[llvm] 219a399 - AMDGPU: Fix assertion on assorted canonicalize vector types
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon May 22 08:45:11 PDT 2023
Author: Matt Arsenault
Date: 2023-05-22T16:44:44+01:00
New Revision: 219a3992ee737491e0fa319286ebafa86a30b203
URL: https://github.com/llvm/llvm-project/commit/219a3992ee737491e0fa319286ebafa86a30b203
DIFF: https://github.com/llvm/llvm-project/commit/219a3992ee737491e0fa319286ebafa86a30b203.diff
LOG: AMDGPU: Fix assertion on assorted canonicalize vector types
This assert should have the same set of vector types as the binary
and ternary case (although this assert is kind of pointless, the code
should work for any vector type as-is).
Fixes part of issue #32650.
Added:
Modified:
llvm/lib/Target/AMDGPU/SIISelLowering.cpp
llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 63176a249bf6..c8329730c678 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -4678,7 +4678,10 @@ SDValue SITargetLowering::splitUnaryVectorOp(SDValue Op,
SelectionDAG &DAG) const {
unsigned Opc = Op.getOpcode();
EVT VT = Op.getValueType();
- assert(VT == MVT::v4f16 || VT == MVT::v4i16);
+ assert(VT == MVT::v4i16 || VT == MVT::v4f16 || VT == MVT::v4f32 ||
+ VT == MVT::v8i16 || VT == MVT::v8f16 || VT == MVT::v16i16 ||
+ VT == MVT::v16f16 || VT == MVT::v8f32 || VT == MVT::v16f32 ||
+ VT == MVT::v32f32);
SDValue Lo, Hi;
std::tie(Lo, Hi) = DAG.SplitVectorOperand(Op.getNode(), 0);
diff --git a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
index eb1914d04bc1..1e9c2c11acb9 100644
--- a/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll
@@ -9,6 +9,12 @@ declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #0
declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
declare <3 x half> @llvm.canonicalize.v3f16(<3 x half>) #0
declare <4 x half> @llvm.canonicalize.v4f16(<4 x half>) #0
+declare <6 x half> @llvm.canonicalize.v6f16(<6 x half>) #0
+declare <8 x half> @llvm.canonicalize.v8f16(<8 x half>) #0
+declare <12 x half> @llvm.canonicalize.v12f16(<12 x half>) #0
+declare <16 x half> @llvm.canonicalize.v16f16(<16 x half>) #0
+declare <32 x half> @llvm.canonicalize.v32f16(<32 x half>) #0
+declare <64 x half> @llvm.canonicalize.v64f16(<64 x half>) #0
declare i32 @llvm.amdgcn.workitem.id.x() #0
define amdgpu_kernel void @test_fold_canonicalize_undef_value_f16(ptr addrspace(1) %out) #1 {
@@ -2227,6 +2233,807 @@ define <4 x half> @v_test_canonicalize_reg_undef_reg_reg_v4f16(half %val0, half
ret <4 x half> %canonicalized
}
+define <6 x half> @v_test_canonicalize_var_v6f16(<6 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v6f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v3, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v4, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v5, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v5
+; VI-NEXT: v_or_b32_e32 v1, v1, v4
+; VI-NEXT: v_or_b32_e32 v2, v2, v3
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v6f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v6f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
+; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <6 x half> @llvm.canonicalize.v6f16(<6 x half> %val)
+ ret <6 x half> %canonicalized
+}
+
+define <8 x half> @v_test_canonicalize_var_v8f16(<8 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v8f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v4, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v5, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v6, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v7, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v3, v3, v3
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v7
+; VI-NEXT: v_or_b32_e32 v1, v1, v6
+; VI-NEXT: v_or_b32_e32 v2, v2, v5
+; VI-NEXT: v_or_b32_e32 v3, v3, v4
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v8f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v8f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
+; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
+; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
+; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <8 x half> @llvm.canonicalize.v8f16(<8 x half> %val)
+ ret <8 x half> %canonicalized
+}
+
+define <12 x half> @v_test_canonicalize_var_v12f16(<12 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v12f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v6, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v7, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v8, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v9, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v10, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v11, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v5, v5, v5
+; VI-NEXT: v_max_f16_e32 v4, v4, v4
+; VI-NEXT: v_max_f16_e32 v3, v3, v3
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v11
+; VI-NEXT: v_or_b32_e32 v1, v1, v10
+; VI-NEXT: v_or_b32_e32 v2, v2, v9
+; VI-NEXT: v_or_b32_e32 v3, v3, v8
+; VI-NEXT: v_or_b32_e32 v4, v4, v7
+; VI-NEXT: v_or_b32_e32 v5, v5, v6
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v12f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
+; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
+; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v12f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
+; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
+; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
+; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
+; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
+; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
+; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
+; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
+; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
+; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
+; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <12 x half> @llvm.canonicalize.v12f16(<12 x half> %val)
+ ret <12 x half> %canonicalized
+}
+
+define <16 x half> @v_test_canonicalize_var_v16f16(<16 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v16f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v8, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v9, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v10, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v11, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v12, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v13, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v14, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v15, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v7, v7, v7
+; VI-NEXT: v_max_f16_e32 v6, v6, v6
+; VI-NEXT: v_max_f16_e32 v5, v5, v5
+; VI-NEXT: v_max_f16_e32 v4, v4, v4
+; VI-NEXT: v_max_f16_e32 v3, v3, v3
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v15
+; VI-NEXT: v_or_b32_e32 v1, v1, v14
+; VI-NEXT: v_or_b32_e32 v2, v2, v13
+; VI-NEXT: v_or_b32_e32 v3, v3, v12
+; VI-NEXT: v_or_b32_e32 v4, v4, v11
+; VI-NEXT: v_or_b32_e32 v5, v5, v10
+; VI-NEXT: v_or_b32_e32 v6, v6, v9
+; VI-NEXT: v_or_b32_e32 v7, v7, v8
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v16f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
+; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
+; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
+; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
+; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v16f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
+; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
+; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
+; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
+; CI-NEXT: v_cvt_f16_f32_e32 v12, v12
+; CI-NEXT: v_cvt_f16_f32_e32 v13, v13
+; CI-NEXT: v_cvt_f16_f32_e32 v14, v14
+; CI-NEXT: v_cvt_f16_f32_e32 v15, v15
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
+; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
+; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
+; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
+; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
+; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
+; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
+; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
+; CI-NEXT: v_cvt_f32_f16_e32 v12, v12
+; CI-NEXT: v_cvt_f32_f16_e32 v13, v13
+; CI-NEXT: v_cvt_f32_f16_e32 v14, v14
+; CI-NEXT: v_cvt_f32_f16_e32 v15, v15
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <16 x half> @llvm.canonicalize.v16f16(<16 x half> %val)
+ ret <16 x half> %canonicalized
+}
+
+define <32 x half> @v_test_canonicalize_var_v32f16(<32 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v32f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v20, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v20
+; VI-NEXT: v_max_f16_sdwa v20, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_or_b32_e32 v1, v1, v20
+; VI-NEXT: v_max_f16_sdwa v20, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_or_b32_e32 v2, v2, v20
+; VI-NEXT: v_max_f16_sdwa v20, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v3, v3, v3
+; VI-NEXT: v_or_b32_e32 v3, v3, v20
+; VI-NEXT: v_max_f16_sdwa v20, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v4, v4, v4
+; VI-NEXT: v_or_b32_e32 v4, v4, v20
+; VI-NEXT: v_max_f16_sdwa v20, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v5, v5, v5
+; VI-NEXT: v_or_b32_e32 v5, v5, v20
+; VI-NEXT: v_max_f16_sdwa v20, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v6, v6, v6
+; VI-NEXT: v_or_b32_e32 v6, v6, v20
+; VI-NEXT: v_max_f16_sdwa v20, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v7, v7, v7
+; VI-NEXT: v_or_b32_e32 v7, v7, v20
+; VI-NEXT: v_max_f16_sdwa v20, v8, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v8, v8, v8
+; VI-NEXT: v_or_b32_e32 v8, v8, v20
+; VI-NEXT: v_max_f16_sdwa v20, v9, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v9, v9, v9
+; VI-NEXT: v_or_b32_e32 v9, v9, v20
+; VI-NEXT: v_max_f16_sdwa v20, v10, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v10, v10, v10
+; VI-NEXT: v_max_f16_sdwa v16, v15, v15 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v17, v14, v14 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v18, v13, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_sdwa v19, v12, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_or_b32_e32 v10, v10, v20
+; VI-NEXT: v_max_f16_sdwa v20, v11, v11 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v15, v15, v15
+; VI-NEXT: v_max_f16_e32 v14, v14, v14
+; VI-NEXT: v_max_f16_e32 v13, v13, v13
+; VI-NEXT: v_max_f16_e32 v12, v12, v12
+; VI-NEXT: v_max_f16_e32 v11, v11, v11
+; VI-NEXT: v_or_b32_e32 v11, v11, v20
+; VI-NEXT: v_or_b32_e32 v12, v12, v19
+; VI-NEXT: v_or_b32_e32 v13, v13, v18
+; VI-NEXT: v_or_b32_e32 v14, v14, v17
+; VI-NEXT: v_or_b32_e32 v15, v15, v16
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v32f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
+; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
+; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
+; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
+; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
+; GFX9-NEXT: v_pk_max_f16 v8, v8, v8
+; GFX9-NEXT: v_pk_max_f16 v9, v9, v9
+; GFX9-NEXT: v_pk_max_f16 v10, v10, v10
+; GFX9-NEXT: v_pk_max_f16 v11, v11, v11
+; GFX9-NEXT: v_pk_max_f16 v12, v12, v12
+; GFX9-NEXT: v_pk_max_f16 v13, v13, v13
+; GFX9-NEXT: v_pk_max_f16 v14, v14, v14
+; GFX9-NEXT: v_pk_max_f16 v15, v15, v15
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v32f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: buffer_load_dword v31, off, s[0:3], s32
+; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v7
+; CI-NEXT: v_cvt_f16_f32_e32 v8, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v9, v9
+; CI-NEXT: v_cvt_f16_f32_e32 v10, v10
+; CI-NEXT: v_cvt_f16_f32_e32 v11, v11
+; CI-NEXT: v_cvt_f16_f32_e32 v12, v12
+; CI-NEXT: v_cvt_f16_f32_e32 v13, v13
+; CI-NEXT: v_cvt_f16_f32_e32 v14, v14
+; CI-NEXT: v_cvt_f16_f32_e32 v15, v15
+; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
+; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
+; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
+; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
+; CI-NEXT: v_cvt_f16_f32_e32 v20, v20
+; CI-NEXT: v_cvt_f16_f32_e32 v21, v21
+; CI-NEXT: v_cvt_f16_f32_e32 v22, v22
+; CI-NEXT: v_cvt_f16_f32_e32 v23, v23
+; CI-NEXT: v_cvt_f16_f32_e32 v24, v24
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v25
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
+; CI-NEXT: v_cvt_f16_f32_e32 v28, v28
+; CI-NEXT: v_cvt_f16_f32_e32 v29, v29
+; CI-NEXT: v_cvt_f16_f32_e32 v30, v30
+; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
+; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
+; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
+; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
+; CI-NEXT: v_cvt_f32_f16_e32 v4, v4
+; CI-NEXT: v_cvt_f32_f16_e32 v5, v5
+; CI-NEXT: v_cvt_f32_f16_e32 v6, v6
+; CI-NEXT: v_cvt_f32_f16_e32 v7, v7
+; CI-NEXT: v_cvt_f32_f16_e32 v8, v8
+; CI-NEXT: v_cvt_f32_f16_e32 v9, v9
+; CI-NEXT: v_cvt_f32_f16_e32 v10, v10
+; CI-NEXT: v_cvt_f32_f16_e32 v11, v11
+; CI-NEXT: v_cvt_f32_f16_e32 v12, v12
+; CI-NEXT: v_cvt_f32_f16_e32 v13, v13
+; CI-NEXT: v_cvt_f32_f16_e32 v14, v14
+; CI-NEXT: v_cvt_f32_f16_e32 v15, v15
+; CI-NEXT: v_cvt_f32_f16_e32 v16, v16
+; CI-NEXT: v_cvt_f32_f16_e32 v17, v17
+; CI-NEXT: v_cvt_f32_f16_e32 v18, v18
+; CI-NEXT: v_cvt_f32_f16_e32 v19, v19
+; CI-NEXT: v_cvt_f32_f16_e32 v20, v20
+; CI-NEXT: v_cvt_f32_f16_e32 v21, v21
+; CI-NEXT: v_cvt_f32_f16_e32 v22, v22
+; CI-NEXT: v_cvt_f32_f16_e32 v23, v23
+; CI-NEXT: v_cvt_f32_f16_e32 v24, v24
+; CI-NEXT: v_cvt_f32_f16_e32 v25, v25
+; CI-NEXT: v_cvt_f32_f16_e32 v26, v26
+; CI-NEXT: v_cvt_f32_f16_e32 v27, v27
+; CI-NEXT: v_cvt_f32_f16_e32 v28, v28
+; CI-NEXT: v_cvt_f32_f16_e32 v29, v29
+; CI-NEXT: v_cvt_f32_f16_e32 v30, v30
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v31, v31
+; CI-NEXT: v_cvt_f32_f16_e32 v31, v31
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <32 x half> @llvm.canonicalize.v32f16(<32 x half> %val)
+ ret <32 x half> %canonicalized
+}
+
+define <64 x half> @v_test_canonicalize_var_v64f16(<64 x half> %val) #1 {
+; VI-LABEL: v_test_canonicalize_var_v64f16:
+; VI: ; %bb.0:
+; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v31, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v0, v0, v0
+; VI-NEXT: v_or_b32_e32 v0, v0, v31
+; VI-NEXT: v_max_f16_sdwa v31, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v1, v1, v1
+; VI-NEXT: v_or_b32_e32 v1, v1, v31
+; VI-NEXT: v_max_f16_sdwa v31, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v2, v2, v2
+; VI-NEXT: v_or_b32_e32 v2, v2, v31
+; VI-NEXT: v_max_f16_sdwa v31, v3, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v3, v3, v3
+; VI-NEXT: v_or_b32_e32 v3, v3, v31
+; VI-NEXT: v_max_f16_sdwa v31, v4, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v4, v4, v4
+; VI-NEXT: v_or_b32_e32 v4, v4, v31
+; VI-NEXT: v_max_f16_sdwa v31, v5, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v5, v5, v5
+; VI-NEXT: v_or_b32_e32 v5, v5, v31
+; VI-NEXT: v_max_f16_sdwa v31, v6, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v6, v6, v6
+; VI-NEXT: v_or_b32_e32 v6, v6, v31
+; VI-NEXT: v_max_f16_sdwa v31, v7, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v7, v7, v7
+; VI-NEXT: v_or_b32_e32 v7, v7, v31
+; VI-NEXT: v_max_f16_sdwa v31, v8, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v8, v8, v8
+; VI-NEXT: v_or_b32_e32 v8, v8, v31
+; VI-NEXT: v_max_f16_sdwa v31, v9, v9 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v9, v9, v9
+; VI-NEXT: v_or_b32_e32 v9, v9, v31
+; VI-NEXT: v_max_f16_sdwa v31, v10, v10 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v10, v10, v10
+; VI-NEXT: v_or_b32_e32 v10, v10, v31
+; VI-NEXT: v_max_f16_sdwa v31, v11, v11 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v11, v11, v11
+; VI-NEXT: v_or_b32_e32 v11, v11, v31
+; VI-NEXT: v_max_f16_sdwa v31, v12, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v12, v12, v12
+; VI-NEXT: v_or_b32_e32 v12, v12, v31
+; VI-NEXT: v_max_f16_sdwa v31, v13, v13 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v13, v13, v13
