[AMDGPU] Do not generate s_set_inst_prefetch_distance for GFX12

[jayfoad]

GFX12 can still encode the s_set_inst_prefetch_distance instruction but
it has no effect.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Jay Foad (jayfoad)

Changes

GFX12 can still encode the s_set_inst_prefetch_distance instruction but
it has no effect.

---

Full diff: https://github.com/llvm/llvm-project/pull/78190.diff

3 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/GCNSubtarget.h (+5-1)
• (modified) llvm/test/CodeGen/AMDGPU/loop-prefetch-data.ll (-1)
• (modified) llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll (+50)

diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index 070d165cdaadb8f..c825ea96b0fe436 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -846,7 +846,11 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
     return getGeneration() < SEA_ISLANDS;
   }
 
-  bool hasInstPrefetch() const { return getGeneration() >= GFX10; }
+  bool hasInstPrefetch() const {
+    // GFX12 can still encode the s_set_inst_prefetch_distance instruction but
+    // it has no effect.
+    return getGeneration() == GFX10 || getGeneration() == GFX11;
+  }
 
   bool hasPrefetch() const { return GFX12Insts; }
 
diff --git a/llvm/test/CodeGen/AMDGPU/loop-prefetch-data.ll b/llvm/test/CodeGen/AMDGPU/loop-prefetch-data.ll
index fb3c04235b8e4d2..02faca2cc191ce8 100644
--- a/llvm/test/CodeGen/AMDGPU/loop-prefetch-data.ll
+++ b/llvm/test/CodeGen/AMDGPU/loop-prefetch-data.ll
@@ -12,7 +12,6 @@ define amdgpu_kernel void @copy_flat(ptr nocapture %d, ptr nocapture readonly %s
 ; GCN-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_add_nc_u64 s[2:3], s[2:3], 0xb0
-; GCN-NEXT:    .p2align 6
 ; GCN-NEXT:  .LBB0_2: ; %for.body
 ; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
diff --git a/llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll b/llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll
index c0d276f5d88cf7d..5741a7b3851d5d4 100644
--- a/llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll
+++ b/llvm/test/CodeGen/AMDGPU/no-dup-inst-prefetch.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx1030 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX12 %s
 
 define amdgpu_cs void @_amdgpu_cs_main(float %0, i32 %1) {
 ; GFX10-LABEL: _amdgpu_cs_main:
@@ -50,6 +51,55 @@ define amdgpu_cs void @_amdgpu_cs_main(float %0, i32 %1) {
 ; GFX10-NEXT:  .LBB0_4: ; %loop0_merge
 ; GFX10-NEXT:    s_inst_prefetch 0x2
 ; GFX10-NEXT:    s_endpgm
+;
+; GFX12-LABEL: _amdgpu_cs_main:
+; GFX12:       ; %bb.0: ; %branch1_true
+; GFX12-NEXT:    v_mov_b32_e32 v2, 0
+; GFX12-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v1
+; GFX12-NEXT:    v_mov_b32_e32 v1, 0
+; GFX12-NEXT:    s_mov_b32 s4, 0
+; GFX12-NEXT:    s_mov_b32 s1, 0
+; GFX12-NEXT:    ; implicit-def: $sgpr2
+; GFX12-NEXT:    s_branch .LBB0_2
+; GFX12-NEXT:  .LBB0_1: ; %Flow
+; GFX12-NEXT:    ; in Loop: Header=BB0_2 Depth=1
+; GFX12-NEXT:    s_or_b32 exec_lo, exec_lo, s3
+; GFX12-NEXT:    v_mov_b32_e32 v1, v0
+; GFX12-NEXT:    s_and_b32 s0, exec_lo, s2
+; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX12-NEXT:    s_or_b32 s1, s0, s1
+; GFX12-NEXT:    s_and_not1_b32 exec_lo, exec_lo, s1
+; GFX12-NEXT:    s_cbranch_execz .LBB0_4
+; GFX12-NEXT:  .LBB0_2: ; %bb
+; GFX12-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GFX12-NEXT:    s_or_b32 s2, s2, exec_lo
+; GFX12-NEXT:    s_and_saveexec_b32 s3, vcc_lo
+; GFX12-NEXT:    s_cbranch_execz .LBB0_1
+; GFX12-NEXT:  ; %bb.3: ; %branch2_merge
+; GFX12-NEXT:    ; in Loop: Header=BB0_2 Depth=1
+; GFX12-NEXT:    s_mov_b32 s5, s4
+; GFX12-NEXT:    s_mov_b32 s6, s4
+; GFX12-NEXT:    s_mov_b32 s7, s4
+; GFX12-NEXT:    s_mov_b32 s8, s4
+; GFX12-NEXT:    s_mov_b32 s9, s4
+; GFX12-NEXT:    s_mov_b32 s10, s4
+; GFX12-NEXT:    s_mov_b32 s11, s4
+; GFX12-NEXT:    s_mov_b32 s12, s4
+; GFX12-NEXT:    s_mov_b32 s13, s4
+; GFX12-NEXT:    s_mov_b32 s14, s4
+; GFX12-NEXT:    s_mov_b32 s15, s4
+; GFX12-NEXT:    s_and_not1_b32 s2, s2, exec_lo
+; GFX12-NEXT:    image_sample_lz v1, [v2, v2, v1], s[8:15], s[4:7] dmask:0x1 dim:SQ_RSRC_IMG_3D
+; GFX12-NEXT:    s_waitcnt vmcnt(0)
+; GFX12-NEXT:    v_fma_f32 v1, v1, v0, 0
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-NEXT:    v_cmp_le_f32_e64 s0, 0, v1
+; GFX12-NEXT:    s_and_b32 s0, s0, exec_lo
+; GFX12-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX12-NEXT:    s_or_b32 s2, s2, s0
+; GFX12-NEXT:    s_branch .LBB0_1
+; GFX12-NEXT:  .LBB0_4: ; %loop0_merge
+; GFX12-NEXT:    s_endpgm
 branch1_true:
   br label %bb