Thanks to visit codestin.com
Credit goes to github.com

Skip to content

[AMDGPU] Fix flat load test to use addrspace 0 #138306

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[AMDGPU] Fix flat load test to use addrspace 0 #138306

wants to merge 1 commit into from

Conversation

jayfoad
Copy link
Contributor

@jayfoad jayfoad commented May 2, 2025

No description provided.

@llvmbot
Copy link
Member

llvmbot commented May 2, 2025 •
edited
Loading

@llvm/pr-subscribers-backend-amdgpu

@llvm/pr-subscribers-llvm-globalisel

Author: Jay Foad (jayfoad)

Changes

Patch is 44.60 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/138306.diff

1 Files Affected:

  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir (+171-171)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
index e1325a0a0bc50..9302f1b400e73 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
@@ -59,7 +59,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
     ; GFX12-NEXT: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 0)
     $vgpr0 = COPY %1
 
@@ -117,7 +117,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_USHORT:%[0-9]+]]:vgpr_32 = FLAT_LOAD_USHORT [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s16))
     ; GFX12-NEXT: $vgpr0 = COPY [[FLAT_LOAD_USHORT]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s32) = G_LOAD %0 :: (load (s16), align 2, addrspace 0)
     $vgpr0 = COPY %1
 
@@ -175,7 +175,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = FLAT_LOAD_UBYTE [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8))
     ; GFX12-NEXT: $vgpr0 = COPY [[FLAT_LOAD_UBYTE]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s32) = G_LOAD %0 :: (load (s8), align 1, addrspace 0)
     $vgpr0 = COPY %1
 
@@ -195,16 +195,16 @@ body: |
     ; GFX7-LABEL: name: load_flat_v2s32
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s32>))
-    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x s32>) = G_LOAD [[COPY]](p0) :: (load (<2 x s32>))
+    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
     ;
     ; GFX8-LABEL: name: load_flat_v2s32
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s32>))
-    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x s32>) = G_LOAD [[COPY]](p0) :: (load (<2 x s32>))
+    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
     ;
     ; GFX9-LABEL: name: load_flat_v2s32
     ; GFX9: liveins: $vgpr0_vgpr1
@@ -233,7 +233,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s32>))
     ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(<2 x s32>) = G_LOAD %0 :: (load (<2 x s32>), align 8, addrspace 0)
     $vgpr0_vgpr1 = COPY %1
 
@@ -253,16 +253,16 @@ body: |
     ; GFX7-LABEL: name: load_flat_v3s32
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[FLAT_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = FLAT_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<3 x s32>), align 4)
-    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[FLAT_LOAD_DWORDX3_]]
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<3 x s32>) = G_LOAD [[COPY]](p0) :: (load (<3 x s32>), align 4)
+    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
     ;
     ; GFX8-LABEL: name: load_flat_v3s32
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = FLAT_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<3 x s32>), align 4)
-    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[FLAT_LOAD_DWORDX3_]]
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<3 x s32>) = G_LOAD [[COPY]](p0) :: (load (<3 x s32>), align 4)
+    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<3 x s32>)
     ;
     ; GFX9-LABEL: name: load_flat_v3s32
     ; GFX9: liveins: $vgpr0_vgpr1
@@ -291,7 +291,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = FLAT_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<3 x s32>), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[FLAT_LOAD_DWORDX3_]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(<3 x  s32>) = G_LOAD %0 :: (load (<3 x s32>), align 4, addrspace 0)
     $vgpr0_vgpr1_vgpr2 = COPY %1
 
@@ -311,16 +311,16 @@ body: |
     ; GFX7-LABEL: name: load_flat_v4s32
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4)
-    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load (<4 x s32>), align 4)
+    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
     ;
     ; GFX8-LABEL: name: load_flat_v4s32
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4)
-    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<4 x s32>) = G_LOAD [[COPY]](p0) :: (load (<4 x s32>), align 4)
+    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<4 x s32>)
     ;
     ; GFX9-LABEL: name: load_flat_v4s32
     ; GFX9: liveins: $vgpr0_vgpr1
@@ -349,7 +349,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<4 x s32>), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(<4 x  s32>) = G_LOAD %0 :: (load (<4 x s32>), align 4, addrspace 0)
     $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
 
