[AMDGPU][DAG] Remove AssertZext before some intrinsics #146052

Pierre-vh · 2025-06-27T10:45:52Z

No description provided.

Pierre-vh · 2025-06-27T10:46:08Z

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-06-27T10:48:01Z

@llvm/pr-subscribers-backend-amdgpu

Author: Pierre van Houtryve (Pierre-vh)

Changes

Patch is 30.11 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/146052.diff

11 Files Affected:

(modified) llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp (+18)
(modified) llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll (+2-2)
(modified) llvm/test/CodeGen/AMDGPU/flat-scratch.ll (+27-27)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll (+3-3)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx11.ll (+8-8)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx12.ll (+4-4)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll (+10-10)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.ll (+8-8)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll (+2-2)
(modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.workitem.id-unsupported-calling-convention.ll (+5-11)
(modified) llvm/test/CodeGen/AMDGPU/memory_clause.ll (+4-4)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 3db2b3bff2d36..7a5619fb1fda1 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -3985,6 +3985,24 @@ SDValue AMDGPUTargetLowering::performAssertSZExtCombine(SDNode *N,
     }
   }
 
+  // AssertZext in front of these intrinsics is not necessary, the lowering of
+  // the intrinsics into a register read will insert one if it is needed.
+  if (N->getOpcode() == ISD::AssertZext &&
+      N0.getOpcode() == ISD::INTRINSIC_WO_CHAIN) {
+    unsigned IID = N0.getConstantOperandVal(0);
+    switch (IID) {
+    case Intrinsic::amdgcn_workitem_id_x:
+    case Intrinsic::amdgcn_workitem_id_y:
+    case Intrinsic::amdgcn_workitem_id_z:
+    case Intrinsic::amdgcn_workgroup_id_x:
+    case Intrinsic::amdgcn_workgroup_id_y:
+    case Intrinsic::amdgcn_workgroup_id_z:
+      return N0;
+    default:
+      break;
+    }
+  }
+
   return SDValue();
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll b/llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll
index 9cf9d81773037..7a542d2e9c514 100644
--- a/llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll
+++ b/llvm/test/CodeGen/AMDGPU/ds-sub-offset.ll
@@ -442,9 +442,9 @@ define amdgpu_kernel void @add_x_shl_neg_to_sub_multi_use() #1 {
 ;
 ; GFX11-LABEL: add_x_shl_neg_to_sub_multi_use:
 ; GFX11:       ; %bb.0:
-; GFX11-NEXT:    v_dual_mov_b32 v1, 13 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX11-NEXT:    v_dual_mov_b32 v1, 13 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-NEXT:    v_sub_nc_u32_e32 v0, 0, v0
 ; GFX11-NEXT:    ds_store_b32 v0, v1 offset:123
 ; GFX11-NEXT:    ds_store_b32 v0, v1 offset:456
diff --git a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
index b5e579b78a59c..b25d9b245f5f6 100644
--- a/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
+++ b/llvm/test/CodeGen/AMDGPU/flat-scratch.ll
@@ -714,10 +714,10 @@ define amdgpu_kernel void @store_load_vindex_kernel(i32 %n) {
 ; GFX11-LABEL: store_load_vindex_kernel:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX11-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
@@ -732,9 +732,9 @@ define amdgpu_kernel void @store_load_vindex_kernel(i32 %n) {
 ; GFX12-LABEL: store_load_vindex_kernel:
 ; GFX12:       ; %bb.0: ; %bb
 ; GFX12-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
@@ -769,8 +769,8 @@ define amdgpu_kernel void @store_load_vindex_kernel(i32 %n) {
 ; GFX942-LABEL: store_load_vindex_kernel:
 ; GFX942:       ; %bb.0: ; %bb
 ; GFX942-NEXT:    s_load_dword s0, s[4:5], 0x24
-; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX942-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX942-NEXT:    v_mov_b32_e32 v1, 15
 ; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX942-NEXT:    s_lshl_b32 s0, s0, 7
@@ -809,10 +809,10 @@ define amdgpu_kernel void @store_load_vindex_kernel(i32 %n) {
 ; GFX11-PAL-LABEL: store_load_vindex_kernel:
 ; GFX11-PAL:       ; %bb.0: ; %bb
 ; GFX11-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX11-PAL-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX11-PAL-NEXT:    v_mov_b32_e32 v2, 15
 ; GFX11-PAL-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-PAL-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX11-PAL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