+; VI-NEXT: v_or_b32_e32 v13, v13, v31
+; VI-NEXT: v_max_f16_sdwa v31, v14, v14 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v14, v14, v14
+; VI-NEXT: v_or_b32_e32 v14, v14, v31
+; VI-NEXT: v_max_f16_sdwa v31, v15, v15 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v15, v15, v15
+; VI-NEXT: v_or_b32_e32 v15, v15, v31
+; VI-NEXT: v_max_f16_sdwa v31, v16, v16 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v16, v16, v16
+; VI-NEXT: v_or_b32_e32 v16, v16, v31
+; VI-NEXT: v_max_f16_sdwa v31, v17, v17 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v17, v17, v17
+; VI-NEXT: v_or_b32_e32 v17, v17, v31
+; VI-NEXT: v_max_f16_sdwa v31, v18, v18 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v18, v18, v18
+; VI-NEXT: v_or_b32_e32 v18, v18, v31
+; VI-NEXT: v_max_f16_sdwa v31, v19, v19 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v19, v19, v19
+; VI-NEXT: v_or_b32_e32 v19, v19, v31
+; VI-NEXT: v_max_f16_sdwa v31, v20, v20 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v20, v20, v20
+; VI-NEXT: v_or_b32_e32 v20, v20, v31
+; VI-NEXT: v_max_f16_sdwa v31, v21, v21 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v21, v21, v21
+; VI-NEXT: v_or_b32_e32 v21, v21, v31
+; VI-NEXT: v_max_f16_sdwa v31, v22, v22 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v22, v22, v22
+; VI-NEXT: v_or_b32_e32 v22, v22, v31
+; VI-NEXT: v_max_f16_sdwa v31, v23, v23 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v23, v23, v23
+; VI-NEXT: v_or_b32_e32 v23, v23, v31
+; VI-NEXT: v_max_f16_sdwa v31, v24, v24 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v24, v24, v24
+; VI-NEXT: v_or_b32_e32 v24, v24, v31
+; VI-NEXT: v_max_f16_sdwa v31, v25, v25 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v25, v25, v25
+; VI-NEXT: v_or_b32_e32 v25, v25, v31
+; VI-NEXT: v_max_f16_sdwa v31, v26, v26 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v26, v26, v26
+; VI-NEXT: v_or_b32_e32 v26, v26, v31
+; VI-NEXT: v_max_f16_sdwa v31, v27, v27 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v27, v27, v27
+; VI-NEXT: v_or_b32_e32 v27, v27, v31
+; VI-NEXT: v_max_f16_sdwa v31, v28, v28 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v28, v28, v28
+; VI-NEXT: v_or_b32_e32 v28, v28, v31
+; VI-NEXT: v_max_f16_sdwa v31, v29, v29 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v29, v29, v29
+; VI-NEXT: v_or_b32_e32 v29, v29, v31
+; VI-NEXT: v_max_f16_sdwa v31, v30, v30 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v30, v30, v30
+; VI-NEXT: v_or_b32_e32 v30, v30, v31
+; VI-NEXT: buffer_load_dword v31, off, s[0:3], s32
+; VI-NEXT: s_waitcnt vmcnt(0)
+; VI-NEXT: v_max_f16_sdwa v32, v31, v31 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT: v_max_f16_e32 v31, v31, v31
+; VI-NEXT: v_or_b32_e32 v31, v31, v32
+; VI-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_test_canonicalize_var_v64f16:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: buffer_load_dword v31, off, s[0:3], s32
+; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
+; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
+; GFX9-NEXT: v_pk_max_f16 v2, v2, v2
+; GFX9-NEXT: v_pk_max_f16 v3, v3, v3
+; GFX9-NEXT: v_pk_max_f16 v4, v4, v4
+; GFX9-NEXT: v_pk_max_f16 v5, v5, v5
+; GFX9-NEXT: v_pk_max_f16 v6, v6, v6
+; GFX9-NEXT: v_pk_max_f16 v7, v7, v7
+; GFX9-NEXT: v_pk_max_f16 v8, v8, v8
+; GFX9-NEXT: v_pk_max_f16 v9, v9, v9
+; GFX9-NEXT: v_pk_max_f16 v10, v10, v10
+; GFX9-NEXT: v_pk_max_f16 v11, v11, v11
+; GFX9-NEXT: v_pk_max_f16 v12, v12, v12
+; GFX9-NEXT: v_pk_max_f16 v13, v13, v13
+; GFX9-NEXT: v_pk_max_f16 v14, v14, v14
+; GFX9-NEXT: v_pk_max_f16 v15, v15, v15
+; GFX9-NEXT: v_pk_max_f16 v16, v16, v16
+; GFX9-NEXT: v_pk_max_f16 v17, v17, v17
+; GFX9-NEXT: v_pk_max_f16 v18, v18, v18
+; GFX9-NEXT: v_pk_max_f16 v19, v19, v19
+; GFX9-NEXT: v_pk_max_f16 v20, v20, v20
+; GFX9-NEXT: v_pk_max_f16 v21, v21, v21
+; GFX9-NEXT: v_pk_max_f16 v22, v22, v22
+; GFX9-NEXT: v_pk_max_f16 v23, v23, v23
+; GFX9-NEXT: v_pk_max_f16 v24, v24, v24
