[llvm] 8e17c84 - [AMDGPU][ISel] Set trunc store action to expand for v4f32->v4bf16 (#90427)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 29 06:08:57 PDT 2024
Author: Shilei Tian
Date: 2024-04-29T09:08:54-04:00
New Revision: 8e17c84836b08be9a23d76c2cc234777712347de
URL: https://github.com/llvm/llvm-project/commit/8e17c84836b08be9a23d76c2cc234777712347de
DIFF: https://github.com/llvm/llvm-project/commit/8e17c84836b08be9a23d76c2cc234777712347de.diff
LOG: [AMDGPU][ISel] Set trunc store action to expand for v4f32->v4bf16 (#90427)
Added:
llvm/test/CodeGen/AMDGPU/fp_trunc_store_bf16.ll
Modified:
llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 7993b63121110c..7f4a2437f62eb4 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -313,11 +313,16 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(const TargetMachine &TM,
setTruncStoreAction(MVT::f32, MVT::bf16, Expand);
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
+ setTruncStoreAction(MVT::v2f32, MVT::v2bf16, Expand);
setTruncStoreAction(MVT::v2f32, MVT::v2f16, Expand);
setTruncStoreAction(MVT::v3f32, MVT::v3f16, Expand);
+ setTruncStoreAction(MVT::v4f32, MVT::v4bf16, Expand);
setTruncStoreAction(MVT::v4f32, MVT::v4f16, Expand);
+ setTruncStoreAction(MVT::v8f32, MVT::v8bf16, Expand);
setTruncStoreAction(MVT::v8f32, MVT::v8f16, Expand);
+ setTruncStoreAction(MVT::v16f32, MVT::v16bf16, Expand);
setTruncStoreAction(MVT::v16f32, MVT::v16f16, Expand);
+ setTruncStoreAction(MVT::v32f32, MVT::v32bf16, Expand);
setTruncStoreAction(MVT::v32f32, MVT::v32f16, Expand);
setTruncStoreAction(MVT::f64, MVT::bf16, Expand);
diff --git a/llvm/test/CodeGen/AMDGPU/fp_trunc_store_bf16.ll b/llvm/test/CodeGen/AMDGPU/fp_trunc_store_bf16.ll
new file mode 100644
index 00000000000000..5aaff773689f96
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fp_trunc_store_bf16.ll
@@ -0,0 +1,426 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a %s -o - | FileCheck %s
+
+define void @v2(<2 x float> %num, ptr addrspace(1) %p) {
+; CHECK-LABEL: v2:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_bfe_u32 v4, v0, 16, 1
+; CHECK-NEXT: s_movk_i32 s4, 0x7fff
+; CHECK-NEXT: v_add3_u32 v4, v4, v0, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v0
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v5, vcc
+; CHECK-NEXT: v_bfe_u32 v4, v1, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v1, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v1
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v4, v5, vcc
+; CHECK-NEXT: s_mov_b32 s4, 0x7060302
+; CHECK-NEXT: v_perm_b32 v0, v1, v0, s4
+; CHECK-NEXT: global_store_dword v[2:3], v0, off
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+entry:
+ %conv = fptrunc <2 x float> %num to <2 x bfloat>
+ store <2 x bfloat> %conv, ptr addrspace(1) %p, align 8
+ ret void
+}
+
+define void @v4(<4 x float> %num, ptr addrspace(1) %p) {
+; CHECK-LABEL: v4:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_bfe_u32 v6, v2, 16, 1
+; CHECK-NEXT: s_movk_i32 s4, 0x7fff
+; CHECK-NEXT: v_add3_u32 v6, v6, v2, s4
+; CHECK-NEXT: v_or_b32_e32 v7, 0x400000, v2
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v2, v2
+; CHECK-NEXT: v_cndmask_b32_e32 v2, v6, v7, vcc
+; CHECK-NEXT: v_bfe_u32 v6, v3, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v3, s4
+; CHECK-NEXT: v_or_b32_e32 v7, 0x400000, v3
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v6, v7, vcc
+; CHECK-NEXT: s_mov_b32 s5, 0x7060302
+; CHECK-NEXT: v_perm_b32 v3, v3, v2, s5
+; CHECK-NEXT: v_bfe_u32 v2, v0, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v0, s4
+; CHECK-NEXT: v_or_b32_e32 v6, 0x400000, v0
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v6, vcc
+; CHECK-NEXT: v_bfe_u32 v2, v1, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v1, s4
+; CHECK-NEXT: v_or_b32_e32 v6, 0x400000, v1
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v2, v6, vcc
+; CHECK-NEXT: v_perm_b32 v2, v1, v0, s5
+; CHECK-NEXT: global_store_dwordx2 v[4:5], v[2:3], off
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+entry:
+ %conv = fptrunc <4 x float> %num to <4 x bfloat>
+ store <4 x bfloat> %conv, ptr addrspace(1) %p, align 8
+ ret void
+}
+
