[llvm] [AMDGPU][NFC] Add GFX13 run lines to bitcast tests (PR #216980)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Aug 18 03:20:25 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Mariusz Sikora (mariusz-sikora-at-amd)
<details>
<summary>Changes</summary>
---
Patch is 7.59 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/216980.diff
24 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll (+37371)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.128bit.ll (+5218)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.160bit.ll (+634)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.16bit.ll (+420)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.192bit.ll (+1625)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.224bit.ll (+704)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.256bit.ll (+8412)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.288bit.ll (+770)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.320bit.ll (+8459)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.32bit.ll (+2723)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.352bit.ll (+822)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.384bit.ll (+2064)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.448bit.ll (+2201)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.48bit.ll (+280-8)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.576bit.ll (+6762)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.640bit.ll (+7348)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.64bit.ll (+4077)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.704bit.ll (+7929)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.768bit.ll (+8503)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.832bit.ll (+9084)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.896bit.ll (+9658)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.960bit.ll (+10251)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.96bit.ll (+2895)
- (modified) llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.ptr.ll (+13)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll b/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll
index 2b1650e429b86..806c2b918ba7d 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll
@@ -7,6 +7,8 @@
; RUN: llc -mtriple=amdgpu9.00 -verify-machineinstrs=0 < %s | FileCheck -check-prefix=GFX9 %s
; RUN: llc -mtriple=amdgpu11.00 -mattr=+real-true16 -verify-machineinstrs=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-TRUE16 %s
; RUN: llc -mtriple=amdgpu11.00 -mattr=-real-true16 -verify-machineinstrs=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-FAKE16 %s
+; RUN: llc -mtriple=amdgpu13.10 -mattr=+real-true16 -verify-machineinstrs=0 < %s | FileCheck -check-prefixes=GFX13,GFX13-TRUE16 %s
+; RUN: llc -mtriple=amdgpu13.10 -mattr=-real-true16 -verify-machineinstrs=0 < %s | FileCheck -check-prefixes=GFX13,GFX13-FAKE16 %s