@@ -827,9 +827,9 @@ define amdgpu_kernel void @store_load_vindex_kernel(i32 %n) {
 ; GFX12-PAL-LABEL: store_load_vindex_kernel:
 ; GFX12-PAL:       ; %bb.0: ; %bb
 ; GFX12-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-PAL-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-PAL-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-PAL-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX12-PAL-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
@@ -1958,10 +1958,10 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel(i32 %n) {
 ; GFX11-LABEL: store_load_vindex_small_offset_kernel:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_b32 v3, off, off glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-NEXT:    scratch_store_b32 v0, v1, off offset:384 dlc
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
@@ -1976,10 +1976,10 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel(i32 %n) {
 ; GFX12-LABEL: store_load_vindex_small_offset_kernel:
 ; GFX12:       ; %bb.0: ; %bb
 ; GFX12-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-NEXT:    scratch_load_b32 v3, off, off scope:SCOPE_SYS
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:384 scope:SCOPE_SYS
 ; GFX12-NEXT:    s_wait_storecnt 0x0
@@ -2021,8 +2021,8 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel(i32 %n) {
 ; GFX942-NEXT:    s_load_dword s0, s[4:5], 0x24
 ; GFX942-NEXT:    scratch_load_dword v1, off, off sc0 sc1
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
-; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX942-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX942-NEXT:    v_mov_b32_e32 v1, 15
 ; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX942-NEXT:    s_lshl_b32 s0, s0, 7
@@ -2092,10 +2092,10 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel(i32 %n) {
 ; GFX11-PAL-LABEL: store_load_vindex_small_offset_kernel:
 ; GFX11-PAL:       ; %bb.0: ; %bb
 ; GFX11-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX11-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX11-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-PAL-NEXT:    scratch_load_b32 v3, off, off glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-PAL-NEXT:    scratch_store_b32 v0, v1, off offset:384 dlc
 ; GFX11-PAL-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-PAL-NEXT:    s_waitcnt lgkmcnt(0)
@@ -2110,10 +2110,10 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel(i32 %n) {
 ; GFX12-PAL-LABEL: store_load_vindex_small_offset_kernel:
 ; GFX12-PAL:       ; %bb.0: ; %bb
 ; GFX12-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-PAL-NEXT:    scratch_load_b32 v3, off, off scope:SCOPE_SYS
 ; GFX12-PAL-NEXT:    s_wait_loadcnt 0x0
-; GFX12-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-PAL-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-PAL-NEXT:    scratch_store_b32 v0, v1, off offset:384 scope:SCOPE_SYS
 ; GFX12-PAL-NEXT:    s_wait_storecnt 0x0
@@ -3254,10 +3254,10 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel(i32 %n) {
 ; GFX11-LABEL: store_load_vindex_large_offset_kernel:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX11-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-NEXT:    scratch_load_b32 v3, off, off offset:4 glc dlc
 ; GFX11-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
@@ -3274,10 +3274,10 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel(i32 %n) {
 ; GFX12-LABEL: store_load_vindex_large_offset_kernel:
 ; GFX12:       ; %bb.0: ; %bb
 ; GFX12-NEXT:    s_load_b32 s0, s[4:5], 0x24
-; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-NEXT:    scratch_load_b32 v3, off, off scope:SCOPE_SYS
 ; GFX12-NEXT:    s_wait_loadcnt 0x0
-; GFX12-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-NEXT:    scratch_store_b32 v0, v1, off offset:16512 scope:SCOPE_SYS
 ; GFX12-NEXT:    s_wait_storecnt 0x0
@@ -3319,8 +3319,8 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel(i32 %n) {
 ; GFX942-NEXT:    s_load_dword s0, s[4:5], 0x24
 ; GFX942-NEXT:    scratch_load_dword v1, off, off offset:4 sc0 sc1
 ; GFX942-NEXT:    s_waitcnt vmcnt(0)
-; GFX942-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GFX942-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX942-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX942-NEXT:    v_mov_b32_e32 v1, 15
 ; GFX942-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX942-NEXT:    s_lshl_b32 s0, s0, 7
