[AMDGPU] Fix RP tracker's live registers after processing a memory clause.

It's incorrect to reuse live registers left from the first instruction in a clause after the clause as they don't contain in-clause defs.

Reviewed By: arsenm

Differential Revision: https://reviews.llvm.org/D137081

diff --git a/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp b/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
index d7ca7f3..471d70a 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
@@ -393,13 +393,11 @@ bool SIFormMemoryClauses::runOnMachineFunction(MachineFunction &MF) {
         Ind->insertMachineInstrInMaps(*Kill);
       }
 
-      if (!Kill) {
-        RPT.reset(MI, &LiveRegsCopy);
-        continue;
-      }
-
       // Restore the state after processing the end of the bundle.
-      RPT.reset(*Kill, &LiveRegsCopy);
+      RPT.reset(MI, &LiveRegsCopy);
+
+      if (!Kill)
+        continue;
 
       for (auto &&R : Defs) {
         Register Reg = R.first;

diff --git a/llvm/test/CodeGen/AMDGPU/spill-vgpr.ll b/llvm/test/CodeGen/AMDGPU/spill-vgpr.ll
index ea2b5f3..e892f41 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/spill-vgpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/spill-vgpr.ll
@@ -85,12 +85,13 @@ define amdgpu_kernel void @max_10_vgprs_spill_v32(<32 x float> addrspace(1)* %p)
 ; GFX908-DAG: v_accvgpr_read_b32
 
 ; GFX900: NumVgprs: 256
-; GFX900: ScratchSize: 148
-; GFX908: NumVgprs: 254
+; GFX900: ScratchSize: 132
+; GFX908: NumVgprs: 252
 ; GFX908: ScratchSize: 0
-; GCN:    VGPRBlocks: 63
+; GCN900:    VGPRBlocks: 63
+; GCN908:    VGPRBlocks: 62
 ; GFX900:    NumVGPRsForWavesPerEU: 256
-; GFX908:    NumVGPRsForWavesPerEU: 254
+; GFX908:    NumVGPRsForWavesPerEU: 252
 define amdgpu_kernel void @max_256_vgprs_spill_9x32(<32 x float> addrspace(1)* %p) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %p1 = getelementptr inbounds <32 x float>, <32 x float> addrspace(1)* %p, i32 %tid
@@ -135,13 +136,13 @@ define amdgpu_kernel void @max_256_vgprs_spill_9x32(<32 x float> addrspace(1)* %
 ; GFX908-DAG: v_accvgpr_read_b32
 
 ; GFX900: NumVgprs: 256
-; GFX908: NumVgprs: 254
-; GFX900: ScratchSize: 1796
+; GFX908: NumVgprs: 252
+; GFX900: ScratchSize: 1668
 ; GFX908: ScratchSize: 0
 ; GFX900: VGPRBlocks: 63
-; GFX908: VGPRBlocks: 63
+; GFX908: VGPRBlocks: 62
 ; GFX900: NumVGPRsForWavesPerEU: 256
-; GFX908: NumVGPRsForWavesPerEU: 25
+; GFX908: NumVGPRsForWavesPerEU: 252
 define amdgpu_kernel void @max_256_vgprs_spill_9x32_2bb(<32 x float> addrspace(1)* %p) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %p1 = getelementptr inbounds <32 x float>, <32 x float> addrspace(1)* %p, i32 %tid

