AMDGPU/SI: Extend promoting alloca to vector to arrays of up to 16 elements

Summary:
  This patch extends the promotion of alloca to vector to the arrays of up to 16 elements. Also we introduce
an option, -disable-promote-alloca-to-vector, to switch promotion to vector off, if needed.

Reviewers:
  arsenm

Differential Revision:
  https://reviews.llvm.org/D33559

llvm-svn: 325372

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index 419d48165e517a218fcd6ec4c5efa49bbb03a416..cf8469f14ba7b95979760880ed022353cfe5b0ea 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -65,6 +65,11 @@ using namespace llvm;
 
 namespace {
 
+static cl::opt<bool> DisablePromoteAllocaToVector(
+  "disable-promote-alloca-to-vector",
+  cl::desc("Disable promote alloca to vector"),
+  cl::init(false));
+
 // FIXME: This can create globals so should be a module pass.
 class AMDGPUPromoteAlloca : public FunctionPass {
 private:
@@ -337,6 +342,12 @@ static bool canVectorizeInst(Instruction *Inst, User *User) {
 }
 
 static bool tryPromoteAllocaToVector(AllocaInst *Alloca, AMDGPUAS AS) {
+
+  if (DisablePromoteAllocaToVector) {
+    DEBUG(dbgs() << "  Promotion alloca to vector is disabled\n");
+    return false;
+  }
+
   ArrayType *AllocaTy = dyn_cast<ArrayType>(Alloca->getAllocatedType());
 
   DEBUG(dbgs() << "Alloca candidate for vectorization\n");
@@ -346,7 +357,7 @@ static bool tryPromoteAllocaToVector(AllocaInst *Alloca, AMDGPUAS AS) {
   // FIXME: We also reject alloca's of the form [ 2 x [ 2 x i32 ]] or equivalent. Potentially these
   // could also be promoted but we don't currently handle this case
   if (!AllocaTy ||
-      AllocaTy->getNumElements() > 4 ||
+      AllocaTy->getNumElements() > 16 ||
       AllocaTy->getNumElements() < 2 ||
       !VectorType::isValidElementType(AllocaTy->getElementType())) {
     DEBUG(dbgs() << "  Cannot convert type to vector\n");

diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu.private-memory.ll b/llvm/test/CodeGen/AMDGPU/amdgpu.private-memory.ll
index f37ba76fc84698bd215934ca038cbdf6382399d1..a8a5e1a1b487c190da8bae423890d756cb08886c 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/amdgpu.private-memory.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu.private-memory.ll
@@ -1,15 +1,16 @@
-; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -march=amdgcn < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE %s
+; RUN: llc -show-mc-encoding -mattr=+promote-alloca -disable-promote-alloca-to-vector -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -march=amdgcn < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -show-mc-encoding -mattr=+promote-alloca -disable-promote-alloca-to-vector -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-PROMOTE %s
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -march=amdgcn < %s | FileCheck %s -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -mcpu=kaveri -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC -check-prefix=HSA-ALLOCA %s
-; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -show-mc-encoding -mattr=+promote-alloca -disable-promote-alloca-to-vector -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -show-mc-encoding -mattr=+promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-PROMOTE-VECT -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -show-mc-encoding -mattr=-promote-alloca -amdgpu-load-store-vectorizer=0 -enable-amdgpu-aa=0 -verify-machineinstrs -mtriple=amdgcn-amdhsa -march=amdgcn -mcpu=tonga -mattr=-unaligned-buffer-access < %s | FileCheck -enable-var-scope -check-prefix=SI-ALLOCA -check-prefix=SI -check-prefix=FUNC %s
 
-; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -data-layout=A5 -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -enable-var-scope -check-prefix=HSAOPT -check-prefix=OPT %s
-; RUN: opt -S -mtriple=amdgcn-unknown-unknown -data-layout=A5 -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck -enable-var-scope -check-prefix=NOHSAOPT -check-prefix=OPT %s
-
-; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -data-layout=A5 -mcpu=kaveri -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck -enable-var-scope -check-prefix=HSAOPT -check-prefix=OPT %s
+; RUN: opt -S -mtriple=amdgcn-unknown-unknown -data-layout=A5 -mcpu=kaveri -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck -enable-var-scope -check-prefix=NOHSAOPT -check-prefix=OPT %s
 
+; RUN: llc -march=r600 -mcpu=cypress -disable-promote-alloca-to-vector < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
+; RUN: llc -march=r600 -mcpu=cypress < %s | FileCheck %s -check-prefix=R600-VECT -check-prefix=FUNC
 
