llvmpipe: Add a few more common arithmetic functions.

We are relying on SSE4.1 for round/trunc/ceil/floor. We'll need to
eventually find alternatives for the rest of the world.

diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
index be7442d00aec585c1eee113329b448a09a2fc10a..ce3e5f91c071997681d4a160cd606a4229603f37 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
@@ -623,6 +623,47 @@ lp_build_abs(struct lp_build_context *bld,
 }
 
 
+LLVMValueRef
+lp_build_sgn(struct lp_build_context *bld,
+             LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMValueRef cond;
+   LLVMValueRef res;
+
+   /* Handle non-zero case */
+   if(!type.sign) {
+      /* if not zero then sign must be positive */
+      res = bld->one;
+   }
+   else if(type.floating) {
+      /* Take the sign bit and add it to 1 constant */
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+      LLVMValueRef sign;
+      LLVMValueRef one;
+      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+      one = LLVMConstBitCast(bld->one, int_vec_type);
+      res = LLVMBuildOr(bld->builder, sign, one, "");
+      res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+   }
+   else
+   {
+      LLVMValueRef minus_one = lp_build_const_scalar(type, -1.0);
+      cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero);
+      res = lp_build_select(bld, cond, bld->one, minus_one);
+   }
+
+   /* Handle zero */
+   cond = lp_build_cmp(bld, PIPE_FUNC_EQUAL, a, bld->zero);
+   res = lp_build_select(bld, cond, bld->zero, bld->one);
+
+   return res;
+}
+
+
 enum lp_build_round_sse41_mode
 {
    LP_BUILD_ROUND_SSE41_NEAREST = 0,
@@ -661,6 +702,24 @@ lp_build_round_sse41(struct lp_build_context *bld,
 }
 
 
+LLVMValueRef
+lp_build_round(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
 LLVMValueRef
 lp_build_floor(struct lp_build_context *bld,
                LLVMValueRef a)
@@ -679,6 +738,42 @@ lp_build_floor(struct lp_build_context *bld,
 }
 
 
+LLVMValueRef
+lp_build_ceil(struct lp_build_context *bld,
+              LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
+LLVMValueRef
+lp_build_trunc(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
 /**
  * Convert to integer, through whichever rounding method that's fastest,
  * typically truncating to zero.

diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.h b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
index 383c3c3313b19749793263b06ee55743e7f482e0..5e083b847fdf66dc67a95f1bfd714751d852bf06 100644 (file)
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
@@ -105,10 +105,26 @@ LLVMValueRef
 lp_build_abs(struct lp_build_context *bld,
              LLVMValueRef a);
 
+LLVMValueRef
+lp_build_sgn(struct lp_build_context *bld,
+             LLVMValueRef a);
+
+LLVMValueRef
+lp_build_round(struct lp_build_context *bld,
+               LLVMValueRef a);
+
 LLVMValueRef
 lp_build_floor(struct lp_build_context *bld,
                LLVMValueRef a);
 
+LLVMValueRef
+lp_build_ceil(struct lp_build_context *bld,
+              LLVMValueRef a);
+
+LLVMValueRef
+lp_build_trunc(struct lp_build_context *bld,
+               LLVMValueRef a);
+
 LLVMValueRef
 lp_build_int(struct lp_build_context *bld,
              LLVMValueRef a);