@@ -3391,10 +3391,10 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel(i32 %n) {
 ; GFX11-PAL-LABEL: store_load_vindex_large_offset_kernel:
 ; GFX11-PAL:       ; %bb.0: ; %bb
 ; GFX11-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX11-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX11-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX11-PAL-NEXT:    scratch_load_b32 v3, off, off offset:4 glc dlc
 ; GFX11-PAL-NEXT:    s_waitcnt vmcnt(0)
-; GFX11-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX11-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX11-PAL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-PAL-NEXT:    s_lshl_b32 s0, s0, 7
 ; GFX11-PAL-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
@@ -3411,10 +3411,10 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel(i32 %n) {
 ; GFX12-PAL-LABEL: store_load_vindex_large_offset_kernel:
 ; GFX12-PAL:       ; %bb.0: ; %bb
 ; GFX12-PAL-NEXT:    s_load_b32 s0, s[4:5], 0x0
-; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_and_b32 v0, 0x3ff, v0
+; GFX12-PAL-NEXT:    v_dual_mov_b32 v1, 15 :: v_dual_lshlrev_b32 v0, 2, v0
 ; GFX12-PAL-NEXT:    scratch_load_b32 v3, off, off scope:SCOPE_SYS
 ; GFX12-PAL-NEXT:    s_wait_loadcnt 0x0
-; GFX12-PAL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GFX12-PAL-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GFX12-PAL-NEXT:    s_wait_kmcnt 0x0
 ; GFX12-PAL-NEXT:    scratch_store_b32 v0, v1, off offset:16512 scope:SCOPE_SYS
 ; GFX12-PAL-NEXT:    s_wait_storecnt 0x0
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll
index 565ad295ebbb3..08c0d15432915 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll
@@ -15,8 +15,8 @@ define amdgpu_kernel void @test_iglp_opt_mfma_gemm(ptr addrspace(3) noalias %in,
 ; GCN-LABEL: test_iglp_opt_mfma_gemm:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 7, v0
+; GCN-NEXT:    v_and_b32_e32 v0, 0x1ff80, v0
 ; GCN-NEXT:    v_mov_b32_e32 v3, 2.0
 ; GCN-NEXT:    ; iglp_opt mask(0x00000000)
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
@@ -153,8 +153,8 @@ define amdgpu_kernel void @test_iglp_opt_rev_mfma_gemm(ptr addrspace(3) noalias
 ; GCN-LABEL: test_iglp_opt_rev_mfma_gemm:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 7, v0
+; GCN-NEXT:    v_and_b32_e32 v0, 0x1ff80, v0
 ; GCN-NEXT:    v_mov_b32_e32 v2, 1.0
 ; GCN-NEXT:    v_mov_b32_e32 v3, 2.0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
@@ -289,8 +289,8 @@ define amdgpu_kernel void @test_iglp_opt_asm_sideeffect(ptr addrspace(3) noalias
 ; GCN-LABEL: test_iglp_opt_asm_sideeffect:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GCN-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; GCN-NEXT:    v_and_b32_e32 v0, 0xffc, v0
 ; GCN-NEXT:    ; iglp_opt mask(0x00000000)
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_add_u32_e32 v1, s0, v0
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx11.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx11.ll
index 6507976872410..46359f7e99059 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx11.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx11.ll
@@ -6,9 +6,9 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_cluster(ptr ad
 ; GCN-LABEL: test_sched_group_barrier_pipeline_WMMA_cluster:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GCN-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
 ; GCN-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GCN-NEXT:    v_lshlrev_b32_e32 v40, 5, v0
+; GCN-NEXT:    v_and_b32_e32 v40, 0x7fe0, v0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_add_nc_u32_e32 v32, s0, v40
 ; GCN-NEXT:    v_dual_mov_b32 v81, s1 :: v_dual_add_nc_u32 v80, s1, v40
@@ -74,9 +74,9 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_cluster(ptr ad
 ; EXACTCUTOFF-LABEL: test_sched_group_barrier_pipeline_WMMA_cluster:
 ; EXACTCUTOFF:       ; %bb.0: ; %entry
 ; EXACTCUTOFF-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; EXACTCUTOFF-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
 ; EXACTCUTOFF-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v40, 5, v0
+; EXACTCUTOFF-NEXT:    v_and_b32_e32 v40, 0x7fe0, v0
 ; EXACTCUTOFF-NEXT:    s_waitcnt lgkmcnt(0)
 ; EXACTCUTOFF-NEXT:    v_add_nc_u32_e32 v32, s0, v40
 ; EXACTCUTOFF-NEXT:    v_dual_mov_b32 v81, s1 :: v_dual_add_nc_u32 v80, s1, v40
