From 01adf1f3e50a87e290cedb47fdf9d1e0cc635015 Mon Sep 17 00:00:00 2001 From: Matt Arsenault Date: Tue, 20 Sep 2022 17:57:40 -0400 Subject: [PATCH] AtomicExpand: Add some more overaligned atomic tests --- .../AtomicExpand/AMDGPU/expand-atomic-i16.ll | 64 ++++++++++++ .../AtomicExpand/AMDGPU/expand-atomic-i8.ll | 110 +++++++++++++++++++++ .../AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll | 17 ++++ .../AtomicExpand/AMDGPU/expand-atomic-rmw-fsub.ll | 27 +++-- 4 files changed, 209 insertions(+), 9 deletions(-) diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i16.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i16.ll index b14f8f8..3dbce74 100644 --- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i16.ll +++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i16.ll @@ -455,6 +455,50 @@ define i16 @test_cmpxchg_i16_global(i16 addrspace(1)* %out, i16 %in, i16 %old) { ret i16 %extract } +define i16 @test_cmpxchg_i16_global_align4(i16 addrspace(1)* %out, i16 %in, i16 %old) { +; CHECK-LABEL: @test_cmpxchg_i16_global_align4( +; CHECK-NEXT: [[GEP:%.*]] = getelementptr i16, i16 addrspace(1)* [[OUT:%.*]], i64 4 +; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i16 addrspace(1)* @llvm.ptrmask.p1i16.i64(i16 addrspace(1)* [[GEP]], i64 -4) +; CHECK-NEXT: [[ALIGNEDADDR1:%.*]] = bitcast i16 addrspace(1)* [[ALIGNEDADDR]] to i32 addrspace(1)* +; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint i16 addrspace(1)* [[GEP]] to i64 +; CHECK-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 +; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 +; CHECK-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 +; CHECK-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]] +; CHECK-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 +; CHECK-NEXT: [[TMP3:%.*]] = zext i16 [[IN:%.*]] to i32 +; CHECK-NEXT: [[TMP4:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP5:%.*]] = zext i16 [[OLD:%.*]] to i32 +; CHECK-NEXT: [[TMP6:%.*]] = shl i32 [[TMP5]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32 addrspace(1)* [[ALIGNEDADDR1]], align 4 +; CHECK-NEXT: [[TMP8:%.*]] = and i32 [[TMP7]], [[INV_MASK]] +; CHECK-NEXT: br label [[PARTWORD_CMPXCHG_LOOP:%.*]] +; CHECK: partword.cmpxchg.loop: +; CHECK-NEXT: [[TMP9:%.*]] = phi i32 [ [[TMP8]], [[TMP0:%.*]] ], [ [[TMP15:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ] +; CHECK-NEXT: [[TMP10:%.*]] = or i32 [[TMP9]], [[TMP4]] +; CHECK-NEXT: [[TMP11:%.*]] = or i32 [[TMP9]], [[TMP6]] +; CHECK-NEXT: [[TMP12:%.*]] = cmpxchg i32 addrspace(1)* [[ALIGNEDADDR1]], i32 [[TMP11]], i32 [[TMP10]] seq_cst seq_cst, align 4 +; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { i32, i1 } [[TMP12]], 0 +; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { i32, i1 } [[TMP12]], 1 +; CHECK-NEXT: br i1 [[TMP14]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]] +; CHECK: partword.cmpxchg.failure: +; CHECK-NEXT: [[TMP15]] = and i32 [[TMP13]], [[INV_MASK]] +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP9]], [[TMP15]] +; CHECK-NEXT: br i1 [[TMP16]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]] +; CHECK: partword.cmpxchg.end: +; CHECK-NEXT: [[SHIFTED:%.*]] = lshr i32 [[TMP13]], [[SHIFTAMT]] +; CHECK-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 +; CHECK-NEXT: [[TMP17:%.*]] = insertvalue { i16, i1 } undef, i16 [[EXTRACTED]], 0 +; CHECK-NEXT: [[TMP18:%.*]] = insertvalue { i16, i1 } [[TMP17]], i1 [[TMP14]], 1 +; CHECK-NEXT: [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP18]], 0 +; CHECK-NEXT: ret i16 [[EXTRACT]] +; + %gep = getelementptr i16, i16 addrspace(1)* %out, i64 4 + %res = cmpxchg i16 addrspace(1)* %gep, i16 %old, i16 %in seq_cst seq_cst, align 4 + %extract = extractvalue {i16, i1} %res, 0 + ret i16 %extract +} + define i16 @test_atomicrmw_xchg_i16_local(i16 addrspace(3)* %ptr, i16 %value) { ; CHECK-LABEL: @test_atomicrmw_xchg_i16_local( ; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i16 addrspace(3)* @llvm.ptrmask.p3i16.i32(i16 addrspace(3)* [[PTR:%.