+; GFX9-NEXT: v_pk_max_f16 v25, v25, v25
+; GFX9-NEXT: v_pk_max_f16 v26, v26, v26
+; GFX9-NEXT: v_pk_max_f16 v27, v27, v27
+; GFX9-NEXT: v_pk_max_f16 v28, v28, v28
+; GFX9-NEXT: v_pk_max_f16 v29, v29, v29
+; GFX9-NEXT: v_pk_max_f16 v30, v30, v30
+; GFX9-NEXT: s_waitcnt vmcnt(0)
+; GFX9-NEXT: v_pk_max_f16 v31, v31, v31
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; CI-LABEL: v_test_canonicalize_var_v64f16:
+; CI: ; %bb.0:
+; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
+; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
+; CI-NEXT: v_or_b32_e32 v1, v1, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v2, v4
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v5, v7
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v9
+; CI-NEXT: v_lshlrev_b32_e32 v2, 16, v2
+; CI-NEXT: v_or_b32_e32 v2, v3, v2
+; CI-NEXT: v_cvt_f16_f32_e32 v3, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v10
+; CI-NEXT: v_cvt_f16_f32_e32 v9, v13
+; CI-NEXT: v_cvt_f16_f32_e32 v10, v18
+; CI-NEXT: v_lshlrev_b32_e32 v3, 16, v3
+; CI-NEXT: v_or_b32_e32 v3, v4, v3
+; CI-NEXT: v_cvt_f16_f32_e32 v4, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v8, v14
+; CI-NEXT: v_cvt_f16_f32_e32 v13, v21
+; CI-NEXT: v_cvt_f16_f32_e32 v14, v26
+; CI-NEXT: v_lshlrev_b32_e32 v4, 16, v4
+; CI-NEXT: v_or_b32_e32 v4, v5, v4
+; CI-NEXT: v_lshlrev_b32_e32 v5, 16, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v6, v12
+; CI-NEXT: v_or_b32_e32 v5, v7, v5
+; CI-NEXT: v_cvt_f16_f32_e32 v7, v11
+; CI-NEXT: v_cvt_f16_f32_e32 v11, v17
+; CI-NEXT: v_lshlrev_b32_e32 v6, 16, v6
+; CI-NEXT: v_cvt_f16_f32_e32 v12, v22
+; CI-NEXT: v_or_b32_e32 v6, v7, v6
+; CI-NEXT: v_lshlrev_b32_e32 v7, 16, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v8, v16
+; CI-NEXT: v_or_b32_e32 v7, v9, v7
+; CI-NEXT: v_cvt_f16_f32_e32 v9, v15
+; CI-NEXT: v_cvt_f16_f32_e32 v15, v25
+; CI-NEXT: v_lshlrev_b32_e32 v8, 16, v8
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v29
+; CI-NEXT: v_or_b32_e32 v8, v9, v8
+; CI-NEXT: v_lshlrev_b32_e32 v9, 16, v10
+; CI-NEXT: v_cvt_f16_f32_e32 v10, v20
+; CI-NEXT: v_or_b32_e32 v9, v11, v9
+; CI-NEXT: v_cvt_f16_f32_e32 v11, v19
+; CI-NEXT: buffer_load_dword v16, off, s[0:3], s32 offset:4
+; CI-NEXT: buffer_load_dword v17, off, s[0:3], s32
+; CI-NEXT: buffer_load_dword v18, off, s[0:3], s32 offset:12
+; CI-NEXT: buffer_load_dword v19, off, s[0:3], s32 offset:8
+; CI-NEXT: v_lshlrev_b32_e32 v10, 16, v10
+; CI-NEXT: v_or_b32_e32 v10, v11, v10
+; CI-NEXT: v_lshlrev_b32_e32 v11, 16, v12
+; CI-NEXT: v_cvt_f16_f32_e32 v12, v24
+; CI-NEXT: v_or_b32_e32 v11, v13, v11
+; CI-NEXT: v_cvt_f16_f32_e32 v13, v23
+; CI-NEXT: buffer_load_dword v20, off, s[0:3], s32 offset:20
+; CI-NEXT: buffer_load_dword v21, off, s[0:3], s32 offset:16
+; CI-NEXT: buffer_load_dword v22, off, s[0:3], s32 offset:28
+; CI-NEXT: buffer_load_dword v23, off, s[0:3], s32 offset:24
+; CI-NEXT: v_lshlrev_b32_e32 v12, 16, v12
+; CI-NEXT: v_cvt_f16_f32_e32 v24, v30
+; CI-NEXT: v_or_b32_e32 v12, v13, v12
+; CI-NEXT: v_lshlrev_b32_e32 v13, 16, v14
+; CI-NEXT: v_or_b32_e32 v13, v15, v13
+; CI-NEXT: v_cvt_f16_f32_e32 v14, v28
+; CI-NEXT: v_cvt_f16_f32_e32 v15, v27
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:36
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:32
+; CI-NEXT: buffer_load_dword v28, off, s[0:3], s32 offset:44
+; CI-NEXT: buffer_load_dword v29, off, s[0:3], s32 offset:40
+; CI-NEXT: v_lshlrev_b32_e32 v14, 16, v14
+; CI-NEXT: v_or_b32_e32 v14, v15, v14
+; CI-NEXT: v_lshlrev_b32_e32 v15, 16, v24
+; CI-NEXT: v_or_b32_e32 v15, v25, v15
+; CI-NEXT: s_waitcnt vmcnt(11)
+; CI-NEXT: v_cvt_f16_f32_e32 v16, v16
+; CI-NEXT: s_waitcnt vmcnt(10)
+; CI-NEXT: v_cvt_f16_f32_e32 v17, v17
+; CI-NEXT: s_waitcnt vmcnt(9)
+; CI-NEXT: v_cvt_f16_f32_e32 v18, v18
+; CI-NEXT: s_waitcnt vmcnt(8)
+; CI-NEXT: v_cvt_f16_f32_e32 v19, v19
+; CI-NEXT: v_lshlrev_b32_e32 v16, 16, v16
+; CI-NEXT: v_or_b32_e32 v16, v17, v16
+; CI-NEXT: v_lshlrev_b32_e32 v17, 16, v18
+; CI-NEXT: v_or_b32_e32 v17, v19, v17
