AMDGPU/SI: Add support for shrinking v_cndmask_b32_e32 instructions

Reviewers: arsenm

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D11061

llvm-svn: 242146

7 files changed, 91 insertions, 89 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/llvm.round.ll b/llvm/test/CodeGen/AMDGPU/llvm.round.ll
index f5f124d915a..d0e49243ffa 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.round.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.round.ll
@@ -9,8 +9,8 @@
 ; SI: v_sub_f32_e32 [[SUB:v[0-9]+]], [[SX]], [[TRUNC]]
 ; SI: v_mov_b32_e32 [[VX:v[0-9]+]], [[SX]]
 ; SI: v_bfi_b32 [[COPYSIGN:v[0-9]+]], [[K]], 1.0, [[VX]]
-; SI: v_cmp_le_f32_e64 [[CMP:s\[[0-9]+:[0-9]+\]]], 0.5, |[[SUB]]|
-; SI: v_cndmask_b32_e64 [[SEL:v[0-9]+]], 0, [[VX]], [[CMP]]
+; SI: v_cmp_le_f32_e64 vcc, 0.5, |[[SUB]]|
+; SI: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, [[VX]]
 ; SI: v_add_f32_e32 [[RESULT:v[0-9]+]], [[SEL]], [[TRUNC]]
 ; SI: buffer_store_dword [[RESULT]]
 
diff --git a/llvm/test/CodeGen/AMDGPU/select-vectors.ll b/llvm/test/CodeGen/AMDGPU/select-vectors.ll
index 59082c65cc8..94758ad84c1 100644
--- a/llvm/test/CodeGen/AMDGPU/select-vectors.ll
+++ b/llvm/test/CodeGen/AMDGPU/select-vectors.ll
@@ -6,10 +6,10 @@
 
 
 ; FUNC-LABEL: {{^}}select_v4i8:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> %a, <4 x i8> %b, i8 %c) nounwind {
   %cmp = icmp eq i8 %c, 0
   %select = select i1 %cmp, <4 x i8> %a, <4 x i8> %b
@@ -18,10 +18,10 @@ define void @select_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> %a, <4 x i8> %b,
 }
 
 ; FUNC-LABEL: {{^}}select_v4i16:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %a, <4 x i16> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x i16> %a, <4 x i16> %b
@@ -30,8 +30,8 @@ define void @select_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %a, <4 x i16>
 }
 
 ; FUNC-LABEL: {{^}}select_v2i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 ; SI: buffer_store_dwordx2
 define void @select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
@@ -41,10 +41,10 @@ define void @select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32>
 }
 
 ; FUNC-LABEL: {{^}}select_v4i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 ; SI: buffer_store_dwordx4
 define void @select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, <4 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
@@ -54,14 +54,14 @@ define void @select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, <4 x i32>
 }
 
 ; FUNC-LABEL: {{^}}select_v8i32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> %a, <8 x i32> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x i32> %a, <8 x i32> %b
@@ -88,14 +88,14 @@ define void @select_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x f
 }
 
 ; FUNC-LABEL: {{^}}select_v8f32:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x float> %a, <8 x float> %b
@@ -104,10 +104,10 @@ define void @select_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x f
 }
 
 ; FUNC-LABEL: {{^}}select_v2f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x double> %a, <2 x double> %b
@@ -116,14 +116,14 @@ define void @select_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x
 }
 
 ; FUNC-LABEL: {{^}}select_v4f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x double> %a, <4 x double> %b
@@ -132,22 +132,22 @@ define void @select_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x
 }
 
 ; FUNC-LABEL: {{^}}select_v8f64:
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
-; SI: v_cndmask_b32_e64
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
+; SI: v_cndmask_b32_e32
 define void @select_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, <8 x double> %b, i32 %c) nounwind {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x double> %a, <8 x double> %b
diff --git a/llvm/test/CodeGen/AMDGPU/select64.ll b/llvm/test/CodeGen/AMDGPU/select64.ll
index 5cebb30dc72..13fb575b2b1 100644
--- a/llvm/test/CodeGen/AMDGPU/select64.ll
+++ b/llvm/test/CodeGen/AMDGPU/select64.ll
@@ -55,8 +55,8 @@ define void @v_select_trunc_i64_2(i32 addrspace(1)* %out, i32 %cond, i64 addrspa
 ; CHECK: s_mov_b32 [[SLO:s[0-9]+]], 0
 ; CHECK-DAG: v_mov_b32_e32 [[VHI:v[0-9]+]], [[SHI]]
 ; CHECK-DAG: v_mov_b32_e32 [[VLO:v[0-9]+]], [[SLO]]
-; CHECK-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, [[VLO]], {{v[0-9]+}}
-; CHECK-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, [[VHI]], {{v[0-9]+}}
+; CHECK-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, [[VLO]], {{v[0-9]+}}
+; CHECK-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, [[VHI]], {{v[0-9]+}}
 ; CHECK: s_endpgm
 define void @v_select_i64_split_imm(i64 addrspace(1)* %out, i32 %cond, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
   %cmp = icmp ugt i32 %cond, 5
diff --git a/llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll b/llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll
index da4e91db3a3..0db7cdc171b 100644
--- a/llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll
@@ -12,11 +12,11 @@ define void @sint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
 