*]], i32 -4) @@ -527,3 +571,23 @@ define i16 @test_cmpxchg_i16_local(i16 addrspace(3)* %out, i16 %in, i16 %old) { %extract = extractvalue {i16, i1} %res, 0 ret i16 %extract } + +define i16 @test_atomicrmw_xor_i16_local_align4(i16 addrspace(3)* %ptr, i16 %value) { +; CHECK-LABEL: @test_atomicrmw_xor_i16_local_align4( +; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i16 addrspace(3)* @llvm.ptrmask.p3i16.i32(i16 addrspace(3)* [[PTR:%.*]], i32 -4) +; CHECK-NEXT: [[ALIGNEDADDR1:%.*]] = bitcast i16 addrspace(3)* [[ALIGNEDADDR]] to i32 addrspace(3)* +; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint i16 addrspace(3)* [[PTR]] to i32 +; CHECK-NEXT: [[PTRLSB:%.*]] = and i32 [[TMP1]], 3 +; CHECK-NEXT: [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3 +; CHECK-NEXT: [[MASK:%.*]] = shl i32 65535, [[TMP2]] +; CHECK-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 +; CHECK-NEXT: [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32 +; CHECK-NEXT: [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = atomicrmw xor i32 addrspace(3)* [[ALIGNEDADDR1]], i32 [[VALOPERAND_SHIFTED]] seq_cst, align 4 +; CHECK-NEXT: [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[TMP2]] +; CHECK-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16 +; CHECK-NEXT: ret i16 [[EXTRACTED]] +; + %res = atomicrmw xor i16 addrspace(3)* %ptr, i16 %value seq_cst, align 4 + ret i16 %res +} diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i8.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i8.ll index f937de9..0e465a1 100644 --- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i8.ll +++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-i8.ll @@ -66,6 +66,72 @@ define i8 @test_atomicrmw_add_i8_global(i8 addrspace(1)* %ptr, i8 %value) { ret i8 %res } +define i8 @test_atomicrmw_add_i8_global_align2(i8 addrspace(1)* %ptr, i8 %value) { +; CHECK-LABEL: @test_atomicrmw_add_i8_global_align2( +; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i8 addrspace(1)* @llvm.ptrmask.p1i8.i64(i8 addrspace(1)* [[PTR:%.*]], i64 -4) +; CHECK-NEXT: [[ALIGNEDADDR1:%.*]] = bitcast i8 addrspace(1)* [[ALIGNEDADDR]] to i32 addrspace(1)* +; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint i8 addrspace(1)* [[PTR]] to i64 +; CHECK-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 +; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 +; CHECK-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 +; CHECK-NEXT: [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]] +; CHECK-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 +; CHECK-NEXT: [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32 +; CHECK-NEXT: [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32 addrspace(1)* [[ALIGNEDADDR1]], align 4 +; CHECK-NEXT: br label [[ATOMICRMW_START:%.*]] +; CHECK: atomicrmw.start: +; CHECK-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] +; CHECK-NEXT: [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]] +; CHECK-NEXT: [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]] +; CHECK-NEXT: [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]] +; CHECK-NEXT: [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]] +; CHECK-NEXT: [[TMP8:%.*]] = cmpxchg i32 addrspace(1)* [[ALIGNEDADDR1]], i32 [[LOADED]], i32 [[TMP7]] seq_cst seq_cst, align 4 +; CHECK-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1 +; CHECK-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0 +; CHECK-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] +; CHECK: atomicrmw.end: +; CHECK-NEXT: [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] +; CHECK-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8 +; CHECK-NEXT: ret i8 [[EXTRACTED]] +; + %res = atomicrmw add i8 addrspace(1)* %ptr, i8 %value seq_cst, align 2 + ret i8 %res +} + +define i8 @test_atomicrmw_add_i8_global_align4(i8 addrspace(1)* %ptr, i8 %value) { +; CHECK-LABEL: @test_atomicrmw_add_i8_global_align4( +; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i8 addrspace(1)* @llvm.ptrmask.p1i8.i64(i8 addrspace(1)* [[PTR:%.