[llvm] b337bce - AMDGPU: Split test functions to avoid dependency on subtarget
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 18 22:07:05 PST 2019
Author: Matt Arsenault
Date: 2019-11-19T11:12:13+05:30
New Revision: b337bce8710f2a7ab8ce9f84c80cfbce1032963c
URL: https://github.com/llvm/llvm-project/commit/b337bce8710f2a7ab8ce9f84c80cfbce1032963c
DIFF: https://github.com/llvm/llvm-project/commit/b337bce8710f2a7ab8ce9f84c80cfbce1032963c.diff
LOG: AMDGPU: Split test functions to avoid dependency on subtarget
Prepare this test for moving tthe denormal setting out of the
subtarget features.
Added:
Modified:
llvm/test/Analysis/CostModel/AMDGPU/fdiv.ll
Removed:
################################################################################
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/fdiv.ll b/llvm/test/Analysis/CostModel/AMDGPU/fdiv.ll
index 7a8012afcdfb..6986a3158461 100644
--- a/llvm/test/Analysis/CostModel/AMDGPU/fdiv.ll
+++ b/llvm/test/Analysis/CostModel/AMDGPU/fdiv.ll
@@ -1,55 +1,91 @@
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,CIFASTF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,CISLOWF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,SIFASTF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=verde -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,SISLOWF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -mattr=+fp32-denormals < %s | FileCheck -check-prefixes=ALL,FP32DENORMS,SLOWFP32DENORMS,NOFP16,NOFP16-FP32DENORM %s
-; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -mattr=+fp32-denormals < %s | FileCheck -check-prefixes=ALL,FP32DENORMS,FASTFP32DENORMS,FP16 %s
-
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,CIFASTF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,CISLOWF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,SIFASTF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=verde -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=ALL,SISLOWF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM %s
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -mattr=+fp32-denormals < %s | FileCheck -check-prefixes=ALL,FP32DENORMS,SLOWFP32DENORMS,NOFP16,NOFP16-FP32DENORM %s
-; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -mattr=+fp32-denormals < %s | FileCheck -check-prefixes=ALL,FP32DENORMS,FASTFP32DENORMS,FP16 %s
-
-; ALL: 'fdiv_f32'
-; NOFP32DENORM: estimated cost of 12 for {{.*}} fdiv float
-; FP32DENORMS: estimated cost of 10 for {{.*}} fdiv float
-define amdgpu_kernel void @fdiv_f32(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #0 {
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=ALL,CIFASTF64,NOFP16,NOFP16-NOFP32DENORM,SLOWFP32DENORMS %s
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri < %s | FileCheck -check-prefixes=ALL,CISLOWF64,NOFP16,NOFP16-NOFP32DENORM,SLOWFP32DENORMS %s
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=ALL,SIFASTF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM,SLOWFP32DENORMS %s
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=verde < %s | FileCheck -check-prefixes=ALL,SISLOWF64,NOFP32DENORM,NOFP16,NOFP16-NOFP32DENORM,SLOWFP32DENORMS %s
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=ALL,NOFP16,NOFP16-FP32DENORM,SLOWFP32DENORMS %s
+; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=ALL,FASTFP32DENORMS,FP16 %s
+
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=ALL,CIFASTF64,NOFP16,NOFP16-NOFP32DENORM %s
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri < %s | FileCheck -check-prefixes=ALL,CISLOWF64,NOFP16,NOFP16-NOFP32DENORM %s
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefixes=ALL,SIFASTF64,NOFP16,NOFP16-NOFP32DENORM %s
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-mesa-mesa3d -mcpu=verde < %s | FileCheck -check-prefixes=ALL,SISLOWF64,NOFP16,NOFP16-NOFP32DENORM %s
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii < %s | FileCheck -check-prefixes=ALL,SLOWFP32DENORMS,NOFP16,NOFP16-FP32DENORM %s
+; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=ALL,FASTFP32DENORMS,FP16 %s
+
+; ALL: 'fdiv_f32_ieee'
+; ALL: estimated cost of 10 for {{.*}} fdiv float
+define amdgpu_kernel void @fdiv_f32_ieee(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #0 {
+ %vec = load float, float addrspace(1)* %vaddr
+ %add = fdiv float %vec, %b
+ store float %add, float addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_f32_ftzdaz'
+; ALL: estimated cost of 12 for {{.*}} fdiv float
+define amdgpu_kernel void @fdiv_f32_ftzdaz(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #1 {
%vec = load float, float addrspace(1)* %vaddr
%add = fdiv float %vec, %b
store float %add, float addrspace(1)* %out
ret void
}
-; ALL: 'fdiv_v2f32'
-; NOFP32DENORM: estimated cost of 24 for {{.*}} fdiv <2 x float>
-; FP32DENORMS: estimated cost of 20 for {{.*}} fdiv <2 x float>
-define amdgpu_kernel void @fdiv_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr, <2 x float> %b) #0 {
+; ALL: 'fdiv_v2f32_ieee'
+; ALL: estimated cost of 20 for {{.*}} fdiv <2 x float>
+define amdgpu_kernel void @fdiv_v2f32_ieee(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr, <2 x float> %b) #0 {
+ %vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr
+ %add = fdiv <2 x float> %vec, %b
+ store <2 x float> %add, <2 x float> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_v2f32_ftzdaz'
+; ALL: estimated cost of 24 for {{.*}} fdiv <2 x float>
+define amdgpu_kernel void @fdiv_v2f32_ftzdaz(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr, <2 x float> %b) #1 {
%vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr
%add = fdiv <2 x float> %vec, %b
store <2 x float> %add, <2 x float> addrspace(1)* %out
ret void
}
-; ALL: 'fdiv_v3f32'
+; ALL: 'fdiv_v3f32_ieee'
; Allow for 48/40 when v3f32 is illegal and TargetLowering thinks it needs widening,
; and 36/30 when it is legal.