@@ -178,9 +178,9 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_interleave(ptr
 ; GCN-LABEL: test_sched_group_barrier_pipeline_WMMA_interleave:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GCN-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
 ; GCN-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GCN-NEXT:    v_lshlrev_b32_e32 v16, 5, v0
+; GCN-NEXT:    v_and_b32_e32 v16, 0x7fe0, v0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_add_nc_u32_e32 v17, s0, v16
 ; GCN-NEXT:    v_add_nc_u32_e32 v16, s1, v16
@@ -260,9 +260,9 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_interleave(ptr
 ; EXACTCUTOFF-LABEL: test_sched_group_barrier_pipeline_WMMA_interleave:
 ; EXACTCUTOFF:       ; %bb.0: ; %entry
 ; EXACTCUTOFF-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; EXACTCUTOFF-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
 ; EXACTCUTOFF-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v16, 5, v0
+; EXACTCUTOFF-NEXT:    v_and_b32_e32 v16, 0x7fe0, v0
 ; EXACTCUTOFF-NEXT:    s_waitcnt lgkmcnt(0)
 ; EXACTCUTOFF-NEXT:    v_add_nc_u32_e32 v17, s0, v16
 ; EXACTCUTOFF-NEXT:    v_add_nc_u32_e32 v16, s1, v16
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx12.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx12.ll
index 02e80b62fed6e..dcc3e0df0c744 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx12.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.gfx12.ll
@@ -8,10 +8,10 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_SWMMAC_cluster(ptr
 ; GCN-LABEL: test_sched_group_barrier_pipeline_SWMMAC_cluster:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; GCN-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; GCN-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; GCN-NEXT:    v_mov_b32_e32 v48, 0
 ; GCN-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; GCN-NEXT:    v_lshlrev_b32_e32 v28, 4, v0
+; GCN-NEXT:    v_and_b32_e32 v28, 0x3ff0, v0
 ; GCN-NEXT:    s_wait_kmcnt 0x0
 ; GCN-NEXT:    v_add_nc_u32_e32 v0, s0, v28
 ; GCN-NEXT:    v_dual_mov_b32 v50, s1 :: v_dual_add_nc_u32 v49, s1, v28
@@ -60,10 +60,10 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_SWMMAC_cluster(ptr
 ; EXACTCUTOFF-LABEL: test_sched_group_barrier_pipeline_SWMMAC_cluster:
 ; EXACTCUTOFF:       ; %bb.0: ; %entry
 ; EXACTCUTOFF-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
-; EXACTCUTOFF-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
+; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; EXACTCUTOFF-NEXT:    v_mov_b32_e32 v48, 0
 ; EXACTCUTOFF-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_1)
-; EXACTCUTOFF-NEXT:    v_lshlrev_b32_e32 v28, 4, v0
+; EXACTCUTOFF-NEXT:    v_and_b32_e32 v28, 0x3ff0, v0
 ; EXACTCUTOFF-NEXT:    s_wait_kmcnt 0x0
 ; EXACTCUTOFF-NEXT:    v_add_nc_u32_e32 v0, s0, v28
 ; EXACTCUTOFF-NEXT:    v_dual_mov_b32 v50, s1 :: v_dual_add_nc_u32 v49, s1, v28
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll
index 371b4f070094d..0764cd5d34d75 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.iterative.ll
@@ -7,8 +7,8 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_interleave(ptr
 ; GCN-MINREG-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
 ; GCN-MINREG:       ; %bb.0: ; %entry
 ; GCN-MINREG-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GCN-MINREG-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
 ; GCN-MINREG-NEXT:    v_lshlrev_b32_e32 v0, 7, v0
+; GCN-MINREG-NEXT:    v_and_b32_e32 v0, 0x1ff80, v0
 ; GCN-MINREG-NEXT:    v_mov_b32_e32 v2, 1.0
 ; GCN-MINREG-NEXT:    v_mov_b32_e32 v1, 2.0
 ; GCN-MINREG-NEXT:    s_waitcnt lgkmcnt(0)
@@ -140,8 +140,8 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_interleave(ptr
 ; GCN-MAXOCC-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
 ; GCN-MAXOCC:       ; %bb.0: ; %entry
 ; GCN-MAXOCC-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GCN-MAXOCC-NEXT:    v_and_b32_e32 v0, 0x3ff, v0
-; GCN-MAXOCC-NEXT:    v_lshlrev_b32_e32 v1, 7, v0
+; GCN-MAXOCC-NEXT:    v_lshlrev_b32_e32 v0, 7, v0
+; GCN-MAXOCC-NEXT:    v_and_b32_e32 v1, 0x1ff80, v0
 ; GCN-MAXOCC-NEXT:    v_mov_b32_e32 v2, 1.0
 ; GCN-MAXOCC-NEXT:    v_mov_b32_e32 v3, 2.0
 ; GCN-MAXOCC-NEXT:    s_waitcnt lgkmcnt(0)
@@ -274,8 +274,8 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_interleave(ptr
 ; GCN-ILP-LABEL: test_sched_group_barrier_pipeline_MFMA_interleave:
 ; GCN-ILP:       ; %bb.0: ; %entry
 ; GCN-ILP-NEXT:    s_...
[truncated]