*]], i64 -4) +; CHECK-NEXT: [[ALIGNEDADDR1:%.*]] = bitcast i8 addrspace(1)* [[ALIGNEDADDR]] to i32 addrspace(1)* +; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint i8 addrspace(1)* [[PTR]] to i64 +; CHECK-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 +; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 +; CHECK-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 +; CHECK-NEXT: [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]] +; CHECK-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 +; CHECK-NEXT: [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32 +; CHECK-NEXT: [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP4:%.*]] = load i32, i32 addrspace(1)* [[ALIGNEDADDR1]], align 4 +; CHECK-NEXT: br label [[ATOMICRMW_START:%.*]] +; CHECK: atomicrmw.start: +; CHECK-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ] +; CHECK-NEXT: [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]] +; CHECK-NEXT: [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]] +; CHECK-NEXT: [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]] +; CHECK-NEXT: [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]] +; CHECK-NEXT: [[TMP8:%.*]] = cmpxchg i32 addrspace(1)* [[ALIGNEDADDR1]], i32 [[LOADED]], i32 [[TMP7]] seq_cst seq_cst, align 4 +; CHECK-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1 +; CHECK-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0 +; CHECK-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]] +; CHECK: atomicrmw.end: +; CHECK-NEXT: [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]] +; CHECK-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8 +; CHECK-NEXT: ret i8 [[EXTRACTED]] +; + %res = atomicrmw add i8 addrspace(1)* %ptr, i8 %value seq_cst, align 4 + ret i8 %res +} + define i8 @test_atomicrmw_sub_i8_global(i8 addrspace(1)* %ptr, i8 %value) { ; CHECK-LABEL: @test_atomicrmw_sub_i8_global( ; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i8 addrspace(1)* @llvm.ptrmask.p1i8.i64(i8 addrspace(1)* [[PTR:%.*]], i64 -4) @@ -388,3 +454,47 @@ define i8 @test_cmpxchg_i8_global(i8 addrspace(1)* %out, i8 %in, i8 %old) { %extract = extractvalue {i8, i1} %res, 0 ret i8 %extract } + +define i8 @test_cmpxchg_i8_local_align2(i8 addrspace(3)* %out, i8 %in, i8 %old) { +; CHECK-LABEL: @test_cmpxchg_i8_local_align2( +; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, i8 addrspace(3)* [[OUT:%.*]], i64 4 +; CHECK-NEXT: [[ALIGNEDADDR:%.*]] = call i8 addrspace(3)* @llvm.ptrmask.p3i8.i64(i8 addrspace(3)* [[GEP]], i64 -4) +; CHECK-NEXT: [[ALIGNEDADDR1:%.*]] = bitcast i8 addrspace(3)* [[ALIGNEDADDR]] to i32 addrspace(3)* +; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint i8 addrspace(3)* [[GEP]] to i64 +; CHECK-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 3 +; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3 +; CHECK-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32 +; CHECK-NEXT: [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]] +; CHECK-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -1 +; CHECK-NEXT: [[TMP3:%.*]] = zext i8 [[IN:%.*]] to i32 +; CHECK-NEXT: [[TMP4:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP5:%.*]] = zext i8 [[OLD:%.*]] to i32 +; CHECK-NEXT: [[TMP6:%.*]] = shl i32 [[TMP5]], [[SHIFTAMT]] +; CHECK-NEXT: [[TMP7:%.*]] = load i32, i32 addrspace(3)* [[ALIGNEDADDR1]], align 4 +; CHECK-NEXT: [[TMP8:%.*]] = and i32 [[TMP7]], [[INV_MASK]] +; CHECK-NEXT: br label [[PARTWORD_CMPXCHG_LOOP:%.*]] +; CHECK: partword.cmpxchg.loop: +; CHECK-NEXT: [[TMP9:%.*]] = phi i32 [ [[TMP8]], [[TMP0:%.