-; NOFP32DENORM: estimated cost of {{36|48}} for {{.*}} fdiv <3 x float>
-; FP32DENORMS: estimated cost of {{30|40}} for {{.*}} fdiv <3 x float>
-define amdgpu_kernel void @fdiv_v3f32(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #0 {
+; ALL: estimated cost of {{30|40}} for {{.*}} fdiv <3 x float>
+define amdgpu_kernel void @fdiv_v3f32_ieee(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #0 {
%vec = load <3 x float>, <3 x float> addrspace(1)* %vaddr
%add = fdiv <3 x float> %vec, %b
store <3 x float> %add, <3 x float> addrspace(1)* %out
ret void
}
-; ALL: 'fdiv_v5f32'
+; ALL: 'fdiv_v3f32_ftzdaz'
+; Allow for 48/40 when v3f32 is illegal and TargetLowering thinks it needs widening,
+; and 36/30 when it is legal.
+; ALL: estimated cost of {{36|48}} for {{.*}} fdiv <3 x float>
+define amdgpu_kernel void @fdiv_v3f32_ftzdaz(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #1 {
+ %vec = load <3 x float>, <3 x float> addrspace(1)* %vaddr
+ %add = fdiv <3 x float> %vec, %b
+ store <3 x float> %add, <3 x float> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_v5f32_ieee'
+; Allow for 96/80 when v5f32 is illegal and TargetLowering thinks it needs widening,
+; and 60/50 when it is legal.
+; ALL: estimated cost of {{80|50}} for {{.*}} fdiv <5 x float>
+define amdgpu_kernel void @fdiv_v5f32_ieee(<5 x float> addrspace(1)* %out, <5 x float> addrspace(1)* %vaddr, <5 x float> %b) #0 {
+ %vec = load <5 x float>, <5 x float> addrspace(1)* %vaddr
+ %add = fdiv <5 x float> %vec, %b
+ store <5 x float> %add, <5 x float> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_v5f32_ftzdaz'
; Allow for 96/80 when v5f32 is illegal and TargetLowering thinks it needs widening,
; and 60/50 when it is legal.