define <32 x float> @bitcast_v32i32_to_v32f32(<32 x i32> %a, i32 %b) #0 {
; SI-LABEL: bitcast_v32i32_to_v32f32:
@@ -210,6 +212,46 @@ define <32 x float> @bitcast_v32i32_to_v32f32(<32 x i32> %a, i32 %b) #0 {
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32i32_to_v32f32:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: s_clause 0x1
+; GFX13-NEXT: scratch_load_b32 v32, off, s32 offset:4
+; GFX13-NEXT: scratch_load_b32 v31, off, s32
+; GFX13-NEXT: s_mov_b32 s0, exec_lo
+; GFX13-NEXT: s_wait_loadcnt 0x1
+; GFX13-NEXT: v_cmpx_ne_u32_e32 0, v32
+; GFX13-NEXT: s_xor_b32 s0, exec_lo, s0
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_not1_saveexec_b32 s0, s0
+; GFX13-NEXT: s_cbranch_execz .LBB0_2
+; GFX13-NEXT: ; %bb.1: ; %cmp.true
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: v_dual_add_nc_u32 v31, 3, v31 :: v_dual_add_nc_u32 v30, 3, v30
+; GFX13-NEXT: v_dual_add_nc_u32 v29, 3, v29 :: v_dual_add_nc_u32 v28, 3, v28
+; GFX13-NEXT: v_dual_add_nc_u32 v27, 3, v27 :: v_dual_add_nc_u32 v26, 3, v26
+; GFX13-NEXT: v_dual_add_nc_u32 v25, 3, v25 :: v_dual_add_nc_u32 v24, 3, v24
+; GFX13-NEXT: v_dual_add_nc_u32 v23, 3, v23 :: v_dual_add_nc_u32 v22, 3, v22
+; GFX13-NEXT: v_dual_add_nc_u32 v21, 3, v21 :: v_dual_add_nc_u32 v20, 3, v20
+; GFX13-NEXT: v_dual_add_nc_u32 v19, 3, v19 :: v_dual_add_nc_u32 v18, 3, v18
+; GFX13-NEXT: v_dual_add_nc_u32 v17, 3, v17 :: v_dual_add_nc_u32 v16, 3, v16
+; GFX13-NEXT: v_dual_add_nc_u32 v15, 3, v15 :: v_dual_add_nc_u32 v14, 3, v14
+; GFX13-NEXT: v_dual_add_nc_u32 v13, 3, v13 :: v_dual_add_nc_u32 v12, 3, v12
+; GFX13-NEXT: v_dual_add_nc_u32 v11, 3, v11 :: v_dual_add_nc_u32 v10, 3, v10
+; GFX13-NEXT: v_dual_add_nc_u32 v9, 3, v9 :: v_dual_add_nc_u32 v8, 3, v8
+; GFX13-NEXT: v_dual_add_nc_u32 v7, 3, v7 :: v_dual_add_nc_u32 v6, 3, v6
+; GFX13-NEXT: v_dual_add_nc_u32 v5, 3, v5 :: v_dual_add_nc_u32 v4, 3, v4
+; GFX13-NEXT: v_dual_add_nc_u32 v3, 3, v3 :: v_dual_add_nc_u32 v2, 3, v2
+; GFX13-NEXT: v_dual_add_nc_u32 v1, 3, v1 :: v_dual_add_nc_u32 v0, 3, v0
+; GFX13-NEXT: .LBB0_2: ; %end
+; GFX13-NEXT: s_or_b32 exec_lo, exec_lo, s0
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: s_set_pc_i64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
@@ -615,6 +657,92 @@ define inreg <32 x float> @bitcast_v32i32_to_v32f32_scalar(<32 x i32> inreg %a,
; GFX11-NEXT: v_dual_mov_b32 v28, s7 :: v_dual_mov_b32 v29, s6
; GFX11-NEXT: v_dual_mov_b32 v30, s5 :: v_dual_mov_b32 v31, s4
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32i32_to_v32f32_scalar:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: v_readfirstlane_b32 s42, v14
+; GFX13-NEXT: v_readfirstlane_b32 s4, v13
+; GFX13-NEXT: v_readfirstlane_b32 s5, v12
+; GFX13-NEXT: v_readfirstlane_b32 s6, v11
+; GFX13-NEXT: v_readfirstlane_b32 s7, v10
+; GFX13-NEXT: v_readfirstlane_b32 s8, v9
+; GFX13-NEXT: v_readfirstlane_b32 s9, v8
