diff options
author | Keith Whitwell <keith@tungstengraphics.com> | 2005-06-08 21:56:21 +0000 |
---|---|---|
committer | Keith Whitwell <keith@tungstengraphics.com> | 2005-06-08 21:56:21 +0000 |
commit | dce3a91a8d755e394d80b6d594fb3114eed67e49 (patch) | |
tree | 346a86826370d1f539f6858a6f0abca5ad29651f /src/mesa | |
parent | 2d2f5a4b19a95477e1cf6915963fe9fdbfc461f7 (diff) |
Add some more opcodes
Diffstat (limited to 'src/mesa')
-rw-r--r-- | src/mesa/x86/rtasm/x86sse.c | 101 | ||||
-rw-r--r-- | src/mesa/x86/rtasm/x86sse.h | 208 |
2 files changed, 149 insertions, 160 deletions
diff --git a/src/mesa/x86/rtasm/x86sse.c b/src/mesa/x86/rtasm/x86sse.c index 4aee89a89a..90145c164d 100644 --- a/src/mesa/x86/rtasm/x86sse.c +++ b/src/mesa/x86/rtasm/x86sse.c @@ -2,7 +2,7 @@ #include "x86sse.h" -#define DISASSEM 0 +#define DISASSEM 1 #define X86_TWOB 0x0f /* Emit bytes to the instruction stream: @@ -21,10 +21,9 @@ static void emit_1i( struct x86_function *p, GLint i0 ) static void disassem( struct x86_function *p, const char *fn ) { #if DISASSEM - static const char *last_fn; - if (fn && fn != last_fn) { + if (fn && fn != p->fn) { _mesa_printf("0x%x: %s\n", p->csr, fn); - last_fn = fn; + p->fn = fn; } #endif } @@ -75,7 +74,8 @@ static void emit_modrm( struct x86_function *p, /* Oh-oh we've stumbled into the SIB thing. */ - if (regmem.idx == reg_SP) { + if (regmem.file == file_REG32 && + regmem.idx == reg_SP) { emit_1ub_fn(p, 0x24, 0); /* simplistic! */ } @@ -357,6 +357,38 @@ void sse_movlps( struct x86_function *p, /* SSE operations often only have one format, with dest constrained to * be a register: */ +void sse_maxps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x5F); + emit_modrm( p, dst, src ); +} + +void sse_divss( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_3ub(p, 0xF3, X86_TWOB, 0x5E); + emit_modrm( p, dst, src ); +} + +void sse_minps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x5D); + emit_modrm( p, dst, src ); +} + +void sse_subps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x5C); + emit_modrm( p, dst, src ); +} + void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) @@ -373,6 +405,39 @@ void sse_addps( struct x86_function *p, emit_modrm( p, dst, src ); } +void sse_addss( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_3ub(p, 0xF3, X86_TWOB, 0x58); + emit_modrm( p, dst, src ); +} + +void sse_andps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_2ub(p, X86_TWOB, 0x54); + emit_modrm( p, dst, src ); +} + +void sse2_rcpss( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_3ub(p, 0xF3, X86_TWOB, 0x53); + emit_modrm( p, dst, src ); +} + +void sse_rsqrtss( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src ) +{ + emit_3ub(p, 0xF3, X86_TWOB, 0x52); + emit_modrm( p, dst, src ); + +} + void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ) @@ -488,14 +553,11 @@ void x86_test( struct x86_function *p, void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, - GLubyte x, - GLubyte y, - GLubyte z, - GLubyte w) + GLubyte shuf) { emit_3ub(p, 0x66, X86_TWOB, 0x70); emit_modrm(p, dest, arg0); - emit_1ub(p, (x|(y<<2)|(z<<4)|w<<6)); + emit_1ub(p, shuf); } @@ -505,14 +567,21 @@ void sse2_pshufd( struct x86_function *p, void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, - GLubyte x, - GLubyte y, - GLubyte z, - GLubyte w) + GLubyte shuf) { emit_2ub(p, X86_TWOB, 0xC6); emit_modrm(p, dest, arg0); - emit_1ub(p, (x|(y<<2)|(z<<4)|w<<6)); + emit_1ub(p, shuf); +} + +void sse_cmpps( struct x86_function *p, + struct x86_reg dest, + struct x86_reg arg0, + GLubyte cc) +{ + emit_2ub(p, X86_TWOB, 0xC2); + emit_modrm(p, dest, arg0); + emit_1ub(p, cc); } @@ -541,6 +610,8 @@ void x86_release_func( struct x86_function *p ) void (*x86_get_func( struct x86_function *p ))(void) { + if (DISASSEM) + _mesa_printf("disassemble %p %p\n", p->store, p->csr); return (void (*)())p->store; } diff --git a/src/mesa/x86/rtasm/x86sse.h b/src/mesa/x86/rtasm/x86sse.h index 8d48e35647..19f8967a2f 100644 --- a/src/mesa/x86/rtasm/x86sse.h +++ b/src/mesa/x86/rtasm/x86sse.h @@ -22,6 +22,7 @@ struct x86_function { GLubyte *csr; GLuint stack_offset; GLint need_emms; + const char *fn; }; enum x86_reg_file { @@ -60,6 +61,17 @@ enum x86_cc { cc_NE /* not equal / not zero */ }; +enum sse_cc { + cc_Equal, + cc_LessThan, + cc_LessThanEqual, + cc_Unordered, + cc_NotEqual, + cc_NotLessThan, + cc_NotLessThanEqual, + cc_Ordered +}; + #define cc_Z cc_E #define cc_NZ cc_NE @@ -86,8 +98,6 @@ struct x86_reg x86_deref( struct