-; NOFP32DENORM: estimated cost of {{96|60}} for {{.*}} fdiv <5 x float>
-; FP32DENORMS: estimated cost of {{80|50}} for {{.*}} fdiv <5 x float>
-define amdgpu_kernel void @fdiv_v5f32(<5 x float> addrspace(1)* %out, <5 x float> addrspace(1)* %vaddr, <5 x float> %b) #0 {
+; ALL: estimated cost of {{96|60}} for {{.*}} fdiv <5 x float>
+define amdgpu_kernel void @fdiv_v5f32_ftzdaz(<5 x float> addrspace(1)* %out, <5 x float> addrspace(1)* %vaddr, <5 x float> %b) #1 {
%vec = load <5 x float>, <5 x float> addrspace(1)* %vaddr
%add = fdiv <5 x float> %vec, %b
store <5 x float> %add, <5 x float> addrspace(1)* %out
@@ -92,55 +128,99 @@ define amdgpu_kernel void @fdiv_v3f64(<3 x double> addrspace(1)* %out, <3 x doub
ret void
}
-; ALL: 'fdiv_f16'
-; NOFP16-NOFP32DENORM: estimated cost of 12 for {{.*}} fdiv half
-; NOFP16-FP32DENORM: estimated cost of 10 for {{.*}} fdiv half
+; ALL: 'fdiv_f16_f32_ieee'
+; NOFP16: estimated cost of 10 for {{.*}} fdiv half
; FP16: estimated cost of 10 for {{.*}} fdiv half
-define amdgpu_kernel void @fdiv_f16(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #0 {
+define amdgpu_kernel void @fdiv_f16_f32_ieee(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #0 {
%vec = load half, half addrspace(1)* %vaddr
%add = fdiv half %vec, %b
store half %add, half addrspace(1)* %out
ret void
}
-; ALL: 'fdiv_v2f16'
-; NOFP16-NOFP32DENORM: estimated cost of 24 for {{.*}} fdiv <2 x half>
-; NOFP16-FP32DENORM: estimated cost of 20 for {{.*}} fdiv <2 x half>
+; ALL: 'fdiv_f16_f32_ftzdaz'
+; NOFP16: estimated cost of 12 for {{.*}} fdiv half
+; FP16: estimated cost of 10 for {{.*}} fdiv half
+define amdgpu_kernel void @fdiv_f16_f32_ftzdaz(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #1 {
+ %vec = load half, half addrspace(1)* %vaddr
+ %add = fdiv half %vec, %b
+ store half %add, half addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_v2f16_f32_ieee'
+; NOFP16: estimated cost of 20 for {{.*}} fdiv <2 x half>
; FP16: estimated cost of 20 for {{.*}} fdiv <2 x half>
-define amdgpu_kernel void @fdiv_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #0 {
+define amdgpu_kernel void @fdiv_v2f16_f32_ieee(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #0 {
%vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr
%add = fdiv <2 x half> %vec, %b
store <2 x half> %add, <2 x half> addrspace(1)* %out
ret void
}
-; ALL: 'fdiv_v4f16'
-; NOFP16-NOFP32DENORM: estimated cost of 48 for {{.*}} fdiv <4 x half>
-; NOFP16-FP32DENORM: estimated cost of 40 for {{.*}} fdiv <4 x half>
+; ALL: 'fdiv_v2f16_f32_ftzdaz'
+; NOFP16: estimated cost of 24 for {{.*}} fdiv <2 x half>
+; FP16: estimated cost of 20 for {{.*}} fdiv <2 x half>
+define amdgpu_kernel void @fdiv_v2f16_f32_ftzdaz(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #1 {
+ %vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr
+ %add = fdiv <2 x half> %vec, %b
+ store <2 x half> %add, <2 x half> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'fdiv_v4f16_f32_ieee'
+; NOFP16: estimated cost of 40 for {{.*}} fdiv <4 x half>
; FP16: estimated cost of 40 for {{.*}} fdiv <4 x half>
-define amdgpu_kernel void @fdiv_v4f16(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #0 {
+define amdgpu_kernel void @fdiv_v4f16_f32_ieee(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #0 {
%vec = load <4 x half>, <4 x half> addrspace(1)* %vaddr
%add = fdiv <4 x half> %vec, %b
store <4 x half> %add, <4 x half> addrspace(1)* %out
ret void
}
-; ALL: 'rcp_f32'
-; NOFP32DENORM: estimated cost of 3 for {{.*}} fdiv float
+; ALL: 'fdiv_v4f16_f32_ftzdaz'
+; NOFP16: estimated cost of 48 for {{.*}} fdiv <4 x half>
+; FP16: estimated cost of 40 for {{.*}} fdiv <4 x half>
+define amdgpu_kernel void @fdiv_v4f16_f32_ftzdaz(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #1 {
+ %vec = load <4 x half>, <4 x half> addrspace(1)* %vaddr
+ %add = fdiv <4 x half> %vec, %b
+ store <4 x half> %add, <4 x half> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'rcp_f32_ieee'
; SLOWFP32DENORMS: estimated cost of 10 for {{.*}} fdiv float
; FASTFP32DENORMS: estimated cost of 10 for {{.*}} fdiv float