+define void @v8(<8 x float> %num, ptr addrspace(1) %p) {
+; CHECK-LABEL: v8:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_bfe_u32 v10, v6, 16, 1
+; CHECK-NEXT: s_movk_i32 s4, 0x7fff
+; CHECK-NEXT: v_add3_u32 v10, v10, v6, s4
+; CHECK-NEXT: v_or_b32_e32 v11, 0x400000, v6
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v6, v6
+; CHECK-NEXT: v_cndmask_b32_e32 v6, v10, v11, vcc
+; CHECK-NEXT: v_bfe_u32 v10, v7, 16, 1
+; CHECK-NEXT: v_add3_u32 v10, v10, v7, s4
+; CHECK-NEXT: v_or_b32_e32 v11, 0x400000, v7
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v7, v7
+; CHECK-NEXT: v_cndmask_b32_e32 v7, v10, v11, vcc
+; CHECK-NEXT: s_mov_b32 s5, 0x7060302
+; CHECK-NEXT: v_perm_b32 v7, v7, v6, s5
+; CHECK-NEXT: v_bfe_u32 v6, v4, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v4, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v4
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
+; CHECK-NEXT: v_cndmask_b32_e32 v4, v6, v10, vcc
+; CHECK-NEXT: v_bfe_u32 v6, v5, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v5, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v5
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v5, v5
+; CHECK-NEXT: v_cndmask_b32_e32 v5, v6, v10, vcc
+; CHECK-NEXT: v_perm_b32 v6, v5, v4, s5
+; CHECK-NEXT: v_bfe_u32 v4, v2, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v2, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v2
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v2, v2
+; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc
+; CHECK-NEXT: v_bfe_u32 v4, v3, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v3, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v3
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v4, v5, vcc
+; CHECK-NEXT: v_perm_b32 v5, v3, v2, s5
+; CHECK-NEXT: v_bfe_u32 v2, v0, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v0, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v0
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; CHECK-NEXT: v_bfe_u32 v2, v1, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v1, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v1
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
+; CHECK-NEXT: v_perm_b32 v4, v1, v0, s5
+; CHECK-NEXT: global_store_dwordx4 v[8:9], v[4:7], off
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+entry:
+ %conv = fptrunc <8 x float> %num to <8 x bfloat>
+ store <8 x bfloat> %conv, ptr addrspace(1) %p, align 8
+ ret void
+}
+
+define void @v16(<16 x float> %num, ptr addrspace(1) %p) {
+; CHECK-LABEL: v16:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_bfe_u32 v18, v6, 16, 1
+; CHECK-NEXT: s_movk_i32 s4, 0x7fff
+; CHECK-NEXT: v_add3_u32 v18, v18, v6, s4
+; CHECK-NEXT: v_or_b32_e32 v19, 0x400000, v6
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v6, v6
+; CHECK-NEXT: v_cndmask_b32_e32 v6, v18, v19, vcc
+; CHECK-NEXT: v_bfe_u32 v18, v7, 16, 1
+; CHECK-NEXT: v_add3_u32 v18, v18, v7, s4
+; CHECK-NEXT: v_or_b32_e32 v19, 0x400000, v7
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v7, v7
+; CHECK-NEXT: v_cndmask_b32_e32 v7, v18, v19, vcc
+; CHECK-NEXT: s_mov_b32 s5, 0x7060302
+; CHECK-NEXT: v_perm_b32 v7, v7, v6, s5
+; CHECK-NEXT: v_bfe_u32 v6, v4, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v4, s4
+; CHECK-NEXT: v_or_b32_e32 v18, 0x400000, v4
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
+; CHECK-NEXT: v_cndmask_b32_e32 v4, v6, v18, vcc
+; CHECK-NEXT: v_bfe_u32 v6, v5, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v5, s4
+; CHECK-NEXT: v_or_b32_e32 v18, 0x400000, v5
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v5, v5
+; CHECK-NEXT: v_cndmask_b32_e32 v5, v6, v18, vcc
+; CHECK-NEXT: v_perm_b32 v6, v5, v4, s5
+; CHECK-NEXT: v_bfe_u32 v4, v2, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v2, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v2
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v2, v2
+; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc
+; CHECK-NEXT: v_bfe_u32 v4, v3, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v3, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v3
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v4, v5, vcc
+; CHECK-NEXT: v_perm_b32 v5, v3, v2, s5