+; GFX13-NEXT: v_readfirstlane_b32 s10, v7
+; GFX13-NEXT: v_readfirstlane_b32 s11, v6
+; GFX13-NEXT: v_readfirstlane_b32 s12, v5
+; GFX13-NEXT: v_readfirstlane_b32 s13, v4
+; GFX13-NEXT: v_readfirstlane_b32 s14, v3
+; GFX13-NEXT: v_readfirstlane_b32 s15, v2
+; GFX13-NEXT: v_readfirstlane_b32 s40, v1
+; GFX13-NEXT: v_readfirstlane_b32 s41, v0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 0
+; GFX13-NEXT: s_mov_b32 s42, 0
+; GFX13-NEXT: s_cbranch_scc1 .LBB1_2
+; GFX13-NEXT: ; %bb.1:
+; GFX13-NEXT: s_mov_b32 s42, -1
+; GFX13-NEXT: .LBB1_2: ; %Flow
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_b32 s42, s42, exec_lo
+; GFX13-NEXT: s_cselect_b32 s42, 1, 0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 1
+; GFX13-NEXT: s_cbranch_scc1 .LBB1_4
+; GFX13-NEXT: ; %bb.3: ; %cmp.true
+; GFX13-NEXT: s_add_co_i32 s4, s4, 3
+; GFX13-NEXT: s_add_co_i32 s5, s5, 3
+; GFX13-NEXT: s_add_co_i32 s6, s6, 3
+; GFX13-NEXT: s_add_co_i32 s7, s7, 3
+; GFX13-NEXT: s_add_co_i32 s8, s8, 3
+; GFX13-NEXT: s_add_co_i32 s9, s9, 3
+; GFX13-NEXT: s_add_co_i32 s10, s10, 3
+; GFX13-NEXT: s_add_co_i32 s11, s11, 3
+; GFX13-NEXT: s_add_co_i32 s12, s12, 3
+; GFX13-NEXT: s_add_co_i32 s13, s13, 3
+; GFX13-NEXT: s_add_co_i32 s14, s14, 3
+; GFX13-NEXT: s_add_co_i32 s15, s15, 3
+; GFX13-NEXT: s_add_co_i32 s40, s40, 3
+; GFX13-NEXT: s_add_co_i32 s41, s41, 3
+; GFX13-NEXT: s_add_co_i32 s29, s29, 3
+; GFX13-NEXT: s_add_co_i32 s28, s28, 3
+; GFX13-NEXT: s_add_co_i32 s27, s27, 3
+; GFX13-NEXT: s_add_co_i32 s26, s26, 3
+; GFX13-NEXT: s_add_co_i32 s25, s25, 3
+; GFX13-NEXT: s_add_co_i32 s24, s24, 3
+; GFX13-NEXT: s_add_co_i32 s23, s23, 3
+; GFX13-NEXT: s_add_co_i32 s22, s22, 3
+; GFX13-NEXT: s_add_co_i32 s21, s21, 3
+; GFX13-NEXT: s_add_co_i32 s20, s20, 3
+; GFX13-NEXT: s_add_co_i32 s19, s19, 3
+; GFX13-NEXT: s_add_co_i32 s18, s18, 3
+; GFX13-NEXT: s_add_co_i32 s17, s17, 3
+; GFX13-NEXT: s_add_co_i32 s16, s16, 3
+; GFX13-NEXT: s_add_co_i32 s3, s3, 3
+; GFX13-NEXT: s_add_co_i32 s2, s2, 3
+; GFX13-NEXT: s_add_co_i32 s1, s1, 3
+; GFX13-NEXT: s_add_co_i32 s0, s0, 3
+; GFX13-NEXT: .LBB1_4: ; %end
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX13-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX13-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX13-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
+; GFX13-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
+; GFX13-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
+; GFX13-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
+; GFX13-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
+; GFX13-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
+; GFX13-NEXT: v_dual_mov_b32 v16, s28 :: v_dual_mov_b32 v17, s29
+; GFX13-NEXT: v_dual_mov_b32 v18, s41 :: v_dual_mov_b32 v19, s40
+; GFX13-NEXT: v_dual_mov_b32 v20, s15 :: v_dual_mov_b32 v21, s14
+; GFX13-NEXT: v_dual_mov_b32 v22, s13 :: v_dual_mov_b32 v23, s12