*]] ], [ [[TMP15:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ] +; CHECK-NEXT: [[TMP10:%.*]] = or i32 [[TMP9]], [[TMP4]] +; CHECK-NEXT: [[TMP11:%.*]] = or i32 [[TMP9]], [[TMP6]] +; CHECK-NEXT: [[TMP12:%.*]] = cmpxchg i32 addrspace(3)* [[ALIGNEDADDR1]], i32 [[TMP11]], i32 [[TMP10]] seq_cst seq_cst, align 4 +; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { i32, i1 } [[TMP12]], 0 +; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { i32, i1 } [[TMP12]], 1 +; CHECK-NEXT: br i1 [[TMP14]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]] +; CHECK: partword.cmpxchg.failure: +; CHECK-NEXT: [[TMP15]] = and i32 [[TMP13]], [[INV_MASK]] +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP9]], [[TMP15]] +; CHECK-NEXT: br i1 [[TMP16]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]] +; CHECK: partword.cmpxchg.end: +; CHECK-NEXT: [[SHIFTED:%.*]] = lshr i32 [[TMP13]], [[SHIFTAMT]] +; CHECK-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8 +; CHECK-NEXT: [[TMP17:%.*]] = insertvalue { i8, i1 } undef, i8 [[EXTRACTED]], 0 +; CHECK-NEXT: [[TMP18:%.*]] = insertvalue { i8, i1 } [[TMP17]], i1 [[TMP14]], 1 +; CHECK-NEXT: [[EXTRACT:%.*]] = extractvalue { i8, i1 } [[TMP18]], 0 +; CHECK-NEXT: ret i8 [[EXTRACT]] +; + %gep = getelementptr i8, i8 addrspace(3)* %out, i64 4 + %res = cmpxchg i8 addrspace(3)* %gep, i8 %old, i8 %in seq_cst seq_cst, align 2 + %extract = extractvalue {i8, i1} %res, 0 + ret i8 %extract +} diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll index 52d37a2..19ce5ef 100644 --- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll +++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fadd.ll @@ -900,6 +900,23 @@ define half @test_atomicrmw_fadd_f16_global(half addrspace(1)* %ptr, half %value ret half %res } +define half @test_atomicrmw_fadd_f16_global_align4(half addrspace(1)* %ptr, half %value) { +; CI-LABEL: @test_atomicrmw_fadd_f16_global_align4( +; CI-NEXT: [[RES:%.*]] = atomicrmw fadd half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 4 +; CI-NEXT: ret half [[RES]] +; +; GFX9-LABEL: @test_atomicrmw_fadd_f16_global_align4( +; GFX9-NEXT: [[RES:%.*]] = atomicrmw fadd half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 4 +; GFX9-NEXT: ret half [[RES]] +; +; GFX908-LABEL: @test_atomicrmw_fadd_f16_global_align4( +; GFX908-NEXT: [[RES:%.*]] = atomicrmw fadd half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 4 +; GFX908-NEXT: ret half [[RES]] +; + %res = atomicrmw fadd half addrspace(1)* %ptr, half %value seq_cst, align 4 + ret half %res +} + define half @test_atomicrmw_fadd_f16_local(half addrspace(3)* %ptr, half %value) { ; CI-LABEL: @test_atomicrmw_fadd_f16_local( ; CI-NEXT: [[RES:%.*]] = atomicrmw fadd half addrspace(3)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 2 diff --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fsub.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fsub.ll index 8a42e05..2fb4c36 100644 --- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fsub.ll +++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fsub.ll @@ -12,7 +12,7 @@ define float @test_atomicrmw_fsub_f32_flat(float* %ptr, float %value) { ; GCN-NEXT: [[TMP2:%.*]] = bitcast float* [[PTR]] to i32* ; GCN-NEXT: [[TMP3:%.*]] = bitcast float [[NEW]] to i32 ; GCN-NEXT: [[TMP4:%.*]] = bitcast float [[LOADED]] to i32 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i32 [[NEWLOADED]] to float @@ -34,7 +34,7 @@ define float @test_atomicrmw_fsub_f32_global(float addrspace(1)* %ptr, float %va ; GCN-NEXT: [[TMP2:%.*]] = bitcast float addrspace(1)* [[PTR]] to i32 addrspace(1)* ; GCN-NEXT: [[TMP3:%.*]] = bitcast float [[NEW]] to i32 ; GCN-NEXT: [[TMP4:%.*]] = bitcast float [[LOADED]] to i32 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32 addrspace(1)* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32 addrspace(1)* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i32 [[NEWLOADED]] to float @@ -56,7 +56,7 @@ define float @test_atomicrmw_fsub_f32_local(float addrspace(3)* %ptr, float %val ; GCN-NEXT: [[TMP2:%.