Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(341)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: test/cctest/test-assembler-ia32.cc

Issue 773783002: [ia32] Introduce FMA3 instructions on scalar data elements. (Closed) Base URL: https://chromium.googlesource.com/v8/v8.git@master
Patch Set: Created 6 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « src/ia32/disasm-ia32.cc ('k') | test/cctest/test-disasm-ia32.cc » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: test/cctest/test-assembler-ia32.cc
diff --git a/test/cctest/test-assembler-ia32.cc b/test/cctest/test-assembler-ia32.cc
index d943297393b2b1be4d6ff4a1cb8898dc9712159b..f59c3c4aa1878e8cf6141fddc2169d7525ae2db8 100644
--- a/test/cctest/test-assembler-ia32.cc
+++ b/test/cctest/test-assembler-ia32.cc
@@ -589,4 +589,458 @@ TEST(AssemblerIa32SSE) {
}
+typedef int (*F9)(double x, double y, double z);
+TEST(AssemblerX64FMA_sd) {
+ CcTest::InitializeVM();
+ if (!CpuFeatures::IsSupported(FMA3)) return;
+
+ Isolate* isolate = reinterpret_cast<Isolate*>(CcTest::isolate());
+ HandleScope scope(isolate);
+ v8::internal::byte buffer[1024];
+ MacroAssembler assm(isolate, buffer, sizeof buffer);
+ {
+ CpuFeatureScope fscope(&assm, FMA3);
+ Label exit;
+ __ movsd(xmm0, Operand(esp, 1 * kPointerSize));
+ __ movsd(xmm1, Operand(esp, 3 * kPointerSize));
+ __ movsd(xmm2, Operand(esp, 5 * kPointerSize));
+ // argument in xmm0, xmm1 and xmm2
+ // xmm0 * xmm1 + xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulsd(xmm3, xmm1);
+ __ addsd(xmm3, xmm2); // Expected result in xmm3
+
+ __ sub(esp, Immediate(kDoubleSize)); // For memory operand
+ // vfmadd132sd
+ __ mov(eax, Immediate(1)); // Test number
+ __ movaps(xmm4, xmm0);
+ __ vfmadd132sd(xmm4, xmm2, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfmadd213sd(xmm4, xmm0, xmm2);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfmadd231sd(xmm4, xmm0, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfmadd132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfmadd132sd(xmm4, xmm2, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movsd(Operand(esp, 0), xmm2);
+ __ vfmadd213sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfmadd231sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // xmm0 * xmm1 - xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulsd(xmm3, xmm1);
+ __ subsd(xmm3, xmm2); // Expected result in xmm3
+
+ // vfmsub132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfmsub132sd(xmm4, xmm2, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfmsub213sd(xmm4, xmm0, xmm2);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfmsub231sd(xmm4, xmm0, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfmsub132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfmsub132sd(xmm4, xmm2, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movsd(Operand(esp, 0), xmm2);
+ __ vfmsub213sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfmsub231sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ // - xmm0 * xmm1 + xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulsd(xmm3, xmm1);
+ __ Move(xmm4, (uint64_t)1 << 63);
+ __ xorpd(xmm3, xmm4);
+ __ addsd(xmm3, xmm2); // Expected result in xmm3
+
+ // vfnmadd132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfnmadd132sd(xmm4, xmm2, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfnmadd213sd(xmm4, xmm0, xmm2);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfnmadd231sd(xmm4, xmm0, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfnmadd132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfnmadd132sd(xmm4, xmm2, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movsd(Operand(esp, 0), xmm2);
+ __ vfnmadd213sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfnmadd231sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ // - xmm0 * xmm1 - xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulsd(xmm3, xmm1);
+ __ Move(xmm4, (uint64_t)1 << 63);
+ __ xorpd(xmm3, xmm4);
+ __ subsd(xmm3, xmm2); // Expected result in xmm3
+
+ // vfnmsub132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfnmsub132sd(xmm4, xmm2, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfnmsub213sd(xmm4, xmm0, xmm2);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfnmsub231sd(xmm4, xmm0, xmm1);
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfnmsub132sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfnmsub132sd(xmm4, xmm2, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub213sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movsd(Operand(esp, 0), xmm2);
+ __ vfnmsub213sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub231sd
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movsd(Operand(esp, 0), xmm1);
+ __ vfnmsub231sd(xmm4, xmm0, Operand(esp, 0));
+ __ ucomisd(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ __ xor_(eax, eax);
+ __ bind(&exit);
+ __ add(esp, Immediate(kDoubleSize));
+ __ ret(0);
+ }
+
+ CodeDesc desc;
+ assm.GetCode(&desc);
+ Handle<Code> code = isolate->factory()->NewCode(