diff --git a/llvm/test/CodeGen/AMDGPU/unallocatable-bundle-regression.mir b/llvm/test/CodeGen/AMDGPU/unallocatable-bundle-regression.mir
index 97c63bd..676cbc5 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/unallocatable-bundle-regression.mir
+++ b/llvm/test/CodeGen/AMDGPU/unallocatable-bundle-regression.mir
@@ -52,13 +52,11 @@ body:             |
     ; CHECK-NEXT: renamable $sgpr0 = S_MOV_B32 1264
     ; CHECK-NEXT: renamable $sgpr1 = COPY renamable $sgpr5
     ; CHECK-NEXT: renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43 = S_LOAD_DWORDX8_IMM renamable $sgpr4_sgpr5, 1216, 0 :: (dereferenceable load (s256), addrspace 6)
-    ; CHECK-NEXT: renamable $sgpr88_sgpr89_sgpr90_sgpr91 = S_LOAD_DWORDX4_IMM renamable $sgpr0_sgpr1, 0, 0 :: (load (s128), addrspace 6)
-    ; CHECK-NEXT: KILL killed renamable $sgpr0, renamable $sgpr1
+    ; CHECK-NEXT: renamable $sgpr88_sgpr89_sgpr90_sgpr91 = S_LOAD_DWORDX4_IMM killed renamable $sgpr0_sgpr1, 0, 0 :: (load (s128), addrspace 6)
     ; CHECK-NEXT: renamable $sgpr0 = S_MOV_B32 1328
     ; CHECK-NEXT: renamable $sgpr1 = COPY renamable $sgpr5
     ; CHECK-NEXT: renamable $sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51 = S_LOAD_DWORDX8_IMM renamable $sgpr4_sgpr5, 1280, 0 :: (dereferenceable load (s256), addrspace 6)
-    ; CHECK-NEXT: renamable $sgpr92_sgpr93_sgpr94_sgpr95 = S_LOAD_DWORDX4_IMM renamable $sgpr0_sgpr1, 0, 0 :: (load (s128), addrspace 6)
-    ; CHECK-NEXT: KILL killed renamable $sgpr0, renamable $sgpr1
+    ; CHECK-NEXT: renamable $sgpr92_sgpr93_sgpr94_sgpr95 = S_LOAD_DWORDX4_IMM killed renamable $sgpr0_sgpr1, 0, 0 :: (load (s128), addrspace 6)
     ; CHECK-NEXT: renamable $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59 = S_LOAD_DWORDX8_IMM renamable $sgpr4_sgpr5, 1344, 0 :: (dereferenceable load (s256), addrspace 6)
     ; CHECK-NEXT: renamable $sgpr0 = S_MOV_B32 1392
     ; CHECK-NEXT: renamable $sgpr1 = COPY renamable $sgpr5
@@ -73,25 +71,19 @@ body:             |
     ; CHECK-NEXT: renamable $vgpr7 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15, killed renamable $sgpr76_sgpr77_sgpr78_sgpr79, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
     ; CHECK-NEXT: renamable $vgpr8 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr16_sgpr17_sgpr18_sgpr19_sgpr20_sgpr21_sgpr22_sgpr23, killed renamable $sgpr80_sgpr81_sgpr82_sgpr83, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
     ; CHECK-NEXT: renamable $vgpr9 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr24_sgpr25_sgpr26_sgpr27_sgpr28_sgpr29_sgpr30_sgpr31, killed renamable $sgpr84_sgpr85_sgpr86_sgpr87, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
-    ; CHECK-NEXT: renamable $vgpr10 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43, renamable $sgpr88_sgpr89_sgpr90_sgpr91, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
-    ; CHECK-NEXT: renamable $vgpr11 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, renamable $sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, renamable $sgpr92_sgpr93_sgpr94_sgpr95, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
-    ; CHECK-NEXT: renamable $vgpr12 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, renamable $sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, renamable $sgpr96_sgpr97_sgpr98_sgpr99, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
+    ; CHECK-NEXT: renamable $vgpr10 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43, killed renamable $sgpr88_sgpr89_sgpr90_sgpr91, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
+    ; CHECK-NEXT: renamable $vgpr11 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51, killed renamable $sgpr92_sgpr93_sgpr94_sgpr95, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
+    ; CHECK-NEXT: renamable $vgpr12 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, killed renamable $sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67, killed renamable $sgpr96_sgpr97_sgpr98_sgpr99, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
     ; CHECK-NEXT: renamable $vgpr13 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, renamable $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75, renamable $sgpr4_sgpr5_sgpr6_sgpr7, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
     ; CHECK-NEXT: renamable $vgpr14 = IMAGE_SAMPLE_LZ_V1_V2_gfx10 renamable $vgpr5_vgpr6, renamable $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59, renamable $sgpr0_sgpr1_sgpr2_sgpr3, 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "ImageResource")
     ; CHECK-NEXT: renamable $sgpr8_sgpr9_sgpr10_sgpr11 = SI_SPILL_S128_RESTORE %stack.0, implicit $exec, implicit $sgpr32 :: (load (s128) from %stack.0, align 4, addrspace 5)
     ; CHECK-NEXT: renamable $vgpr1_vgpr2_vgpr3_vgpr4 = BUFFER_LOAD_FORMAT_XYZW_IDXEN renamable $vgpr0, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s128) from custom "BufferResource", align 1, addrspace 4)
     ; CHECK-NEXT: KILL killed renamable $sgpr4_sgpr5_sgpr6_sgpr7
-    ; CHECK-NEXT: KILL killed renamable $sgpr92_sgpr93_sgpr94_sgpr95
     ; CHECK-NEXT: KILL killed renamable $sgpr68_sgpr69_sgpr70_sgpr71_sgpr72_sgpr73_sgpr74_sgpr75
-    ; CHECK-NEXT: KILL killed renamable $sgpr60_sgpr61_sgpr62_sgpr63_sgpr64_sgpr65_sgpr66_sgpr67
-    ; CHECK-NEXT: KILL killed renamable $sgpr36_sgpr37_sgpr38_sgpr39_sgpr40_sgpr41_sgpr42_sgpr43
     ; CHECK-NEXT: KILL killed renamable $vgpr5_vgpr6
     ; CHECK-NEXT: KILL killed renamable $sgpr0_sgpr1_sgpr2_sgpr3
     ; CHECK-NEXT: KILL killed renamable $sgpr52_sgpr53_sgpr54_sgpr55_sgpr56_sgpr57_sgpr58_sgpr59
-    ; CHECK-NEXT: KILL killed renamable $sgpr96_sgpr97_sgpr98_sgpr99
-    ; CHECK-NEXT: KILL killed renamable $sgpr44_sgpr45_sgpr46_sgpr47_sgpr48_sgpr49_sgpr50_sgpr51
     ; CHECK-NEXT: KILL killed renamable $sgpr8_sgpr9_sgpr10_sgpr11
-    ; CHECK-NEXT: KILL killed renamable $sgpr88_sgpr89_sgpr90_sgpr91
     ; CHECK-NEXT: KILL killed renamable $vgpr0
     ; CHECK-NEXT: renamable $vgpr0 = nofpexcept V_MAX_F32_e32 killed $vgpr7, killed $vgpr8, implicit $mode, implicit $exec
     ; CHECK-NEXT: renamable $vgpr0 = V_MAX3_F32_e64 0, killed $vgpr0, 0, killed $vgpr9, 0, killed $vgpr10, 0, 0, implicit $mode, implicit $exec