Closed llvmbot closed 6 years ago
Verified that the problem is now fixed. Thanks!
Fixed. Uncovered additional argument handling bug in LowerFORMAL_ARGUMENTS as well, which was also fixed.
Given the following C-language, test_float4.c, which is compiled with clang
typedef float float4 attribute((ext_vector_type(4)));
int check(float4 a) { return a.x == 0.0f; }
Or given the following IR (after optimization)
; ModuleID = 'test-float4' target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128" target triple = "ppc-unknown-linux-gnu"
define i32 @check(<4 x float> %a) nounwind readnone { entry: %0 = extractelement <4 x float> %a, i32 0 ;
[#uses=1] %1 = fcmp oeq float %0, 0.000000e+00 ; [#uses=1] %2 = zext i1 %1 to i32 ; [#uses=1] ret i32 %2 } The CellSPU backend generates an assert failure (llc -march=cellspu test-float4)
llc: SPUISelLowering.cpp:2070: llvm::SDValue LowerEXTRACT_VECTOR_ELT(llvm::SDValue, llvm::SelectionDAG&): Assertion `false && "Invalid value type!"' failed.
I brief review of the code and it appears that the LowerEXTRACT_VECTOR_ELT function needs cases added for MVT::f32 and MVT::f64
Extended Description
Given the following C-language, test_float4.c, which is compiled with clang
typedef float float4 attribute((ext_vector_type(4)));
int check(float4 a) { return a.x == 0.0f; }
Or given the following IR (after optimization)
; ModuleID = 'test-float4' target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128" target triple = "ppc-unknown-linux-gnu"
define i32 @check(<4 x float> %a) nounwind readnone { entry: %0 = extractelement <4 x float> %a, i32 0 ; [#uses=1]
%1 = fcmp oeq float %0, 0.000000e+00 ; [#uses=1]
%2 = zext i1 %1 to i32 ; [#uses=1]
ret i32 %2
}
The CellSPU backend generates an assert failure (llc -march=cellspu test-float4)
llc: SPUISelLowering.cpp:2070: llvm::SDValue LowerEXTRACT_VECTOR_ELT(llvm::SDValue, llvm::SelectionDAG&): Assertion `false && "Invalid value type!"' failed.
I brief review of the code and it appears that the LowerEXTRACT_VECTOR_ELT function needs cases added for MVT::f32 and MVT::f64