mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 11:13:28 +01:00
30d519d924
The code could issue a truncate from a small type to larger type. We need to extend in that case instead. llvm-svn: 333460
128 lines
4.3 KiB
LLVM
128 lines
4.3 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
|
|
|
|
define <4 x float> @select_mask_add_ss(<4 x float> %w, i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_mask_add_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vaddss %xmm2, %xmm1, %xmm0 {%k1}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fadd float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = extractelement <4 x float> %w, i32 0
|
|
%6 = select i1 %4, float %5, float %2
|
|
%7 = insertelement <4 x float> %a, float %6, i32 0
|
|
ret <4 x float> %7
|
|
}
|
|
|
|
define <4 x float> @select_maskz_add_ss(i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_maskz_add_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vaddss %xmm1, %xmm0, %xmm0 {%k1} {z}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fadd float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = select i1 %4, float 0.000000e+00, float %2
|
|
%6 = insertelement <4 x float> %a, float %5, i32 0
|
|
ret <4 x float> %6
|
|
}
|
|
|
|
define <4 x float> @select_mask_sub_ss(<4 x float> %w, i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_mask_sub_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vsubss %xmm2, %xmm1, %xmm0 {%k1}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fsub float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = extractelement <4 x float> %w, i32 0
|
|
%6 = select i1 %4, float %5, float %2
|
|
%7 = insertelement <4 x float> %a, float %6, i32 0
|
|
ret <4 x float> %7
|
|
}
|
|
|
|
define <4 x float> @select_maskz_sub_ss(i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_maskz_sub_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vsubss %xmm1, %xmm0, %xmm0 {%k1} {z}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fsub float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = select i1 %4, float 0.000000e+00, float %2
|
|
%6 = insertelement <4 x float> %a, float %5, i32 0
|
|
ret <4 x float> %6
|
|
}
|
|
|
|
define <4 x float> @select_mask_mul_ss(<4 x float> %w, i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_mask_mul_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vmulss %xmm2, %xmm1, %xmm0 {%k1}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fmul float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = extractelement <4 x float> %w, i32 0
|
|
%6 = select i1 %4, float %5, float %2
|
|
%7 = insertelement <4 x float> %a, float %6, i32 0
|
|
ret <4 x float> %7
|
|
}
|
|
|
|
define <4 x float> @select_maskz_mul_ss(i8 zeroext %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_maskz_mul_ss:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vmulss %xmm1, %xmm0, %xmm0 {%k1} {z}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fmul float %1, %0
|
|
%3 = and i8 %u, 1
|
|
%4 = icmp eq i8 %3, 0
|
|
%5 = select i1 %4, float 0.000000e+00, float %2
|
|
%6 = insertelement <4 x float> %a, float %5, i32 0
|
|
ret <4 x float> %6
|
|
}
|
|
|
|
; Make sure we don't crash trying to truncate the and instruction i4->i8. We need to extend instead.
|
|
define <4 x float> @select_mask_add_ss_small_mask_type(<4 x float> %w, i4 %u, <4 x float> %a, <4 x float> %b) {
|
|
; CHECK-LABEL: select_mask_add_ss_small_mask_type:
|
|
; CHECK: ## %bb.0: ## %entry
|
|
; CHECK-NEXT: kmovw %edi, %k1
|
|
; CHECK-NEXT: vaddss %xmm2, %xmm1, %xmm0 {%k1}
|
|
; CHECK-NEXT: retq
|
|
entry:
|
|
%0 = extractelement <4 x float> %b, i32 0
|
|
%1 = extractelement <4 x float> %a, i32 0
|
|
%2 = fadd float %1, %0
|
|
%3 = and i4 %u, 1
|
|
%4 = icmp eq i4 %3, 0
|
|
%5 = extractelement <4 x float> %w, i32 0
|
|
%6 = select i1 %4, float %5, float %2
|
|
%7 = insertelement <4 x float> %a, float %6, i32 0
|
|
ret <4 x float> %7
|
|
}
|
|
|