+; GFX13-NEXT: v_dual_mov_b32 v24, s11 :: v_dual_mov_b32 v25, s10
+; GFX13-NEXT: v_dual_mov_b32 v26, s9 :: v_dual_mov_b32 v27, s8
+; GFX13-NEXT: v_dual_mov_b32 v28, s7 :: v_dual_mov_b32 v29, s6
+; GFX13-NEXT: v_dual_mov_b32 v30, s5 :: v_dual_mov_b32 v31, s4
+; GFX13-NEXT: s_set_pc_i64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
@@ -818,6 +946,46 @@ define <32 x i32> @bitcast_v32f32_to_v32i32(<32 x float> %a, i32 %b) #0 {
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32f32_to_v32i32:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: s_clause 0x1
+; GFX13-NEXT: scratch_load_b32 v32, off, s32 offset:4
+; GFX13-NEXT: scratch_load_b32 v31, off, s32
+; GFX13-NEXT: s_mov_b32 s0, exec_lo
+; GFX13-NEXT: s_wait_loadcnt 0x1
+; GFX13-NEXT: v_cmpx_ne_u32_e32 0, v32
+; GFX13-NEXT: s_xor_b32 s0, exec_lo, s0
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_not1_saveexec_b32 s0, s0
+; GFX13-NEXT: s_cbranch_execz .LBB2_2
+; GFX13-NEXT: ; %bb.1: ; %cmp.true
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: v_dual_add_f32 v31, 1.0, v31 :: v_dual_add_f32 v30, 1.0, v30
+; GFX13-NEXT: v_dual_add_f32 v29, 1.0, v29 :: v_dual_add_f32 v28, 1.0, v28
+; GFX13-NEXT: v_dual_add_f32 v27, 1.0, v27 :: v_dual_add_f32 v26, 1.0, v26
+; GFX13-NEXT: v_dual_add_f32 v25, 1.0, v25 :: v_dual_add_f32 v24, 1.0, v24
+; GFX13-NEXT: v_dual_add_f32 v23, 1.0, v23 :: v_dual_add_f32 v22, 1.0, v22
+; GFX13-NEXT: v_dual_add_f32 v21, 1.0, v21 :: v_dual_add_f32 v20, 1.0, v20
+; GFX13-NEXT: v_dual_add_f32 v19, 1.0, v19 :: v_dual_add_f32 v18, 1.0, v18
+; GFX13-NEXT: v_dual_add_f32 v17, 1.0, v17 :: v_dual_add_f32 v16, 1.0, v16
+; GFX13-NEXT: v_dual_add_f32 v15, 1.0, v15 :: v_dual_add_f32 v14, 1.0, v14
+; GFX13-NEXT: v_dual_add_f32 v13, 1.0, v13 :: v_dual_add_f32 v12, 1.0, v12
+; GFX13-NEXT: v_dual_add_f32 v11, 1.0, v11 :: v_dual_add_f32 v10, 1.0, v10
+; GFX13-NEXT: v_dual_add_f32 v9, 1.0, v9 :: v_dual_add_f32 v8, 1.0, v8
+; GFX13-NEXT: v_dual_add_f32 v7, 1.0, v7 :: v_dual_add_f32 v6, 1.0, v6
+; GFX13-NEXT: v_dual_add_f32 v5, 1.0, v5 :: v_dual_add_f32 v4, 1.0, v4
+; GFX13-NEXT: v_dual_add_f32 v3, 1.0, v3 :: v_dual_add_f32 v2, 1.0, v2
+; GFX13-NEXT: v_dual_add_f32 v1, 1.0, v1 :: v_dual_add_f32 v0, 1.0, v0
+; GFX13-NEXT: .LBB2_2: ; %end
+; GFX13-NEXT: s_or_b32 exec_lo, exec_lo, s0
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: s_set_pc_i64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
@@ -1447,6 +1615,92 @@ define inreg <32 x i32> @bitcast_v32f32_to_v32i32_scalar(<32 x float> inreg %a,
; GFX11-NEXT: s_mov_b32 exec_lo, s0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32f32_to_v32i32_scalar:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: v_readfirstlane_b32 s42, v14
+; GFX13-NEXT: v_readfirstlane_b32 s4, v13
+; GFX13-NEXT: v_readfirstlane_b32 s5, v12
+; GFX13-NEXT: v_readfirstlane_b32 s6, v11
+; GFX13-NEXT: v_readfirstlane_b32 s7, v10