+; CHECK-NEXT: v_bfe_u32 v2, v0, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v0, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v0
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; CHECK-NEXT: v_bfe_u32 v2, v1, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v1, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v1
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
+; CHECK-NEXT: v_perm_b32 v4, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v14, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v14, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v14
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v14, v14
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v15, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v15, s4
+; CHECK-NEXT: v_or_b32_e32 v2, 0x400000, v15
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v15, v15
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
+; CHECK-NEXT: v_perm_b32 v3, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v12, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v12, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v12
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v12, v12
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v13, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v13, s4
+; CHECK-NEXT: v_or_b32_e32 v2, 0x400000, v13
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v13, v13
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
+; CHECK-NEXT: v_perm_b32 v2, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v10, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v10, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v10
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v10, v10
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v11, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v11, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v11
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v11, v11
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v10, vcc
+; CHECK-NEXT: v_perm_b32 v1, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v8, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v8, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v8
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v8, v8
+; CHECK-NEXT: v_bfe_u32 v8, v9, 16, 1
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v10, vcc
+; CHECK-NEXT: v_add3_u32 v8, v8, v9, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v9
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v9, v9
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
+; CHECK-NEXT: v_perm_b32 v0, v8, v0, s5
+; CHECK-NEXT: global_store_dwordx4 v[16:17], v[0:3], off offset:16
+; CHECK-NEXT: global_store_dwordx4 v[16:17], v[4:7], off
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+entry:
+ %conv = fptrunc <16 x float> %num to <16 x bfloat>
+ store <16 x bfloat> %conv, ptr addrspace(1) %p, align 8
+ ret void
+}
+
+define void @v32(<32 x float> %num, ptr addrspace(1) %p) {
+; CHECK-LABEL: v32:
+; CHECK: ; %bb.0: ; %entry
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: buffer_load_dword v33, off, s[0:3], s32 offset:8
+; CHECK-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:4
+; CHECK-NEXT: buffer_load_dword v31, off, s[0:3], s32
+; CHECK-NEXT: v_bfe_u32 v34, v6, 16, 1
+; CHECK-NEXT: s_movk_i32 s4, 0x7fff
+; CHECK-NEXT: v_add3_u32 v34, v34, v6, s4
+; CHECK-NEXT: v_or_b32_e32 v35, 0x400000, v6
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v6, v6
+; CHECK-NEXT: v_cndmask_b32_e32 v6, v34, v35, vcc
+; CHECK-NEXT: v_bfe_u32 v34, v7, 16, 1
+; CHECK-NEXT: v_add3_u32 v34, v34, v7, s4
+; CHECK-NEXT: v_or_b32_e32 v35, 0x400000, v7
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v7, v7
+; CHECK-NEXT: v_cndmask_b32_e32 v7, v34, v35, vcc
+; CHECK-NEXT: s_mov_b32 s5, 0x7060302
+; CHECK-NEXT: v_perm_b32 v7, v7, v6, s5
+; CHECK-NEXT: v_bfe_u32 v6, v4, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v4, s4
+; CHECK-NEXT: v_or_b32_e32 v34, 0x400000, v4
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v4, v4
+; CHECK-NEXT: v_cndmask_b32_e32 v4, v6, v34, vcc
+; CHECK-NEXT: v_bfe_u32 v6, v5, 16, 1
