From 45bae0c70f43dc04ccf485e2ea54b892a072cb0c Mon Sep 17 00:00:00 2001 From: Gabe Black Date: Mon, 17 Aug 2009 20:15:16 -0700 Subject: X86: Implement the multiply and add instructions. --- .../simd128/integer/arithmetic/multiply_add.py | 32 +++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-) (limited to 'src/arch/x86/isa/insts/simd128/integer/arithmetic/multiply_add.py') diff --git a/src/arch/x86/isa/insts/simd128/integer/arithmetic/multiply_add.py b/src/arch/x86/isa/insts/simd128/integer/arithmetic/multiply_add.py index 40b38867b..f157d165f 100644 --- a/src/arch/x86/isa/insts/simd128/integer/arithmetic/multiply_add.py +++ b/src/arch/x86/isa/insts/simd128/integer/arithmetic/multiply_add.py @@ -54,5 +54,35 @@ # Authors: Gabe Black microcode = ''' -# PMADDWD +def macroop PMADDWD_XMM_XMM { + mmuli ufp3, xmml, xmmlm, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmml, xmmlm, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmml, ufp3, ufp4, size=4, ext=0 + mmuli ufp3, xmmh, xmmhm, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmmh, xmmhm, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmmh, ufp3, ufp4, size=4, ext=0 +}; + +def macroop PMADDWD_XMM_M { + ldfp ufp1, seg, sib, "DISPLACEMENT", dataSize=8 + ldfp ufp2, seg, sib, "DISPLACEMENT + 8", dataSize=8 + mmuli ufp3, xmml, ufp1, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmml, ufp1, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmml, ufp3, ufp4, size=4, ext=0 + mmuli ufp3, xmmh, ufp2, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmmh, ufp2, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmmh, ufp3, ufp4, size=4, ext=0 +}; + +def macroop PMADDWD_XMM_P { + rdip t7 + ldfp ufp1, seg, riprel, "DISPLACEMENT", dataSize=8 + ldfp ufp2, seg, riprel, "DISPLACEMENT + 8", dataSize=8 + mmuli ufp3, xmml, ufp1, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmml, ufp1, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmml, ufp3, ufp4, size=4, ext=0 + mmuli ufp3, xmmh, ufp2, srcSize=2, destSize=4, ext=(0x2 | 0x10 | 0x20) + mmuli ufp4, xmmh, ufp2, srcSize=2, destSize=4, ext=(0x2 | 0x10) + maddi xmmh, ufp3, ufp4, size=4, ext=0 +}; ''' -- cgit v1.2.3