-define amdgpu_kernel void @rcp_f32(float addrspace(1)* %out, float addrspace(1)* %vaddr) #0 {
+define amdgpu_kernel void @rcp_f32_ieee(float addrspace(1)* %out, float addrspace(1)* %vaddr) #0 {
%vec = load float, float addrspace(1)* %vaddr
%add = fdiv float 1.0, %vec
store float %add, float addrspace(1)* %out
ret void
}
-; ALL: 'rcp_f16'
-; NOFP16-NOFP32DENORM: estimated cost of 3 for {{.*}} fdiv half
-; NOFP16-FP32DENORM: estimated cost of 10 for {{.*}} fdiv half
+; ALL: 'rcp_f32_ftzdaz'
+; ALL: estimated cost of 3 for {{.*}} fdiv float
+define amdgpu_kernel void @rcp_f32_ftzdaz(float addrspace(1)* %out, float addrspace(1)* %vaddr) #1 {
+ %vec = load float, float addrspace(1)* %vaddr
+ %add = fdiv float 1.0, %vec
+ store float %add, float addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'rcp_f16_f32_ieee'
+; NOFP16: estimated cost of 10 for {{.*}} fdiv half
+; FP16: estimated cost of 3 for {{.*}} fdiv half
+define amdgpu_kernel void @rcp_f16_f32_ieee(half addrspace(1)* %out, half addrspace(1)* %vaddr) #0 {
+ %vec = load half, half addrspace(1)* %vaddr
+ %add = fdiv half 1.0, %vec
+ store half %add, half addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'rcp_f16_f32_ftzdaz'
+; NOFP16: estimated cost of 3 for {{.*}} fdiv half
; FP16: estimated cost of 3 for {{.*}} fdiv half
-define amdgpu_kernel void @rcp_f16(half addrspace(1)* %out, half addrspace(1)* %vaddr) #0 {
+define amdgpu_kernel void @rcp_f16_f32_ftzdaz(half addrspace(1)* %out, half addrspace(1)* %vaddr) #1 {
%vec = load half, half addrspace(1)* %vaddr
%add = fdiv half 1.0, %vec
store half %add, half addrspace(1)* %out
@@ -159,26 +239,44 @@ define amdgpu_kernel void @rcp_f64(double addrspace(1)* %out, double addrspace(1
ret void
}
-; ALL: 'rcp_v2f32'
-; NOFP32DENORM: estimated cost of 6 for {{.*}} fdiv <2 x float>
+; ALL: 'rcp_v2f32_ieee'
; SLOWFP32DENORMS: estimated cost of 20 for {{.*}} fdiv <2 x float>
; FASTFP32DENORMS: estimated cost of 20 for {{.*}} fdiv <2 x float>
-define amdgpu_kernel void @rcp_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr) #0 {
+define amdgpu_kernel void @rcp_v2f32_ieee(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr) #0 {
%vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr
%add = fdiv <2 x float> <float 1.0, float 1.0>, %vec
store <2 x float> %add, <2 x float> addrspace(1)* %out
ret void
}
-; ALL: 'rcp_v2f16'
-; NOFP16-NOFP32DENORM: estimated cost of 6 for {{.*}} fdiv <2 x half>
-; NOFP16-FP32DENORM: estimated cost of 20 for {{.*}} fdiv <2 x half>
+; ALL: 'rcp_v2f32_ftzdaz'
+; ALL: estimated cost of 6 for {{.*}} fdiv <2 x float>
+define amdgpu_kernel void @rcp_v2f32_ftzdaz(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr) #1 {
+ %vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr
+ %add = fdiv <2 x float> <float 1.0, float 1.0>, %vec
+ store <2 x float> %add, <2 x float> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'rcp_v2f16_f32_ieee'
+; NOFP16: estimated cost of 20 for {{.*}} fdiv <2 x half>
+; FP16: estimated cost of 6 for {{.*}} fdiv <2 x half>
+define amdgpu_kernel void @rcp_v2f16_f32_ieee(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr) #0 {
+ %vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr
+ %add = fdiv <2 x half> <half 1.0, half 1.0>, %vec
+ store <2 x half> %add, <2 x half> addrspace(1)* %out
+ ret void
+}
+
+; ALL: 'rcp_v2f16_f32_ftzdaz'
+; NOFP16: estimated cost of 6 for {{.*}} fdiv <2 x half>
; FP16: estimated cost of 6 for {{.*}} fdiv <2 x half>
-define amdgpu_kernel void @rcp_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr) #0 {
+define amdgpu_kernel void @rcp_v2f16_f32_ftzdaz(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr) #1 {
%vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr
%add = fdiv <2 x half> <half 1.0, half 1.0>, %vec
store <2 x half> %add, <2 x half> addrspace(1)* %out
ret void
}
-attributes #0 = { nounwind }
+attributes #0 = { nounwind "target-features"="+fp32-denormals" }
+attributes #1 = { nounwind "target-features"="-fp32-denormals" }
More information about the llvm-commits
mailing list