@@ -369,16 +369,16 @@ body: |
     ; GFX7-LABEL: name: load_flat_s64
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s64))
-    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(s64) = G_LOAD [[COPY]](p0) :: (load (s64))
+    ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
     ;
     ; GFX8-LABEL: name: load_flat_s64
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s64))
-    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(s64) = G_LOAD [[COPY]](p0) :: (load (s64))
+    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](s64)
     ;
     ; GFX9-LABEL: name: load_flat_s64
     ; GFX9: liveins: $vgpr0_vgpr1
@@ -407,7 +407,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORDX2_:%[0-9]+]]:vreg_64 = FLAT_LOAD_DWORDX2 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s64))
     ; GFX12-NEXT: $vgpr0_vgpr1 = COPY [[FLAT_LOAD_DWORDX2_]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s64) = G_LOAD %0 :: (load (s64), align 8, addrspace 0)
     $vgpr0_vgpr1 = COPY %1
 
@@ -427,16 +427,16 @@ body: |
     ; GFX7-LABEL: name: load_flat_v2s64
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s64>), align 4)
-    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load (<2 x s64>), align 4)
+    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x s64>)
     ;
     ; GFX8-LABEL: name: load_flat_v2s64
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s64>), align 4)
-    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x s64>) = G_LOAD [[COPY]](p0) :: (load (<2 x s64>), align 4)
+    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x s64>)
     ;
     ; GFX9-LABEL: name: load_flat_v2s64
     ; GFX9: liveins: $vgpr0_vgpr1
@@ -465,7 +465,7 @@ body: |
     ; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX12-NEXT: [[FLAT_LOAD_DWORDX4_:%[0-9]+]]:vreg_128 = FLAT_LOAD_DWORDX4 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s64>), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[FLAT_LOAD_DWORDX4_]]
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(<2 x s64>) = G_LOAD %0 :: (load (<2 x s64>), align 4, addrspace 0)
     $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
 
@@ -485,45 +485,45 @@ body: |
     ; GFX7-LABEL: name: load_flat_v2p1
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
     ;
     ; GFX8-LABEL: name: load_flat_v2p1
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
     ;
     ; GFX9-LABEL: name: load_flat_v2p1
     ; GFX9: liveins: $vgpr0_vgpr1
     ; GFX9-NEXT: {{  $}}
-    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
     ;
     ; GFX10-LABEL: name: load_flat_v2p1
     ; GFX10: liveins: $vgpr0_vgpr1
     ; GFX10-NEXT: {{  $}}
-    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
     ;
     ; GFX11-LABEL: name: load_flat_v2p1
     ; GFX11: liveins: $vgpr0_vgpr1
     ; GFX11-NEXT: {{  $}}
-    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
     ;
     ; GFX12-LABEL: name: load_flat_v2p1
     ; GFX12: liveins: $vgpr0_vgpr1
     ; GFX12-NEXT: {{  $}}
-    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p1) :: (load (<2 x p1>), align 4)
+    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_128(<2 x p1>) = G_LOAD [[COPY]](p0) :: (load (<2 x p1>), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](<2 x p1>)
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(<2 x p1>) = G_LOAD %0 :: (load (<2 x p1>), align 4, addrspace 0)
     $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
 