 ; HSAOPT: @mova_same_clause.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4
 ; HSAOPT: @high_alignment.stack = internal unnamed_addr addrspace(3) global [256 x [8 x i32]] undef, align 16
@@ -223,16 +224,16 @@ for.end:
 
 ; FUNC-LABEL: {{^}}short_array:
 
-; R600: MOVA_INT
+; R600-VECT: MOVA_INT
 
 ; SI-ALLOCA-DAG: buffer_store_short v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:6 ; encoding: [0x06,0x00,0x68,0xe0
 ; SI-ALLOCA-DAG: buffer_store_short v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding: [0x04,0x00,0x68,0xe0
 ; Loaded value is 0 or 1, so sext will become zext, so we get buffer_load_ushort instead of buffer_load_sshort.
 ; SI-ALLOCA: buffer_load_sshort v{{[0-9]+}}, v{{[0-9]+}}, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}}
 
-; SI-PROMOTE: s_load_dword [[IDX:s[0-9]+]]
-; SI-PROMOTE: s_lshl_b32 [[SCALED_IDX:s[0-9]+]], [[IDX]], 16
-; SI-PROMOTE: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[SCALED_IDX]], 16
+; SI-PROMOTE-VECT: s_load_dword [[IDX:s[0-9]+]]
+; SI-PROMOTE-VECT: s_lshl_b32 [[SCALED_IDX:s[0-9]+]], [[IDX]], 16
+; SI-PROMOTE-VECT: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, [[SCALED_IDX]], 16
 define amdgpu_kernel void @short_array(i32 addrspace(1)* %out, i32 %index) #0 {
 entry:
   %0 = alloca [2 x i16], addrspace(5)
@@ -249,10 +250,10 @@ entry:
 
 ; FUNC-LABEL: {{^}}char_array:
 
-; R600: MOVA_INT
+; R600-VECT: MOVA_INT
 
-; SI-PROMOTE-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding:
-; SI-PROMOTE-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:5 ; encoding:
+; SI-PROMOTE-VECT-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding:
+; SI-PROMOTE-VECT-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:5 ; encoding:
 
 ; SI-ALLOCA-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:4 ; encoding: [0x04,0x00,0x60,0xe0
 ; SI-ALLOCA-DAG: buffer_store_byte v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s{{[0-9]+}} offset:5 ; encoding: [0x05,0x00,0x60,0xe0

diff --git a/llvm/test/CodeGen/AMDGPU/indirect-private-64.ll b/llvm/test/CodeGen/AMDGPU/indirect-private-64.ll
index b3a33968fbdad468dfac8cd22f6940e4fe1f4e76..699f2060bf26a4f8abc59bb4319ed28440eed84d 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/indirect-private-64.ll
+++ b/llvm/test/CodeGen/AMDGPU/indirect-private-64.ll
@@ -1,8 +1,8 @@
 ; RUN: llc -march=amdgcn -mattr=-promote-alloca,+max-private-element-size-16 -verify-machineinstrs < %s | FileCheck -check-prefix=SI-ALLOCA16 -check-prefix=SI %s
 ; RUN: llc -march=amdgcn -mattr=-promote-alloca,+max-private-element-size-4 -verify-machineinstrs < %s | FileCheck -check-prefix=SI-ALLOCA4 -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mattr=+promote-alloca -disable-promote-alloca-to-vector -verify-machineinstrs < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -mattr=-promote-alloca,+max-private-element-size-16 -verify-machineinstrs < %s | FileCheck -check-prefix=CI-ALLOCA16 -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -mattr=+promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefix=CI-PROMOTE -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -mattr=+promote-alloca -disable-promote-alloca-to-vector -verify-machineinstrs < %s | FileCheck -check-prefix=CI-PROMOTE -check-prefix=SI %s
 
 declare void @llvm.amdgcn.s.barrier() #0
 

diff --git a/llvm/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll b/llvm/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
index 13dd7058c50a5966c02f068d24c2ab4d280db4e7..4fbf5e657dffcce4aeec74d19810b7deb97aede5 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
+++ b/llvm/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
@@ -1,5 +1,5 @@
-; RUN: opt -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck --check-prefix=SI --check-prefix=ALL %s
-; RUN: opt -S -mcpu=tonga -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck --check-prefix=CI --check-prefix=ALL %s
+; RUN: opt -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck --check-prefix=SI --check-prefix=ALL %s
+; RUN: opt -S -mcpu=tonga -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck --check-prefix=CI --check-prefix=ALL %s
 