+; CHECK-NEXT: v_add3_u32 v6, v6, v5, s4
+; CHECK-NEXT: v_or_b32_e32 v34, 0x400000, v5
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v5, v5
+; CHECK-NEXT: v_cndmask_b32_e32 v5, v6, v34, vcc
+; CHECK-NEXT: v_perm_b32 v6, v5, v4, s5
+; CHECK-NEXT: v_bfe_u32 v4, v2, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v2, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v2
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v2, v2
+; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc
+; CHECK-NEXT: v_bfe_u32 v4, v3, 16, 1
+; CHECK-NEXT: v_add3_u32 v4, v4, v3, s4
+; CHECK-NEXT: v_or_b32_e32 v5, 0x400000, v3
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v3, v3
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v4, v5, vcc
+; CHECK-NEXT: v_perm_b32 v5, v3, v2, s5
+; CHECK-NEXT: v_bfe_u32 v2, v0, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v0, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v0
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v0, v0
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
+; CHECK-NEXT: v_bfe_u32 v2, v1, 16, 1
+; CHECK-NEXT: v_add3_u32 v2, v2, v1, s4
+; CHECK-NEXT: v_or_b32_e32 v3, 0x400000, v1
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v1, v1
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
+; CHECK-NEXT: v_perm_b32 v4, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v14, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v14, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v14
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v14, v14
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v15, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v15, s4
+; CHECK-NEXT: v_or_b32_e32 v2, 0x400000, v15
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v15, v15
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
+; CHECK-NEXT: v_perm_b32 v3, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v12, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v12, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v12
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v12, v12
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v13, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v13, s4
+; CHECK-NEXT: v_or_b32_e32 v2, 0x400000, v13
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v13, v13
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
+; CHECK-NEXT: v_perm_b32 v2, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v10, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v10, s4
+; CHECK-NEXT: v_or_b32_e32 v1, 0x400000, v10
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v10, v10
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; CHECK-NEXT: v_bfe_u32 v1, v11, 16, 1
+; CHECK-NEXT: v_add3_u32 v1, v1, v11, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v11
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v11, v11
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v10, vcc
+; CHECK-NEXT: v_perm_b32 v1, v1, v0, s5
+; CHECK-NEXT: v_bfe_u32 v0, v8, 16, 1
+; CHECK-NEXT: v_add3_u32 v0, v0, v8, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v8
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v8, v8
+; CHECK-NEXT: v_bfe_u32 v8, v9, 16, 1
+; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v10, vcc
+; CHECK-NEXT: v_add3_u32 v8, v8, v9, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v9
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v9, v9
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
+; CHECK-NEXT: v_perm_b32 v0, v8, v0, s5
+; CHECK-NEXT: v_bfe_u32 v8, v22, 16, 1
+; CHECK-NEXT: v_add3_u32 v8, v8, v22, s4
+; CHECK-NEXT: v_or_b32_e32 v9, 0x400000, v22
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v22, v22
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc
+; CHECK-NEXT: v_bfe_u32 v9, v23, 16, 1
+; CHECK-NEXT: v_add3_u32 v9, v9, v23, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v23
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v23, v23
+; CHECK-NEXT: v_cndmask_b32_e32 v9, v9, v10, vcc
+; CHECK-NEXT: v_perm_b32 v11, v9, v8, s5
+; CHECK-NEXT: v_bfe_u32 v8, v20, 16, 1
+; CHECK-NEXT: v_add3_u32 v8, v8, v20, s4
+; CHECK-NEXT: v_or_b32_e32 v9, 0x400000, v20
