freedreno/ir3: optimize sam.s2en to sam

Detect when sampler/texture idx are immediate and switch to non s2en
encoding.

Signed-off-by: Rob Clark <robdclark@gmail.com>
This commit is contained in:
Rob Clark
2019-03-19 13:30:03 -04:00
parent 1443694ee5
commit bcd81d2387
3 changed files with 36 additions and 6 deletions
+7 -6
View File
@@ -29,12 +29,13 @@
#include "ir3_compiler.h"
static const struct debug_named_value shader_debug_options[] = {
{"vs", IR3_DBG_SHADER_VS, "Print shader disasm for vertex shaders"},
{"fs", IR3_DBG_SHADER_FS, "Print shader disasm for fragment shaders"},
{"cs", IR3_DBG_SHADER_CS, "Print shader disasm for compute shaders"},
{"disasm", IR3_DBG_DISASM, "Dump NIR and adreno shader disassembly"},
{"optmsgs", IR3_DBG_OPTMSGS,"Enable optimizer debug messages"},
DEBUG_NAMED_VALUE_END
{"vs", IR3_DBG_SHADER_VS, "Print shader disasm for vertex shaders"},
{"fs", IR3_DBG_SHADER_FS, "Print shader disasm for fragment shaders"},
{"cs", IR3_DBG_SHADER_CS, "Print shader disasm for compute shaders"},
{"disasm", IR3_DBG_DISASM, "Dump NIR and adreno shader disassembly"},
{"optmsgs", IR3_DBG_OPTMSGS, "Enable optimizer debug messages"},
{"forces2en", IR3_DBG_FORCES2EN, "Force s2en mode for tex sampler instructions"},
DEBUG_NAMED_VALUE_END
};
DEBUG_GET_ONCE_FLAGS_OPTION(ir3_shader_debug, "IR3_SHADER_DEBUG", shader_debug_options, 0)
+1
View File
@@ -76,6 +76,7 @@ enum ir3_shader_debug {
IR3_DBG_SHADER_CS = 0x04,
IR3_DBG_DISASM = 0x08,
IR3_DBG_OPTMSGS = 0x10,
IR3_DBG_FORCES2EN = 0x20,
};
extern enum ir3_shader_debug ir3_shader_debug;
+28
View File
@@ -613,6 +613,34 @@ instr_cp(struct ir3_cp_ctx *ctx, struct ir3_instruction *instr)
break;
}
}
/* Handle converting a sam.s2en (taking samp/tex idx params via
* register) into a normal sam (encoding immediate samp/tex idx)
* if they are immediate. This saves some instructions and regs
* in the common case where we know samp/tex at compile time:
*/
if (is_tex(instr) && (instr->flags & IR3_INSTR_S2EN) &&
!(ir3_shader_debug & IR3_DBG_FORCES2EN)) {
/* The first src will be a fan-in (collect), if both of it's
* two sources are mov from imm, then we can
*/
struct ir3_instruction *samp_tex = ssa(instr->regs[1]);
debug_assert(samp_tex->opc == OPC_META_FI);
struct ir3_instruction *samp = ssa(samp_tex->regs[1]);
struct ir3_instruction *tex = ssa(samp_tex->regs[2]);
if ((samp->opc == OPC_MOV) &&
(samp->regs[1]->flags & IR3_REG_IMMED) &&
(tex->opc == OPC_MOV) &&
(tex->regs[1]->flags & IR3_REG_IMMED)) {
instr->flags &= ~IR3_INSTR_S2EN;
instr->cat5.samp = samp->regs[1]->iim_val;
instr->cat5.tex = tex->regs[1]->iim_val;
instr->regs[1]->instr = NULL;
}
}
}
void