+ desc, Code::ComputeFlags(Code::STUB), Handle<Code>());
+#ifdef OBJECT_PRINT
+ OFStream os(stdout);
+ code->Print(os);
+#endif
+
+ F9 f = FUNCTION_CAST<F9>(code->entry());
+ CHECK_EQ(0, f(0.000092662107262076, -2.460774966188315, -1.0958787393627414));
+}
+
+
+typedef int (*F10)(float x, float y, float z);
+TEST(AssemblerX64FMA_ss) {
+ CcTest::InitializeVM();
+ if (!CpuFeatures::IsSupported(FMA3)) return;
+
+ Isolate* isolate = reinterpret_cast<Isolate*>(CcTest::isolate());
+ HandleScope scope(isolate);
+ v8::internal::byte buffer[1024];
+ MacroAssembler assm(isolate, buffer, sizeof buffer);
+ {
+ CpuFeatureScope fscope(&assm, FMA3);
+ Label exit;
+ __ movss(xmm0, Operand(esp, 1 * kPointerSize));
+ __ movss(xmm1, Operand(esp, 2 * kPointerSize));
+ __ movss(xmm2, Operand(esp, 3 * kPointerSize));
+ // arguments in xmm0, xmm1 and xmm2
+ // xmm0 * xmm1 + xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulss(xmm3, xmm1);
+ __ addss(xmm3, xmm2); // Expected result in xmm3
+
+ __ sub(esp, Immediate(kDoubleSize)); // For memory operand
+ // vfmadd132ss
+ __ mov(eax, Immediate(1)); // Test number
+ __ movaps(xmm4, xmm0);
+ __ vfmadd132ss(xmm4, xmm2, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfmadd213ss(xmm4, xmm0, xmm2);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfmadd231ss(xmm4, xmm0, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfmadd132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfmadd132ss(xmm4, xmm2, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movss(Operand(esp, 0), xmm2);
+ __ vfmadd213ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfmadd231ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // xmm0 * xmm1 - xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulss(xmm3, xmm1);
+ __ subss(xmm3, xmm2); // Expected result in xmm3
+
+ // vfmsub132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfmsub132ss(xmm4, xmm2, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfmsub213ss(xmm4, xmm0, xmm2);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfmsub231ss(xmm4, xmm0, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfmsub132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfmsub132ss(xmm4, xmm2, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movss(Operand(esp, 0), xmm2);
+ __ vfmsub213ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfmsub231ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ // - xmm0 * xmm1 + xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulss(xmm3, xmm1);
+ __ Move(xmm4, (uint32_t)1 << 31);
+ __ xorps(xmm3, xmm4);
+ __ addss(xmm3, xmm2); // Expected result in xmm3
+
+ // vfnmadd132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfnmadd132ss(xmm4, xmm2, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmadd213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfnmadd213ss(xmm4, xmm0, xmm2);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfnmadd231ss(xmm4, xmm0, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfnmadd132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfnmadd132ss(xmm4, xmm2, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movss(Operand(esp, 0), xmm2);
+ __ vfnmadd213ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmadd231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfnmadd231ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ // - xmm0 * xmm1 - xmm2
+ __ movaps(xmm3, xmm0);
+ __ mulss(xmm3, xmm1);
+ __ Move(xmm4, (uint32_t)1 << 31);
+ __ xorps(xmm3, xmm4);
+ __ subss(xmm3, xmm2); // Expected result in xmm3
+
+ // vfnmsub132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ vfnmsub132ss(xmm4, xmm2, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfmsub213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ vfnmsub213ss(xmm4, xmm0, xmm2);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ vfnmsub231ss(xmm4, xmm0, xmm1);
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+ // vfnmsub132ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm0);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfnmsub132ss(xmm4, xmm2, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub213ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm1);
+ __ movss(Operand(esp, 0), xmm2);
+ __ vfnmsub213ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+ // vfnmsub231ss
+ __ inc(eax);
+ __ movaps(xmm4, xmm2);
+ __ movss(Operand(esp, 0), xmm1);
+ __ vfnmsub231ss(xmm4, xmm0, Operand(esp, 0));
+ __ ucomiss(xmm4, xmm3);
+ __ j(not_equal, &exit);
+
+
+ __ xor_(eax, eax);
+ __ bind(&exit);
+ __ add(esp, Immediate(kDoubleSize));
+ __ ret(0);
+ }
+
+ CodeDesc desc;
+ assm.GetCode(&desc);
+ Handle<Code> code = isolate->factory()->NewCode(
+ desc, Code::ComputeFlags(Code::STUB), Handle<Code>());
+#ifdef OBJECT_PRINT
+ OFStream os(stdout);
+ code->Print(os);
+#endif
+
+ F10 f = FUNCTION_CAST<F10>(code->entry());
+ CHECK_EQ(0, f(9.26621069e-05f, -2.4607749f, -1.09587872f));
+}
#undef __
« no previous file with comments | « src/ia32/disasm-ia32.cc ('k') | test/cctest/test-disasm-ia32.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698