jayfoad · 2025-06-27T11:09:45Z

llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp

+  // AssertZext in front of these intrinsics is not necessary, the lowering of
+  // the intrinsics into a register read will insert one if it is needed.


I don't understand. AssertZext is just an assertion, it does not generate any code, so removing it is presumably just working around some other part of the backend that doesn't know to look through it?

so removing it is presumably just working around some other part of the backend that doesn't know to look through it?

It's a bit of both really

There's a lot of combines that can't look through it, yes

It's also really not necessary here because these intrinsics are lowered immediately after and that code will already insert an AssertZext if one is required. Without this we get (and (assert_zext v31, i10), 1023) for example.

I'm also not sure AssertZext is strictly an assertion that doesn't generate code. I think some backends use it to guide ISel, and it also has an effect on poison apparently.

It is an assertion that doesn't generate code. If it does that's a bug. We should probably skip adding it from the original callsite if the known range is already implied by the intrinsic declaration

[AMDGPU][DAG] Remove AssertZext before some intrinsics

9ad4aa6

This was referenced Jun 27, 2025

[AMDGPU] Add tests for workgroup/workitem intrinsic optimizations #146053

Open

[DAG] Fold (setcc ((x | x >> c0 | ...) & mask)) sequences #146054

Open

[GISel] Combine compare of bitfield extracts or'd together. #146055

Open

Pierre-vh requested review from arsenm, jayfoad and shiltian June 27, 2025 10:46

Pierre-vh marked this pull request as ready for review June 27, 2025 10:47

llvmbot added the backend:AMDGPU label Jun 27, 2025

jayfoad reviewed Jun 27, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[AMDGPU][DAG] Remove AssertZext before some intrinsics #146052

[AMDGPU][DAG] Remove AssertZext before some intrinsics #146052

Pierre-vh commented Jun 27, 2025

Uh oh!

Pierre-vh commented Jun 27, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jun 27, 2025

Uh oh!

jayfoad Jun 27, 2025

Uh oh!

Pierre-vh Jun 27, 2025

Uh oh!

arsenm Jun 27, 2025

Uh oh!

Uh oh!

		// AssertZext in front of these intrinsics is not necessary, the lowering of
		// the intrinsics into a register read will insert one if it is needed.

[AMDGPU][DAG] Remove AssertZext before some intrinsics #146052

Are you sure you want to change the base?

[AMDGPU][DAG] Remove AssertZext before some intrinsics #146052

Conversation

Pierre-vh commented Jun 27, 2025

Uh oh!

Pierre-vh commented Jun 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jun 27, 2025

Uh oh!

jayfoad Jun 27, 2025

Choose a reason for hiding this comment

Uh oh!

Pierre-vh Jun 27, 2025

Choose a reason for hiding this comment

Uh oh!

arsenm Jun 27, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Pierre-vh commented Jun 27, 2025 •

edited

Loading