x86_reg reg ); struct x86_reg x86_get_base_reg( struct x86_reg reg ); - - /* Labels, jumps and fixup: */ GLubyte *x86_get_label( struct x86_function *p ); @@ -96,162 +106,70 @@ void x86_jcc( struct x86_function *p, enum x86_cc cc, GLubyte *label ); -/* Always use a 32bit offset for forward jumps: - */ GLubyte *x86_jcc_forward( struct x86_function *p, enum x86_cc cc ); -/* Fixup offset from forward jump: - */ void x86_fixup_fwd_jump( struct x86_function *p, GLubyte *fixup ); -void x86_push( struct x86_function *p, - struct x86_reg reg ); - -void x86_pop( struct x86_function *p, - struct x86_reg reg ); - -void x86_inc( struct x86_function *p, - struct x86_reg reg ); - -void x86_dec( struct x86_function *p, - struct x86_reg reg ); - -void x86_ret( struct x86_function *p ); - -void mmx_emms( struct x86_function *p ); - -void x86_mov( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void x86_xor( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void x86_cmp( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse2_movd( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void mmx_movd( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void mmx_movq( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movss( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movaps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movups( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); -void sse_movhps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movlps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -/* SSE operations often only have one format, with dest constrained to - * be a register: - */ -void sse_mulps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_addps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movhlps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_movlhps( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse2_cvtps2dq( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse2_packssdw( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse2_packsswb( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse2_packuswb( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void sse_cvtps2pi( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void mmx_packssdw( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void mmx_packuswb( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - - -/* Load effective address: +/* Macro for sse_shufps() and sse2_pshufd(): */ -void x86_lea( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); - -void x86_test( struct x86_function *p, - struct x86_reg dst, - struct x86_reg src ); +#define SHUF(_x,_y,_z,_w) (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6)) +#define SHUF_NOOP RSW(0,1,2,3) +#define GET_SHUF(swz, idx) (((swz) >> ((idx)*2)) & 0x3) -/* Perform a reduced swizzle in a single sse instruction: - */ -void sse2_pshufd( struct x86_function *p, - struct x86_reg dest, - struct x86_reg arg0, - GLubyte x, - GLubyte y, - GLubyte z, - GLubyte w ); - - -/* Shufps can also be used to implement a reduced swizzle when dest == - * arg0. - */ -void sse_shufps( struct x86_function *p, - struct x86_reg dest, - struct x86_reg arg0, - GLubyte x, - GLubyte y, - GLubyte z, - GLubyte w ); +void mmx_emms( struct x86_function *p ); +void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); + +void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf ); +void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); + +void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src, GLubyte cc ); +void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0, GLubyte shuf ); + +void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void x86_dec( struct x86_function *p, struct x86_reg reg ); +void x86_inc( struct x86_function *p, struct x86_reg reg ); +void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void x86_pop( struct x86_function *p, struct x86_reg reg ); +void x86_push( struct x86_function *p, struct x86_reg reg ); +void x86_ret( struct x86_function *p ); +void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); +void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); /* Retreive a reference to one of the function arguments, taking into - * account any push/pop activity: + * account any push/pop activity. Note - doesn't track explict + * manipulation of ESP by other instructions. */ -struct x86_reg x86_fn_arg( struct x86_function *p, - GLuint arg ); +struct x86_reg x86_fn_arg( struct x86_function *p, GLuint arg ); #endif #endif |