+; GFX13-NEXT: v_readfirstlane_b32 s8, v9
+; GFX13-NEXT: v_readfirstlane_b32 s9, v8
+; GFX13-NEXT: v_readfirstlane_b32 s10, v7
+; GFX13-NEXT: v_readfirstlane_b32 s11, v6
+; GFX13-NEXT: v_readfirstlane_b32 s12, v5
+; GFX13-NEXT: v_readfirstlane_b32 s13, v4
+; GFX13-NEXT: v_readfirstlane_b32 s14, v3
+; GFX13-NEXT: v_readfirstlane_b32 s15, v2
+; GFX13-NEXT: v_readfirstlane_b32 s40, v1
+; GFX13-NEXT: v_readfirstlane_b32 s41, v0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 0
+; GFX13-NEXT: s_mov_b32 s42, 0
+; GFX13-NEXT: s_cbranch_scc1 .LBB3_2
+; GFX13-NEXT: ; %bb.1:
+; GFX13-NEXT: s_mov_b32 s42, -1
+; GFX13-NEXT: .LBB3_2: ; %Flow
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_b32 s42, s42, exec_lo
+; GFX13-NEXT: s_cselect_b32 s42, 1, 0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 1
+; GFX13-NEXT: s_cbranch_scc1 .LBB3_4
+; GFX13-NEXT: ; %bb.3: ; %cmp.true
+; GFX13-NEXT: s_add_f32 s4, s4, 1.0
+; GFX13-NEXT: s_add_f32 s5, s5, 1.0
+; GFX13-NEXT: s_add_f32 s6, s6, 1.0
+; GFX13-NEXT: s_add_f32 s7, s7, 1.0
+; GFX13-NEXT: s_add_f32 s8, s8, 1.0
+; GFX13-NEXT: s_add_f32 s9, s9, 1.0
+; GFX13-NEXT: s_add_f32 s10, s10, 1.0
+; GFX13-NEXT: s_add_f32 s11, s11, 1.0
+; GFX13-NEXT: s_add_f32 s12, s12, 1.0
+; GFX13-NEXT: s_add_f32 s13, s13, 1.0
+; GFX13-NEXT: s_add_f32 s14, s14, 1.0
+; GFX13-NEXT: s_add_f32 s15, s15, 1.0
+; GFX13-NEXT: s_add_f32 s40, s40, 1.0
+; GFX13-NEXT: s_add_f32 s41, s41, 1.0
+; GFX13-NEXT: s_add_f32 s29, s29, 1.0
+; GFX13-NEXT: s_add_f32 s28, s28, 1.0
+; GFX13-NEXT: s_add_f32 s27, s27, 1.0
+; GFX13-NEXT: s_add_f32 s26, s26, 1.0
+; GFX13-NEXT: s_add_f32 s25, s25, 1.0
+; GFX13-NEXT: s_add_f32 s24, s24, 1.0
+; GFX13-NEXT: s_add_f32 s23, s23, 1.0
+; GFX13-NEXT: s_add_f32 s22, s22, 1.0
+; GFX13-NEXT: s_add_f32 s21, s21, 1.0
+; GFX13-NEXT: s_add_f32 s20, s20, 1.0
+; GFX13-NEXT: s_add_f32 s19, s19, 1.0
+; GFX13-NEXT: s_add_f32 s18, s18, 1.0
+; GFX13-NEXT: s_add_f32 s17, s17, 1.0
+; GFX13-NEXT: s_add_f32 s16, s16, 1.0
+; GFX13-NEXT: s_add_f32 s3, s3, 1.0
+; GFX13-NEXT: s_add_f32 s2, s2, 1.0
+; GFX13-NEXT: s_add_f32 s1, s1, 1.0
+; GFX13-NEXT: s_add_f32 s0, s0, 1.0
+; GFX13-NEXT: .LBB3_4: ; %end
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX13-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX13-NEXT: v_dual_mov_b32 v2, s2 :: v_dual_mov_b32 v3, s3
+; GFX13-NEXT: v_dual_mov_b32 v4, s16 :: v_dual_mov_b32 v5, s17
+; GFX13-NEXT: v_dual_mov_b32 v6, s18 :: v_dual_mov_b32 v7, s19
+; GFX13-NEXT: v_dual_mov_b32 v8, s20 :: v_dual_mov_b32 v9, s21
+; GFX13-NEXT: v_dual_mov_b32 v10, s22 :: v_dual_mov_b32 v11, s23
+; GFX13-NEXT: v_dual_mov_b32 v12, s24 :: v_dual_mov_b32 v13, s25
+; GFX13-NEXT: v_dual_mov_b32 v14, s26 :: v_dual_mov_b32 v15, s27
+; GFX13-NEXT: v_dual_mov_b32 v16, s28 :: v_dual_mov_b32 v17, s29
+; GFX13-NEXT: v_dual_mov_b32 v18, s41 :: v_dual_mov_b32 v19, s40
+; GFX13-NEXT: v_dual_mov_b32 v20, s15 :: v_dual_mov_b32 v21, s14