 ; SI-NOT: @promote_alloca_size_63.stack = internal unnamed_addr addrspace(3) global [63 x [5 x i32]] undef, align 4
 ; CI: @promote_alloca_size_63.stack = internal unnamed_addr addrspace(3) global [63 x [5 x i32]] undef, align 4

diff --git a/llvm/test/CodeGen/AMDGPU/private-memory-r600.ll b/llvm/test/CodeGen/AMDGPU/private-memory-r600.ll
index 65e728174291a2fd71c84e30e6c58cbf6499a39e..5e03b8a6e410bfea6f7e53bee1e98219646a8627 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/private-memory-r600.ll
+++ b/llvm/test/CodeGen/AMDGPU/private-memory-r600.ll
@@ -1,5 +1,6 @@
-; RUN: llc -march=r600 -mtriple=r600---amdgiz -mcpu=redwood < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
-; RUN: opt -S -mtriple=r600-unknown-unknown-amdgiz -mcpu=redwood -amdgpu-promote-alloca < %s | FileCheck -check-prefix=OPT %s
+; RUN: llc -march=r600 -mcpu=redwood -disable-promote-alloca-to-vector < %s | FileCheck %s -check-prefix=R600 -check-prefix=FUNC
+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s -check-prefix=R600-VECT -check-prefix=FUNC
+; RUN: opt -S -mtriple=r600-unknown-unknown -mcpu=redwood -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck -check-prefix=OPT %s
 target datalayout = "A5"
 
 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
@@ -110,7 +111,7 @@ for.end:
 
 ; FUNC-LABEL: {{^}}short_array:
 
-; R600: MOVA_INT
+; R600-VECT: MOVA_INT
 define amdgpu_kernel void @short_array(i32 addrspace(1)* %out, i32 %index) #0 {
 entry:
   %0 = alloca [2 x i16], addrspace(5)
@@ -127,7 +128,7 @@ entry:
 
 ; FUNC-LABEL: {{^}}char_array:
 
-; R600: MOVA_INT
+; R600-VECT: MOVA_INT
 define amdgpu_kernel void @char_array(i32 addrspace(1)* %out, i32 %index) #0 {
 entry:
   %0 = alloca [2 x i8], addrspace(5)

diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-calling-conv.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-calling-conv.ll
index 51aacf56a40f013690a7d7eea0ee85aa1bbcacf3..2db95cd820f42ff0eaeaf01d290c2505edd3c588 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-calling-conv.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-calling-conv.ll
@@ -1,5 +1,5 @@
-; RUN: opt -data-layout=A5 -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck -check-prefix=IR %s
-; RUN: llc -march=amdgcn -mcpu=fiji < %s | FileCheck -check-prefix=ASM %s
+; RUN: opt -data-layout=A5 -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck -check-prefix=IR %s
+; RUN: llc -march=amdgcn -mcpu=fiji -disable-promote-alloca-to-vector < %s | FileCheck -check-prefix=ASM %s
 
 ; IR-LABEL: define amdgpu_vs void @promote_alloca_shaders(i32 addrspace(1)* inreg %out, i32 addrspace(1)* inreg %in) #0 {
 ; IR: alloca [5 x i32]
@@ -32,7 +32,7 @@ entry:
 
 ; ASM-LABEL: {{^}}promote_to_vector_call_c:
 ; ASM-NOT: LDSByteSize
-; ASM: ; ScratchSize: 0
+; ASM: ; ScratchSize: 12
 define void @promote_to_vector_call_c(i32 addrspace(1)* %out, i32 %in) #0 {
 entry:
   %tmp = alloca [2 x i32], addrspace(5)

diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-padding-size-estimate.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-padding-size-estimate.ll
index 445bfa5c3ed3aa3a4c42aa807b28f35674100495..e8dcb50a3c1fe400275464db7584aea4d3b639d5 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-padding-size-estimate.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-padding-size-estimate.ll
@@ -1,4 +1,4 @@
-; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri -disable-promote-alloca-to-vector < %s | FileCheck -check-prefix=GCN %s
 
 ; This shows that the amount of LDS estimate is sensitive to the order
 ; of the LDS globals.

diff --git a/llvm/test/CodeGen/AMDGPU/target-cpu.ll b/llvm/test/CodeGen/AMDGPU/target-cpu.ll
index bf1525e6163217082205a5b7032c791144094ecf..cc4c98bb6789129ce7d27f2d037ef37f005e09ea 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/target-cpu.ll
+++ b/llvm/test/CodeGen/AMDGPU/target-cpu.ll
@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -march=amdgcn -disable-promote-alloca-to-vector -verify-machineinstrs < %s | FileCheck %s
 
 declare i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr() #1