 ; FIXME: select on 0, 0
 ; SI-LABEL: {{^}}sint_to_fp_i1_f64:
-; SI: v_cmp_eq_i32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
+; SI: v_cmp_eq_i32_e64 vcc,
 ; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
-; uses an SGPR for [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, v{{[0-9]+}}, [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, [[CMP]]
+; uses an SGPR (implicit vcc).
+; SI: v_cndmask_b32_e32 v{{[0-9]+}}, 0, v{{[0-9]+}}
+; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, vcc
 ; SI: buffer_store_dwordx2
 ; SI: s_endpgm
 define void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
diff --git a/llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll b/llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll
index dfec8eb15cb..6f608df5e9f 100644
--- a/llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll
@@ -72,11 +72,11 @@ define void @s_uint_to_fp_v4i32_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i
 
 ; FIXME: select on 0, 0
 ; SI-LABEL: {{^}}uint_to_fp_i1_to_f64:
-; SI: v_cmp_eq_i32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
-; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
-; uses an SGPR for [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, v{{[0-9]+}}, [[CMP]]
-; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, [[CMP]]
+; SI: v_cmp_eq_i32_e64 vcc
+; We can't fold the SGPRs into v_cndmask_b32_e32, because it already
+; uses an SGPR (implicit vcc).
+; SI: v_cndmask_b32_e32 v{{[0-9]+}}, 0, v{{[0-9]+}}
+; SI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 0, vcc
 ; SI: buffer_store_dwordx2
 ; SI: s_endpgm
 define void @uint_to_fp_i1_to_f64(double addrspace(1)* %out, i32 %in) {
diff --git a/llvm/test/CodeGen/AMDGPU/vselect.ll b/llvm/test/CodeGen/AMDGPU/vselect.ll
index a3014b03d2b..dc1f1ea11b0 100644
--- a/llvm/test/CodeGen/AMDGPU/vselect.ll
+++ b/llvm/test/CodeGen/AMDGPU/vselect.ll
@@ -1,14 +1,14 @@
-;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG %s
-;RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI %s
-;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=SI %s
+;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG --check-prefix=FUNC %s
+;RUN: llc < %s -march=amdgcn -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
+;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=VI --check-prefix=FUNC %s
+
+;FUNC-LABEL: {{^}}test_select_v2i32:
 
-;EG: {{^}}test_select_v2i32:
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v2i32:
-;SI: v_cndmask_b32_e64
 ;SI: v_cndmask_b32_e64
+;SI: v_cndmask_b32_e32
 
 define void @test_select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in0, <2 x i32> addrspace(1)* %in1) {
 entry:
@@ -20,13 +20,13 @@ entry:
   ret void
 }
 
-;EG: {{^}}test_select_v2f32:
+;FUNC-LABEL: {{^}}test_select_v2f32:
+
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v2f32:
-;SI: v_cndmask_b32_e64
 ;SI: v_cndmask_b32_e64
+;SI: v_cndmask_b32_e32
 
 define void @test_select_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %in0, <2 x float> addrspace(1)* %in1) {
 entry:
@@ -38,17 +38,19 @@ entry:
   ret void
 }
 
-;EG: {{^}}test_select_v4i32:
+;FUNC-LABEL: {{^}}test_select_v4i32:
+
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
-;SI: {{^}}test_select_v4i32:
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
-;SI: v_cndmask_b32_e64
+; FIXME: The shrinking does not happen on tonga
+
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
+;SI: v_cndmask_b32
 
 define void @test_select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in0, <4 x i32> addrspace(1)* %in1) {
 entry:
@@ -60,7 +62,7 @@ entry:
   ret void
 }
 
-;EG: {{^}}test_select_v4f32:
+;FUNC-LABEL: {{^}}test_select_v4f32:
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;EG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
diff --git a/llvm/test/CodeGen/AMDGPU/xor.ll b/llvm/test/CodeGen/AMDGPU/xor.ll
index 089db59eabc..ddb920af29d 100644
--- a/llvm/test/CodeGen/AMDGPU/xor.ll
+++ b/llvm/test/CodeGen/AMDGPU/xor.ll
@@ -42,8 +42,8 @@ define void @xor_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in
 
 ; SI-DAG: v_cmp_le_f32_e32 [[CMP0:vcc]], 0, {{v[0-9]+}}
 ; SI-DAG: v_cmp_le_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]], 1.0, {{v[0-9]+}}
-; SI: s_xor_b64 [[XOR:s\[[0-9]+:[0-9]+\]]], [[CMP0]], [[CMP1]]
-; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, [[XOR]]
+; SI: s_xor_b64 [[XOR:vcc]], [[CMP0]], [[CMP1]]
+; SI: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}
 ; SI: buffer_store_dword [[RESULT]]
 ; SI: s_endpgm
 define void @xor_i1(float addrspace(1)* %out, float addrspace(1)* %in0, float addrspace(1)* %in1) {