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v20, v20
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc
+; CHECK-NEXT: v_bfe_u32 v9, v21, 16, 1
+; CHECK-NEXT: v_add3_u32 v9, v9, v21, s4
+; CHECK-NEXT: v_or_b32_e32 v10, 0x400000, v21
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v21, v21
+; CHECK-NEXT: v_cndmask_b32_e32 v9, v9, v10, vcc
+; CHECK-NEXT: v_perm_b32 v10, v9, v8, s5
+; CHECK-NEXT: v_bfe_u32 v8, v18, 16, 1
+; CHECK-NEXT: v_add3_u32 v8, v8, v18, s4
+; CHECK-NEXT: v_or_b32_e32 v9, 0x400000, v18
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v18, v18
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc
+; CHECK-NEXT: v_bfe_u32 v9, v19, 16, 1
+; CHECK-NEXT: v_add3_u32 v9, v9, v19, s4
+; CHECK-NEXT: v_or_b32_e32 v12, 0x400000, v19
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v19, v19
+; CHECK-NEXT: v_cndmask_b32_e32 v9, v9, v12, vcc
+; CHECK-NEXT: v_perm_b32 v9, v9, v8, s5
+; CHECK-NEXT: v_bfe_u32 v8, v16, 16, 1
+; CHECK-NEXT: v_add3_u32 v8, v8, v16, s4
+; CHECK-NEXT: v_or_b32_e32 v12, 0x400000, v16
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v16, v16
+; CHECK-NEXT: v_cndmask_b32_e32 v8, v8, v12, vcc
+; CHECK-NEXT: v_bfe_u32 v12, v17, 16, 1
+; CHECK-NEXT: v_add3_u32 v12, v12, v17, s4
+; CHECK-NEXT: v_or_b32_e32 v13, 0x400000, v17
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v17, v17
+; CHECK-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc
+; CHECK-NEXT: v_perm_b32 v8, v12, v8, s5
+; CHECK-NEXT: v_bfe_u32 v12, v30, 16, 1
+; CHECK-NEXT: v_add3_u32 v12, v12, v30, s4
+; CHECK-NEXT: v_or_b32_e32 v13, 0x400000, v30
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v30, v30
+; CHECK-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: v_bfe_u32 v13, v31, 16, 1
+; CHECK-NEXT: v_add3_u32 v13, v13, v31, s4
+; CHECK-NEXT: v_or_b32_e32 v14, 0x400000, v31
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v31, v31
+; CHECK-NEXT: v_cndmask_b32_e32 v13, v13, v14, vcc
+; CHECK-NEXT: v_perm_b32 v15, v13, v12, s5
+; CHECK-NEXT: v_bfe_u32 v12, v28, 16, 1
+; CHECK-NEXT: v_add3_u32 v12, v12, v28, s4
+; CHECK-NEXT: v_or_b32_e32 v13, 0x400000, v28
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v28, v28
+; CHECK-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc
+; CHECK-NEXT: v_bfe_u32 v13, v29, 16, 1
+; CHECK-NEXT: v_add3_u32 v13, v13, v29, s4
+; CHECK-NEXT: v_or_b32_e32 v14, 0x400000, v29
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v29, v29
+; CHECK-NEXT: v_cndmask_b32_e32 v13, v13, v14, vcc
+; CHECK-NEXT: v_perm_b32 v14, v13, v12, s5
+; CHECK-NEXT: v_bfe_u32 v12, v26, 16, 1
+; CHECK-NEXT: v_add3_u32 v12, v12, v26, s4
+; CHECK-NEXT: v_or_b32_e32 v13, 0x400000, v26
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v26, v26
+; CHECK-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc
+; CHECK-NEXT: v_bfe_u32 v13, v27, 16, 1
+; CHECK-NEXT: v_add3_u32 v13, v13, v27, s4
+; CHECK-NEXT: v_or_b32_e32 v16, 0x400000, v27
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v27, v27
+; CHECK-NEXT: v_cndmask_b32_e32 v13, v13, v16, vcc
+; CHECK-NEXT: v_perm_b32 v13, v13, v12, s5
+; CHECK-NEXT: v_bfe_u32 v12, v24, 16, 1
+; CHECK-NEXT: v_add3_u32 v12, v12, v24, s4
+; CHECK-NEXT: v_or_b32_e32 v16, 0x400000, v24
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v24, v24
+; CHECK-NEXT: v_cndmask_b32_e32 v12, v12, v16, vcc
+; CHECK-NEXT: v_bfe_u32 v16, v25, 16, 1
+; CHECK-NEXT: v_add3_u32 v16, v16, v25, s4
+; CHECK-NEXT: v_or_b32_e32 v17, 0x400000, v25
+; CHECK-NEXT: v_cmp_u_f32_e32 vcc, v25, v25
+; CHECK-NEXT: v_cndmask_b32_e32 v16, v16, v17, vcc
+; CHECK-NEXT: v_perm_b32 v12, v16, v12, s5
+; CHECK-NEXT: global_store_dwordx4 v[32:33], v[12:15], off offset:48
+; CHECK-NEXT: global_store_dwordx4 v[32:33], v[8:11], off offset:32
+; CHECK-NEXT: global_store_dwordx4 v[32:33], v[0:3], off offset:16
+; CHECK-NEXT: global_store_dwordx4 v[32:33], v[4:7], off
+; CHECK-NEXT: s_waitcnt vmcnt(0)
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+entry:
+ %conv = fptrunc <32 x float> %num to <32 x bfloat>
+ store <32 x bfloat> %conv, ptr addrspace(1) %p, align 8
+ ret void
+}
More information about the llvm-commits
mailing list