16Predicate = NotHasTrue16BitInsts in
   def : PackB32Pat<V_PACK_B32_F16_e64>;
 
+let True16Predicate = UseRealTrue16Insts in
+  def : PackB32Pat<V_PACK_B32_F16_t16_e64>;
+
 let True16Predicate = UseFakeTrue16Insts in
   def : PackB32Pat<V_PACK_B32_F16_fake16_e64>;
 } // End SubtargetPredicate = isGFX9Plus
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
@@ -160,14 +160,9 @@ define amdgpu_kernel void @ceil_v2f16(
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-NEXT:    v_ceil_f16_e32 v0.l, v0.l
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-NEXT:    v_ceil_f16_e32 v0.h, v1.l
-; GFX11-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-NEXT:    s_endpgm
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll b/llvm/test/CodeGen/AMDGPU/llvm.floor.f16.ll
@@ -161,14 +161,9 @@ define amdgpu_kernel void @floor_v2f16(
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-NEXT:    v_floor_f16_e32 v0.l, v0.l
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-NEXT:    v_floor_f16_e32 v0.h, v1.l
-; GFX11-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-NEXT:    s_endpgm
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll b/llvm/test/CodeGen/AMDGPU/llvm.ldexp.ll
@@ -480,9 +480,8 @@ define <2 x half> @test_ldexp_v2f16_v2i32(<2 x half> %a, <2 x i32> %b) {
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v3.l, v2.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v2f16_v2i32:
@@ -610,9 +609,7 @@ define <2 x half> @test_ldexp_v2f16_v2i16(<2 x half> %a, <2 x i16> %b) {
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v1.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v3.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v1, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v2f16_v2i16:
@@ -737,15 +734,13 @@ define <3 x half> @test_ldexp_v3f16_v3i32(<3 x half> %a, <3 x i32> %b) {
 ; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v3, v3, s0, 0x7fff
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v2, v2, s0, 0x7fff
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v4, v4, s0, 0x7fff
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v5.l, v3.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v2, v4, s0, 0x7fff
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v1.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v3, v0, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v1.l, v4.l
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v3f16_v3i32:
@@ -891,12 +886,9 @@ define <3 x half> @test_ldexp_v3f16_v3i16(<3 x half> %a, <3 x i16> %b) {
 ; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v5, 16, v0
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v2.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v1.l, v3.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v5.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v0.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v0, v2, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v3f16_v3i16:
@@ -1036,24 +1028,21 @@ define <4 x half> @test_ldexp_v4f16_v4i32(<4 x half> %a, <4 x i32> %b) {
 ; GFX11-SDAG-TRUE16:       ; %bb.0:
 ; GFX11-SDAG-TRUE16-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-SDAG-TRUE16-NEXT:    s_movk_i32 s0, 0x8000
-; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v6, 16, v1
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v5, v5, s0, 0x7fff
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v3, v3, s0, 0x7fff
-; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v7, 16, v0
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v2, v2, s0, 0x7fff
 ; GFX11-SDAG-TRUE16-NEXT:    v_med3_i32 v4, v4, s0, 0x7fff
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.h, v6.l, v5.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v6.l, v3.l
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.h, v7.l, v5.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v7.l, v3.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v2.l
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_2)
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v1.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v1.h
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v2, v0, 0x5040100
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v1, v3, v1, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v1, v1.l, v1.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v4f16_v4i32:
@@ -1238,20 +1227,14 @@ define <4 x half> @test_ldexp_v4f16_v4i16(<4 x half> %a, <4 x i16> %b) {
 ; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v5, 16, v2
 ; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v6, 16, v0
 ; GFX11-SDAG-TRUE16-NEXT:    v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v1.l, v3.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.l, v0.l, v2.l
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v1.l, v3.l
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.l, v6.l, v5.l
+; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v0.h, v6.l, v5.l
 ; GFX11-SDAG-TRUE16-NEXT:    v_ldexp_f16_e32 v1.h, v7.l, v4.l
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v2.l, v0.l
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v3.l, v0.h
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v1.l
-; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v1.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v0, v0, v2, 0x5040100
-; GFX11-SDAG-TRUE16-NEXT:    v_perm_b32 v1, v1, v3, 0x5040100
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
+; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v1, v1.l, v1.h
 ; GFX11-SDAG-TRUE16-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-SDAG-FAKE16-LABEL: test_ldexp_v4f16_v4i16:
diff --git a/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll b/llvm/test/CodeGen/AMDGPU/sitofp.f16.ll
@@ -237,14 +237,9 @@ define amdgpu_kernel void @sitofp_v2i16_to_v2f16(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_f16_i16_e32 v0.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_cvt_f16_i16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -338,17 +333,13 @@ define amdgpu_kernel void @sitofp_v2i32_to_v2f16(
 ; GFX11-TRUE16-NEXT:    buffer_load_b64 v[0:1], off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s5, s1
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_cvt_f32_i32_e32 v0, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_f32_i32_e32 v1, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f32_i32_e32 v2, v0
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v0
-; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.h, v1
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.h, v2
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.h, v0.l
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll b/llvm/test/CodeGen/AMDGPU/uitofp.f16.ll
@@ -237,14 +237,9 @@ define amdgpu_kernel void @uitofp_v2i16_to_v2f16(
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-TRUE16-NEXT:    v_lshrrev_b32_e32 v1, 16, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_f16_u16_e32 v0.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX11-TRUE16-NEXT:    v_cvt_f16_u16_e32 v0.h, v1.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.l, v0.h
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
@@ -338,17 +333,13 @@ define amdgpu_kernel void @uitofp_v2i32_to_v2f16(
 ; GFX11-TRUE16-NEXT:    buffer_load_b64 v[0:1], off, s[8:11], 0
 ; GFX11-TRUE16-NEXT:    s_mov_b32 s5, s1
 ; GFX11-TRUE16-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-TRUE16-NEXT:    v_cvt_f32_u32_e32 v0, v0
 ; GFX11-TRUE16-NEXT:    v_cvt_f32_u32_e32 v1, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f32_u32_e32 v2, v0
 ; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v0
-; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.h, v1
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v1.l, v0.l
-; GFX11-TRUE16-NEXT:    v_mov_b16_e32 v0.l, v0.h
-; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-TRUE16-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX11-TRUE16-NEXT:    v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.l, v1
+; GFX11-TRUE16-NEXT:    v_cvt_f16_f32_e32 v0.h, v2
+; GFX11-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT:    v_pack_b32_f16 v0, v0.h, v0.l
 ; GFX11-TRUE16-NEXT:    buffer_store_b32 v0, off, s[4:7], 0
 ; GFX11-TRUE16-NEXT:    s_endpgm
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/v_pack.ll b/llvm/test/CodeGen/AMDGPU/v_pack.ll
