================
@@ -69,7 +68,52 @@ float16x8_t test_vfmaq_f16(float16x8_t a, float16x8_t b,
float16x8_t c) {
return vfmaq_f16(a, b, c);
}
-// ALL-LABEL: @test_vfma_lane_f16(
+// LLVM-LABEL: @test_vfms_f16(
+// CIR-LABEL: @vfms_f16(
+float16x4_t test_vfms_f16(float16x4_t a, float16x4_t b, float16x4_t c) {
+// CIR: [[FNEG:%.*]] = cir.fneg %{{.*}} : !cir.vector<4 x !cir.f16>
+// CIR: cir.call @vfma_f16(%{{.*}}, [[FNEG]], %{{.*}}) :
+
+// LLVM-SAME: <4 x half> {{.*}} [[A:%.*]], <4 x half> {{.*}} [[B:%.*]], <4 x
half> {{.*}} [[C:%.*]]) {{.*}} {
+// LLVM: [[FNEG:%.*]] = fneg <4 x half> [[B]]
+// LLVM-NEXT: [[A_I:%.*]] = bitcast <4 x half> [[A]] to <4 x i16>
+// LLVM-NEXT: [[B_I:%.*]] = bitcast <4 x half> [[FNEG]] to <4 x i16>
+// LLVM-NEXT: [[C_I:%.*]] = bitcast <4 x half> [[C]] to <4 x i16>
+// LLVM-NEXT: [[A_BYTES:%.*]] = bitcast <4 x i16> [[A_I]] to <8 x i8>
+// LLVM-NEXT: [[B_BYTES:%.*]] = bitcast <4 x i16> [[B_I]] to <8 x i8>
+// LLVM-NEXT: [[C_BYTES:%.*]] = bitcast <4 x i16> [[C_I]] to <8 x i8>
+// LLVM-NEXT: [[A_CAST:%.*]] = bitcast <8 x i8> [[A_BYTES]] to <4 x half>
+// LLVM-NEXT: [[B_CAST:%.*]] = bitcast <8 x i8> [[B_BYTES]] to <4 x half>
+// LLVM-NEXT: [[C_CAST:%.*]] = bitcast <8 x i8> [[C_BYTES]] to <4 x half>
+// LLVM-NEXT: [[FMA:%.*]] = call <4 x half> @llvm.fma.v4f16(<4 x half>
[[B_CAST]], <4 x half> [[C_CAST]], <4 x half> [[A_CAST]])
+// LLVM-NEXT: ret <4 x half> [[FMA]]
+ return vfms_f16(a, b, c);
+}
+
+// LLVM-LABEL: @test_vfmsq_f16(
+// CIR-LABEL: @vfmsq_f16(
+float16x8_t test_vfmsq_f16(float16x8_t a, float16x8_t b, float16x8_t c) {
+// CIR: [[FNEG:%.*]] = cir.fneg %{{.*}} : !cir.vector<8 x !cir.f16>
+// CIR: cir.call @vfmaq_f16(%{{.*}}, [[FNEG]], %{{.*}}) :
+
+// LLVM-SAME: <8 x half> {{.*}} [[A:%.*]], <8 x half> {{.*}} [[B:%.*]], <8 x
half> {{.*}} [[C:%.*]]) {{.*}} {
+// LLVM: [[FNEG:%.*]] = fneg <8 x half> [[B]]
+// LLVM-NEXT: [[A_I:%.*]] = bitcast <8 x half> [[A]] to <8 x i16>
+// LLVM-NEXT: [[B_I:%.*]] = bitcast <8 x half> [[FNEG]] to <8 x i16>
+// LLVM-NEXT: [[C_I:%.*]] = bitcast <8 x half> [[C]] to <8 x i16>
+// LLVM-NEXT: [[A_BYTES:%.*]] = bitcast <8 x i16> [[A_I]] to <16 x i8>
+// LLVM-NEXT: [[B_BYTES:%.*]] = bitcast <8 x i16> [[B_I]] to <16 x i8>
+// LLVM-NEXT: [[C_BYTES:%.*]] = bitcast <8 x i16> [[C_I]] to <16 x i8>
+// LLVM-NEXT: [[A_CAST:%.*]] = bitcast <16 x i8> [[A_BYTES]] to <8 x half>
+// LLVM-NEXT: [[B_CAST:%.*]] = bitcast <16 x i8> [[B_BYTES]] to <8 x half>
+// LLVM-NEXT: [[C_CAST:%.*]] = bitcast <16 x i8> [[C_BYTES]] to <8 x half>
+// LLVM-NEXT: [[FMA:%.*]] = call <8 x half> @llvm.fma.v8f16(<8 x half>
[[B_CAST]], <8 x half> [[C_CAST]], <8 x half> [[A_CAST]])
+// LLVM-NEXT: ret <8 x half> [[FMA]]
+ return vfmsq_f16(a, b, c);
+}
+
+// LLVM-LABEL: @test_vfma_lane_f16(
----------------
E00N777 wrote:
Sorry, my earlier comment was mistaken🥲. I assumed that the actual CIR lowering
logic lived in a separate vfma_lane_f16 function. After checking the emitted
CIR, I realized that vfma_lane_f16 is a macro and its logic is expanded
directly into test_vfma_lane_f16. Therefore, the original ALL-LABEL was
correct, and my request to change it was unnecessary. Apologies for the
confusion and the extra churn. Could you please revert that part of the update?
🙏🏻
https://github.com/llvm/llvm-project/pull/210359
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