+; CI-NEXT: s_waitcnt vmcnt(7)
+; CI-NEXT: v_cvt_f16_f32_e32 v18, v20
+; CI-NEXT: s_waitcnt vmcnt(6)
+; CI-NEXT: v_cvt_f16_f32_e32 v19, v21
+; CI-NEXT: s_waitcnt vmcnt(5)
+; CI-NEXT: v_cvt_f16_f32_e32 v20, v22
+; CI-NEXT: s_waitcnt vmcnt(4)
+; CI-NEXT: v_cvt_f16_f32_e32 v21, v23
+; CI-NEXT: v_lshlrev_b32_e32 v18, 16, v18
+; CI-NEXT: v_or_b32_e32 v18, v19, v18
+; CI-NEXT: v_lshlrev_b32_e32 v19, 16, v20
+; CI-NEXT: v_or_b32_e32 v19, v21, v19
+; CI-NEXT: s_waitcnt vmcnt(3)
+; CI-NEXT: v_cvt_f16_f32_e32 v20, v26
+; CI-NEXT: s_waitcnt vmcnt(2)
+; CI-NEXT: v_cvt_f16_f32_e32 v21, v27
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v28
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v29
+; CI-NEXT: v_lshlrev_b32_e32 v20, 16, v20
+; CI-NEXT: v_or_b32_e32 v20, v21, v20
+; CI-NEXT: v_lshlrev_b32_e32 v21, 16, v26
+; CI-NEXT: buffer_load_dword v24, off, s[0:3], s32 offset:52
+; CI-NEXT: buffer_load_dword v25, off, s[0:3], s32 offset:48
+; CI-NEXT: buffer_load_dword v23, off, s[0:3], s32 offset:60
+; CI-NEXT: buffer_load_dword v22, off, s[0:3], s32 offset:56
+; CI-NEXT: v_or_b32_e32 v21, v27, v21
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:132
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:128
+; CI-NEXT: s_waitcnt vmcnt(5)
+; CI-NEXT: v_cvt_f16_f32_e32 v24, v24
+; CI-NEXT: s_waitcnt vmcnt(4)
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v25
+; CI-NEXT: s_waitcnt vmcnt(3)
+; CI-NEXT: v_cvt_f16_f32_e32 v23, v23
+; CI-NEXT: s_waitcnt vmcnt(2)
+; CI-NEXT: v_cvt_f16_f32_e32 v22, v22
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
+; CI-NEXT: v_lshlrev_b32_e32 v24, 16, v24
+; CI-NEXT: v_or_b32_e32 v24, v25, v24
+; CI-NEXT: v_lshlrev_b32_e32 v26, 16, v26
+; CI-NEXT: v_or_b32_e32 v26, v27, v26
+; CI-NEXT: v_add_i32_e32 v27, vcc, 0x7c, v0
+; CI-NEXT: buffer_store_dword v26, v27, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:124
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:120
+; CI-NEXT: v_lshlrev_b32_e32 v23, 16, v23
+; CI-NEXT: v_or_b32_e32 v22, v22, v23
+; CI-NEXT: buffer_load_dword v23, off, s[0:3], s32 offset:88
+; CI-NEXT: s_waitcnt vmcnt(2)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
+; CI-NEXT: v_lshlrev_b32_e32 v26, 16, v26
+; CI-NEXT: v_or_b32_e32 v26, v27, v26
+; CI-NEXT: v_add_i32_e32 v27, vcc, 0x78, v0
+; CI-NEXT: buffer_store_dword v26, v27, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:116
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:112
+; CI-NEXT: s_waitcnt vmcnt(3)
+; CI-NEXT: v_cvt_f16_f32_e32 v23, v23
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
+; CI-NEXT: v_lshlrev_b32_e32 v26, 16, v26
+; CI-NEXT: v_or_b32_e32 v26, v27, v26
+; CI-NEXT: v_add_i32_e32 v27, vcc, 0x74, v0
+; CI-NEXT: buffer_store_dword v26, v27, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:108
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:104
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v26
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v27
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:92
+; CI-NEXT: v_lshlrev_b32_e32 v25, 16, v25
+; CI-NEXT: v_or_b32_e32 v25, v26, v25
+; CI-NEXT: v_add_i32_e32 v26, vcc, 0x70, v0
+; CI-NEXT: buffer_store_dword v25, v26, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v25, off, s[0:3], s32 offset:100
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:96
+; CI-NEXT: s_waitcnt vmcnt(3)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v27
+; CI-NEXT: v_lshlrev_b32_e32 v27, 16, v27
+; CI-NEXT: v_or_b32_e32 v23, v23, v27
+; CI-NEXT: s_waitcnt vmcnt(1)
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v25
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: v_add_i32_e32 v27, vcc, 0x68, v0
+; CI-NEXT: v_lshlrev_b32_e32 v25, 16, v25
+; CI-NEXT: v_or_b32_e32 v25, v26, v25
+; CI-NEXT: v_add_i32_e32 v26, vcc, 0x6c, v0
+; CI-NEXT: buffer_store_dword v25, v26, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v25, off, s[0:3], s32 offset:68
+; CI-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:64
+; CI-NEXT: buffer_store_dword v23, v27, s[0:3], 0 offen