@@ -543,45 +543,45 @@ body: |
     ; GFX7-LABEL: name: load_flat_s96
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
     ;
     ; GFX8-LABEL: name: load_flat_s96
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
     ;
     ; GFX9-LABEL: name: load_flat_s96
     ; GFX9: liveins: $vgpr0_vgpr1
     ; GFX9-NEXT: {{  $}}
-    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
     ;
     ; GFX10-LABEL: name: load_flat_s96
     ; GFX10: liveins: $vgpr0_vgpr1
     ; GFX10-NEXT: {{  $}}
-    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
     ;
     ; GFX11-LABEL: name: load_flat_s96
     ; GFX11: liveins: $vgpr0_vgpr1
     ; GFX11-NEXT: {{  $}}
-    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX11-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
     ;
     ; GFX12-LABEL: name: load_flat_s96
     ; GFX12: liveins: $vgpr0_vgpr1
     ; GFX12-NEXT: {{  $}}
-    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4)
+    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p0) :: (load (s96), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s96) = G_LOAD %0 :: (load (s96), align 4, addrspace 0)
     $vgpr0_vgpr1_vgpr2 = COPY %1
 
@@ -601,45 +601,45 @@ body: |
     ; GFX7-LABEL: name: load_flat_s128
     ; GFX7: liveins: $vgpr0_vgpr1
     ; GFX7-NEXT: {{  $}}
-    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
     ;
     ; GFX8-LABEL: name: load_flat_s128
     ; GFX8: liveins: $vgpr0_vgpr1
     ; GFX8-NEXT: {{  $}}
-    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vgpr(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
     ;
     ; GFX9-LABEL: name: load_flat_s128
     ; GFX9: liveins: $vgpr0_vgpr1
     ; GFX9-NEXT: {{  $}}
-    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
     ;
     ; GFX10-LABEL: name: load_flat_s128
     ; GFX10: liveins: $vgpr0_vgpr1
     ; GFX10-NEXT: {{  $}}
-    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
     ;
     ; GFX11-LABEL: name: load_flat_s128
     ; GFX11: liveins: $vgpr0_vgpr1
     ; GFX11-NEXT: {{  $}}
-    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX11-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX11-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
     ;
     ; GFX12-LABEL: name: load_flat_s128
     ; GFX12: liveins: $vgpr0_vgpr1
     ; GFX12-NEXT: {{  $}}
-    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
-    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p1) :: (load (s128), align 4)
+    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
+    ; GFX12-NEXT: [[LOAD:%[0-9]+]]:vreg_128(s128) = G_LOAD [[COPY]](p0) :: (load (s128), align 4)
     ; GFX12-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[LOAD]](s128)
-    %0:vgpr(p1) = COPY $vgpr0_vgpr1
+    %0:vgpr(p0) = COPY $vgpr0_vgpr1
     %1:vgpr(s128) = G_LOAD %0 :: (loa...
[truncated]

jayfoad
jayfoad commented May 2, 2025
View reviewed changes
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
Comment on lines +198 to +200
; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p0) = COPY $vgpr0_vgpr1
; GFX7-NEXT: [[LOAD:%[0-9]+]]:vgpr(<2 x s32>) = G_LOAD [[COPY]](p0) :: (load (<2 x s32>))
; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[LOAD]](<2 x s32>)
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand what happened here.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm guessing it's because FlatLoadPat uses i64:$vaddr. I'm not sure how the pointer types and integer types interact exactly, but this seems like a bug that p1 would work but not p0. Does it work correctly if you change that to p0?

shiltian
shiltian reviewed May 2, 2025
View reviewed changes
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir
@@ -59,7 +59,7 @@ body: |
; GFX12-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX12-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
; GFX12-NEXT: $vgpr0 = COPY [[FLAT_LOAD_DWORD]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%0:vgpr(p0) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 0)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

surprised to see that even with p1, it can generate FLAT_LOAD_DWORD

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The patterns don't know the pointer type, only the memory operand address space

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm arsenm left review comments

@shiltian shiltian shiltian left review comments

Assignees
No one assigned
Labels
backend:AMDGPU llvm:globalisel
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@jayfoad @llvmbot @arsenm @shiltian