*]] = bitcast float addrspace(3)* [[PTR]] to i32 addrspace(3)* ; GCN-NEXT: [[TMP3:%.*]] = bitcast float [[NEW]] to i32 ; GCN-NEXT: [[TMP4:%.*]] = bitcast float [[LOADED]] to i32 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32 addrspace(3)* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i32 addrspace(3)* [[TMP2]], i32 [[TMP4]], i32 [[TMP3]] seq_cst seq_cst, align 4 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i32, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i32 [[NEWLOADED]] to float @@ -70,7 +70,7 @@ define float @test_atomicrmw_fsub_f32_local(float addrspace(3)* %ptr, float %val define half @test_atomicrmw_fsub_f16_flat(half* %ptr, half %value) { ; GCN-LABEL: @test_atomicrmw_fsub_f16_flat( -; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst +; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 2 ; GCN-NEXT: ret half [[RES]] ; %res = atomicrmw fsub half* %ptr, half %value seq_cst @@ -79,16 +79,25 @@ define half @test_atomicrmw_fsub_f16_flat(half* %ptr, half %value) { define half @test_atomicrmw_fsub_f16_global(half addrspace(1)* %ptr, half %value) { ; GCN-LABEL: @test_atomicrmw_fsub_f16_global( -; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst +; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 2 ; GCN-NEXT: ret half [[RES]] ; %res = atomicrmw fsub half addrspace(1)* %ptr, half %value seq_cst ret half %res } +define half @test_atomicrmw_fsub_f16_global_align4(half addrspace(1)* %ptr, half %value) { +; GCN-LABEL: @test_atomicrmw_fsub_f16_global_align4( +; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half addrspace(1)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 4 +; GCN-NEXT: ret half [[RES]] +; + %res = atomicrmw fsub half addrspace(1)* %ptr, half %value seq_cst, align 4 + ret half %res +} + define half @test_atomicrmw_fsub_f16_local(half addrspace(3)* %ptr, half %value) { ; GCN-LABEL: @test_atomicrmw_fsub_f16_local( -; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half addrspace(3)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst +; GCN-NEXT: [[RES:%.*]] = atomicrmw fsub half addrspace(3)* [[PTR:%.*]], half [[VALUE:%.*]] seq_cst, align 2 ; GCN-NEXT: ret half [[RES]] ; %res = atomicrmw fsub half addrspace(3)* %ptr, half %value seq_cst @@ -105,7 +114,7 @@ define double @test_atomicrmw_fsub_f64_flat(double* %ptr, double %value) { ; GCN-NEXT: [[TMP2:%.*]] = bitcast double* [[PTR]] to i64* ; GCN-NEXT: [[TMP3:%.*]] = bitcast double [[NEW]] to i64 ; GCN-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double @@ -127,7 +136,7 @@ define double @test_atomicrmw_fsub_f64_global(double addrspace(1)* %ptr, double ; GCN-NEXT: [[TMP2:%.*]] = bitcast double addrspace(1)* [[PTR]] to i64 addrspace(1)* ; GCN-NEXT: [[TMP3:%.*]] = bitcast double [[NEW]] to i64 ; GCN-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64 addrspace(1)* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64 addrspace(1)* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double @@ -149,7 +158,7 @@ define double @test_atomicrmw_fsub_f64_local(double addrspace(3)* %ptr, double % ; GCN-NEXT: [[TMP2:%.*]] = bitcast double addrspace(3)* [[PTR]] to i64 addrspace(3)* ; GCN-NEXT: [[TMP3:%.*]] = bitcast double [[NEW]] to i64 ; GCN-NEXT: [[TMP4:%.*]] = bitcast double [[LOADED]] to i64 -; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64 addrspace(3)* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst +; GCN-NEXT: [[TMP5:%.*]] = cmpxchg i64 addrspace(3)* [[TMP2]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8 ; GCN-NEXT: [[SUCCESS:%.*]] = extractvalue { i64, i1 } [[TMP5]], 1 ; GCN-NEXT: [[NEWLOADED:%.*]] = extractvalue { i64, i1 } [[TMP5]], 0 ; GCN-NEXT: [[TMP6]] = bitcast i64 [[NEWLOADED]] to double -- 2.7.4