+; CI-NEXT: buffer_load_dword v23, off, s[0:3], s32 offset:76
+; CI-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:72
+; CI-NEXT: buffer_load_dword v28, off, s[0:3], s32 offset:84
+; CI-NEXT: buffer_load_dword v29, off, s[0:3], s32 offset:80
+; CI-NEXT: s_waitcnt vmcnt(3)
+; CI-NEXT: v_cvt_f16_f32_e32 v23, v23
+; CI-NEXT: v_cvt_f16_f32_e32 v25, v25
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v26
+; CI-NEXT: v_lshlrev_b32_e32 v23, 16, v23
+; CI-NEXT: v_lshlrev_b32_e32 v25, 16, v25
+; CI-NEXT: v_or_b32_e32 v25, v26, v25
+; CI-NEXT: s_waitcnt vmcnt(2)
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v27
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: v_cvt_f16_f32_e32 v27, v29
+; CI-NEXT: v_or_b32_e32 v23, v26, v23
+; CI-NEXT: v_cvt_f16_f32_e32 v26, v28
+; CI-NEXT: v_lshlrev_b32_e32 v26, 16, v26
+; CI-NEXT: v_or_b32_e32 v26, v27, v26
+; CI-NEXT: v_add_i32_e32 v27, vcc, 0x64, v0
+; CI-NEXT: buffer_store_dword v26, v27, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v26, vcc, 0x60, v0
+; CI-NEXT: buffer_store_dword v23, v26, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v23, vcc, 0x5c, v0
+; CI-NEXT: buffer_store_dword v25, v23, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v23, vcc, 0x58, v0
+; CI-NEXT: buffer_store_dword v22, v23, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v22, vcc, 0x54, v0
+; CI-NEXT: buffer_store_dword v24, v22, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v22, vcc, 0x50, v0
+; CI-NEXT: buffer_store_dword v21, v22, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v21, vcc, 0x4c, v0
+; CI-NEXT: buffer_store_dword v20, v21, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v20, vcc, 0x48, v0
+; CI-NEXT: buffer_store_dword v19, v20, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v19, vcc, 0x44, v0
+; CI-NEXT: buffer_store_dword v18, v19, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v18, vcc, 64, v0
+; CI-NEXT: buffer_store_dword v17, v18, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v17, vcc, 60, v0
+; CI-NEXT: buffer_store_dword v16, v17, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v16, vcc, 56, v0
+; CI-NEXT: buffer_store_dword v15, v16, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v15, vcc, 52, v0
+; CI-NEXT: buffer_store_dword v14, v15, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v14, vcc, 48, v0
+; CI-NEXT: buffer_store_dword v13, v14, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v13, vcc, 44, v0
+; CI-NEXT: buffer_store_dword v12, v13, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v12, vcc, 40, v0
+; CI-NEXT: buffer_store_dword v11, v12, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v11, vcc, 36, v0
+; CI-NEXT: buffer_store_dword v10, v11, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v10, vcc, 32, v0
+; CI-NEXT: buffer_store_dword v9, v10, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v9, vcc, 28, v0
+; CI-NEXT: buffer_store_dword v8, v9, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v8, vcc, 24, v0
+; CI-NEXT: buffer_store_dword v7, v8, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v7, vcc, 20, v0
+; CI-NEXT: buffer_store_dword v6, v7, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v6, vcc, 16, v0
+; CI-NEXT: buffer_store_dword v5, v6, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v5, vcc, 12, v0
+; CI-NEXT: buffer_store_dword v4, v5, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v4, vcc, 8, v0
+; CI-NEXT: buffer_store_dword v3, v4, s[0:3], 0 offen
+; CI-NEXT: v_add_i32_e32 v3, vcc, 4, v0
+; CI-NEXT: buffer_store_dword v2, v3, s[0:3], 0 offen
+; CI-NEXT: buffer_store_dword v1, v0, s[0:3], 0 offen
+; CI-NEXT: s_waitcnt vmcnt(0)
+; CI-NEXT: s_setpc_b64 s[30:31]
+ %canonicalized = call <64 x half> @llvm.canonicalize.v64f16(<64 x half> %val)
+ ret <64 x half> %canonicalized
+}
+
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
attributes #2 = { nounwind "denormal-fp-math"="preserve-sign,preserve-sign" }
More information about the llvm-commits
mailing list