+; GFX13-NEXT: v_dual_mov_b32 v22, s13 :: v_dual_mov_b32 v23, s12
+; GFX13-NEXT: v_dual_mov_b32 v24, s11 :: v_dual_mov_b32 v25, s10
+; GFX13-NEXT: v_dual_mov_b32 v26, s9 :: v_dual_mov_b32 v27, s8
+; GFX13-NEXT: v_dual_mov_b32 v28, s7 :: v_dual_mov_b32 v29, s6
+; GFX13-NEXT: v_dual_mov_b32 v30, s5 :: v_dual_mov_b32 v31, s4
+; GFX13-NEXT: s_set_pc_i64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
@@ -1666,6 +1920,46 @@ define <16 x i64> @bitcast_v32i32_to_v16i64(<32 x i32> %a, i32 %b) #0 {
; GFX11-NEXT: s_or_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32i32_to_v16i64:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: s_clause 0x1
+; GFX13-NEXT: scratch_load_b32 v32, off, s32 offset:4
+; GFX13-NEXT: scratch_load_b32 v31, off, s32
+; GFX13-NEXT: s_mov_b32 s0, exec_lo
+; GFX13-NEXT: s_wait_loadcnt 0x1
+; GFX13-NEXT: v_cmpx_ne_u32_e32 0, v32
+; GFX13-NEXT: s_xor_b32 s0, exec_lo, s0
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_not1_saveexec_b32 s0, s0
+; GFX13-NEXT: s_cbranch_execz .LBB4_2
+; GFX13-NEXT: ; %bb.1: ; %cmp.true
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: v_dual_add_nc_u32 v31, 3, v31 :: v_dual_add_nc_u32 v30, 3, v30
+; GFX13-NEXT: v_dual_add_nc_u32 v29, 3, v29 :: v_dual_add_nc_u32 v28, 3, v28
+; GFX13-NEXT: v_dual_add_nc_u32 v27, 3, v27 :: v_dual_add_nc_u32 v26, 3, v26
+; GFX13-NEXT: v_dual_add_nc_u32 v25, 3, v25 :: v_dual_add_nc_u32 v24, 3, v24
+; GFX13-NEXT: v_dual_add_nc_u32 v23, 3, v23 :: v_dual_add_nc_u32 v22, 3, v22
+; GFX13-NEXT: v_dual_add_nc_u32 v21, 3, v21 :: v_dual_add_nc_u32 v20, 3, v20
+; GFX13-NEXT: v_dual_add_nc_u32 v19, 3, v19 :: v_dual_add_nc_u32 v18, 3, v18
+; GFX13-NEXT: v_dual_add_nc_u32 v17, 3, v17 :: v_dual_add_nc_u32 v16, 3, v16
+; GFX13-NEXT: v_dual_add_nc_u32 v15, 3, v15 :: v_dual_add_nc_u32 v14, 3, v14
+; GFX13-NEXT: v_dual_add_nc_u32 v13, 3, v13 :: v_dual_add_nc_u32 v12, 3, v12
+; GFX13-NEXT: v_dual_add_nc_u32 v11, 3, v11 :: v_dual_add_nc_u32 v10, 3, v10
+; GFX13-NEXT: v_dual_add_nc_u32 v9, 3, v9 :: v_dual_add_nc_u32 v8, 3, v8
+; GFX13-NEXT: v_dual_add_nc_u32 v7, 3, v7 :: v_dual_add_nc_u32 v6, 3, v6
+; GFX13-NEXT: v_dual_add_nc_u32 v5, 3, v5 :: v_dual_add_nc_u32 v4, 3, v4
+; GFX13-NEXT: v_dual_add_nc_u32 v3, 3, v3 :: v_dual_add_nc_u32 v2, 3, v2
+; GFX13-NEXT: v_dual_add_nc_u32 v1, 3, v1 :: v_dual_add_nc_u32 v0, 3, v0
+; GFX13-NEXT: .LBB4_2: ; %end
+; GFX13-NEXT: s_or_b32 exec_lo, exec_lo, s0
+; GFX13-NEXT: s_wait_loadcnt 0x0
+; GFX13-NEXT: s_set_pc_i64 s[30:31]
%cmp = icmp eq i32 %b, 0
br i1 %cmp, label %cmp.true, label %cmp.false
@@ -2071,6 +2365,92 @@ define inreg <16 x i64> @bitcast_v32i32_to_v16i64_scalar(<32 x i32> inreg %a, i3
; GFX11-NEXT: v_dual_mov_b32 v28, s7 :: v_dual_mov_b32 v29, s6
; GFX11-NEXT: v_dual_mov_b32 v30, s5 :: v_dual_mov_b32 v31, s4
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX13-LABEL: bitcast_v32i32_to_v16i64_scalar:
+; GFX13: ; %bb.0:
+; GFX13-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX13-NEXT: s_wait_expcnt 0x0
+; GFX13-NEXT: s_wait_samplecnt 0x0
+; GFX13-NEXT: s_wait_bvhcnt 0x0
+; GFX13-NEXT: s_wait_kmcnt 0x0
+; GFX13-NEXT: v_readfirstlane_b32 s42, v14
+; GFX13-NEXT: v_readfirstlane_b32 s4, v13
+; GFX13-NEXT: v_readfirstlane_b32 s5, v12
+; GFX13-NEXT: v_readfirstlane_b32 s6, v11
+; GFX13-NEXT: v_readfirstlane_b32 s7, v10
+; GFX13-NEXT: v_readfirstlane_b32 s8, v9
+; GFX13-NEXT: v_readfirstlane_b32 s9, v8
+; GFX13-NEXT: v_readfirstlane_b32 s10, v7
+; GFX13-NEXT: v_readfirstlane_b32 s11, v6
+; GFX13-NEXT: v_readfirstlane_b32 s12, v5
+; GFX13-NEXT: v_readfirstlane_b32 s13, v4
+; GFX13-NEXT: v_readfirstlane_b32 s14, v3
+; GFX13-NEXT: v_readfirstlane_b32 s15, v2
+; GFX13-NEXT: v_readfirstlane_b32 s40, v1
+; GFX13-NEXT: v_readfirstlane_b32 s41, v0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 0
+; GFX13-NEXT: s_mov_b32 s42, 0
+; GFX13-NEXT: s_cbranch_scc1 .LBB5_2
+; GFX13-NEXT: ; %bb.1:
+; GFX13-NEXT: s_mov_b32 s42, -1
+; GFX13-NEXT: .LBB5_2: ; %Flow
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GFX13-NEXT: s_and_b32 s42, s42, exec_lo
+; GFX13-NEXT: s_cselect_b32 s42, 1, 0
+; GFX13-NEXT: s_cmp_lg_u32 s42, 1
+; GFX13-NEXT: s_cbranch_scc1 .LBB5_4
+; GFX13-NEXT: ; %bb.3: ; %cmp.true
+; GFX13-NEXT: s_add_co_i32 s4, s4, 3
+; GFX13-NEXT: s_add_co_i32 s5, s5, 3
+; GFX13-NEXT: s_add_co_i32 s6, s6, 3
+; GFX13-NEXT: s_add_co_i32 s7, s7, 3
+; GFX13-NEXT: s_add_co_i32 s8, s8, 3
+; GFX13-NEXT: s_add_co_i32 s9, s9, 3
+; GFX13-NEXT: s_add_co_i32 s10, s10, 3
+; GFX13-NEXT: s_add_co_i32 s11, s11, 3
+; GFX13-NEXT: s_add_co_i32 s12, s12, 3
+; GFX13-NEXT: s_add_co_i32 s13, s13, 3
+; GFX13-NEXT: s_add_co_i32 s14, s14, 3
+; GFX13-NEXT: s_add_co_i32 s15, s15, 3
+; GFX13-NEXT: s_add_co_i32 s40, s40, 3
+; GFX13-NEXT: s_add_co_i32 s41, s41, 3
+; GFX13-NEXT: s_add_co_i32 s29, s29, 3
+; GFX13-NEXT: s_add_co_i32 s28, s28, 3
+; GFX13-NEXT: s_add_co_i32 s27, s27, 3
+; GFX13-NEXT: s_add_co_i32 s26, s26, 3
+; GFX13-NEXT: s_add_co_i32 s25, s25, 3
+; GFX13-NEXT: s_add_co_i32 s24, s24, 3
+; GFX13-NEXT: s_add_co_i32 s23, s23, 3
+; GFX13-NEXT: s_add_co_i32 s22, s22, 3
+; GFX13-NEXT: s_add_co_i32 s21, s21, 3
+; GFX13-NEXT: s_add_co_i32 s20, s20, 3
+; GFX13-NEXT: s_add_co_i32 s19, s19, 3
+; GFX13-NEXT: s_add_co_i32 s18, s18, 3
+; GFX13-NEXT: s_add_co_i32 s17, s17, 3
+; GFX13-NEXT: s_add_co_i32 s16, s16, 3
+; GFX13-NEXT: s_add_co_i32 s3, s3, 3
+; GFX13-NEXT: s_add_co_i32 s2, s2, 3
+; GFX13-NEXT: s_add_co_i32 s1, s1, 3
+; GFX13-NEXT: s_add_co_i32 s0, s0, 3
+; GFX13-NEXT: .LBB5_4: ; %end
+; GFX13-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX13-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
+; GFX13-NEXT: v_dual_mov_b32 v2, s2 ::...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/216980
More information about the llvm-commits
mailing list